AI for Multifamily Innovation and Data-Driven Success

In this episode of the Multifamily Innovation® Podcast, host Patrick Antrim engages with Mukund Chopra, a serial entrepreneur and early investor in companies like Groupon and Slack. Mukund shares his insights on leveraging technology and data to enhance business efficiency in the multifamily industry. The conversation delves into the core purpose of a company, emphasizing the importance of maximizing return on equity through efficient resource deployment.

Mukund explains the significance of data ownership and its transformative potential when combined with AI, highlighting that the most valuable companies today are those that can access and unlock their data effectively. He provides a nuanced perspective on when not to use AI, stressing that its application should be strategic and not just a trend-driven decision.

Mukund recounts his experiences in various industries, including his leadership role at Groupon during its historic IPO and his work in big data and AI. He discusses how conversational AI and machine learning have been integrated into business processes long before the current hype, advocating for a thoughtful approach to adopting new technologies.

The episode covers practical frameworks for identifying areas where AI can add value, emphasizing the importance of starting with first principles thinking. Mukund also talks about the cultural shift required within organizations to embrace efficiency and innovation. He shares examples from his current venture, Blue Lake, which helps multifamily operators monetize customer data without invasive practices, illustrating how data can be a significant asset beyond traditional real estate operations.

The discussion wraps up with reflections on the evolving nature of companies, comparing traditional resource-based models with modern data-driven enterprises. Mukund underscores the need for multifamily leaders to view technology as a critical component of their business strategy, not just an IT function.

Overall, this episode offers a masterclass in understanding the intersection of technology, data, and business efficiency, providing valuable insights for multifamily executives looking to innovate and stay ahead in a competitive market.

About the Multifamily Innovation® Council:

The Multifamily Innovation® Council is the executive level membership organization that makes a difference in your bottom line, drives a better experience for your employees, and allows you an experience that keeps demand strong for your company. The council is uniquely positioned to focus on the intersection of Leadership, Technology, AI, and Innovation.

The Multifamily Innovation® Council is for Multifamily Business leaders who want to unlock value inside their organization so they can create better experiences and drive profitability inside their company.

To learn more or to join, visit https://multifamilyinnovation.com.

For more information and to engage with leaders shaping the future of multifamily innovation, visit https://multifamilyinnovation.com/.

Connect:
Multifamily Innovation® Council: https://multifamilyinnovation.com/
Patrick Antrim: https://www.linkedin.com/in/patrickantrim/

Episode Transcript

Speaker 1 (00:00):
Welcome back to another episode of the
Multifamily Innovation Podcast.
I'm your host, patrick Antrim,and I have a very exciting guest
for you today.
Mukand Chopra is a serialentrepreneur and investor.
He was activated early withGroupon as a chief revenue
officer, also an early investorin Slack, and you know he brings

(00:21):
this really interestingfundamental approach to
investments and the applicationto business efficiency for
multifamily.
We discussed what really is acompany right.
What is the primary purpose itserves.
He breaks this down in thesimplest form and we explored
questions like do executiveseven want to find efficiencies

(00:44):
in the business?
And, if they do, what are thoseroles?
How do those roles play outtoday?
Should they exist the way thatthey exist?
So, for example, do I need ahead of marketing or do I really
just want leads?
And so, using first principles,thinking through some of those
things and he goes into later inthe episode around how the most

(01:06):
valuable companies in the worldtoday have data, own their data
, can access their data and canunlock their data with AI we go
into really the turning datainto cash and then where not to
use AI.
Right, that's an interestingperspective and it's just really

(01:27):
more than just a conversation.
It's a masterclass on how tobuild and think through
enterprise value in a businesstoday.
So with that, have a listen.
Great interview with MukanChopra.
All right, so here we are.
We're talking about somethingdifferent when to not use AI.
Should we start there?

(01:47):
Yeah, actually, you know what?
Let me pause, pump the brakes alittle bit, because some of our
audience our viewers, listenersmay have not had the
opportunity to share time withyou.
Let's pull back and talk aboutyour background a little bit.

Speaker 2 (02:01):
Sure background a little bit, sure, um.
So, patrick, I um am a recentuh, I've had a recent entry to
sort of the multi-familyindustry.
I've been an entrepreneur um,kind of an entrepreneur born out
of hubris.
I think of some some probablypoor career decisions I made
when I was younger.
I'm some canadian of indianorigin.

(02:22):
I went to school there, I did acouple grad degrees, um and uh,
in canada and in paris, and Iended up doing a lot of work
around econometrics, measurementand sorts of things that were
more management science oriented.
Um, this led me to some somereally cool experiences in tech,
including.

(02:42):
I was really fortunate to bepart of the senior leadership
team on the revenue side ofGroupon when that company right
in and around when that companywas going public which people
forget about now was the largesttech IPO after Google at the
time.
So we were actually beforeFacebook was 20 billion, and
then I had a bunch of businessesin between something in the
healthcare space and a bunch ofbusinesses in between something

(03:05):
in the healthcare space and alot of stuff that was called at
the time big data, right, sure.
So you know AI as we call itnow, and, I think, a lot of the
excitement around AI.
When we look at the likes ofChatGPT and the large language
models, a lot of it is really aninterface layer or sort of a UI

(03:30):
change, and I think the actuallike machinations of how to use
models to make decisionsconsistently have actually
pre-existed for a super longtime, and it's often been useful
to have programmaticdecision-making functions
whenever you're dealing withlarge volumes of data, short

(03:52):
decision-making windows andthings that are repetitive
enough for it to make sense.
So I was a business that wejust sold Well, not we.
I was chief revenue officer ofmaybe 2012 called Granify, and
granify was venture funded incanada, um, I'm now an lp and a
number of these venture fundssold for 80 million cash.

(04:12):
What granify did was granifydid um e-commerce uh checkout,
um optimization.
So it would determine whensomebody was coming to an
e-commerce store and then notlikely to buy because they got
pants in their cart or whatever,but now they're like, oh, these
look expensive.
So then you shoot them an offerat the sort of right time.

(04:34):
So that's the sort of stuffthat's been termed AI in the
past.
I think that right now there'squite a lot of hype in the media
because the chatbot interfacehas made this stuff more
accessible, so it's in manycases a solution sort of looking
for a problem.
So I've been working in thisspace and actually, you know, my

(04:58):
day-to-day operating businesshas been using conversational
agents as well.
But suddenly people think AI isgreat at making decisions.
When you look at thefundamentals of what a large
language model actually is, alarge language model is just a
token input-output mechanism.
Right, it just says the catjumped over, and then you leave

(05:19):
it blank and it says the dog.
It just looks for the mostlogical sequence.
That doesn't mean it'sreasoning, it just means that
it's making something come outthat looks plausible.
So these conversationalinterfaces, left in isolation,
are prone to error, prone tohallucinations.
Many of them, unless they'reattached to a knowledge
repository that is relevant inyour business, can actually be

(05:43):
pretty damaging.
If you're sort of like goingoff of what chat gpt says in a
in a raw sense.
And I'm, you know, since I'vebeen working in this space and
there's a lot of excitement, I'mgetting 10, 20, um probably
requests a month, withoutexaggeration, from you know,

(06:04):
folks that we work with orconnected to, or multifamily
operators who are looking for areason to use it, and probably
nine times out of ten I.

Speaker 1 (06:13):
My response is please do not use ai for that sure, or
do not try to yeah, andspeaking of the conversational
ai and and what is um, where you, like you, mentioned it's been,
you know, we've been doingmachine learning and big data
and running programs way, wayback Now that the UI ChatGPT

(06:38):
made it easy to converse andbasically program, because
that's what they're doing whenthey're putting something in
there.
Essentially a programmer yes,I'm always I'm really fascinated
, even in the work that we do.
It's like how the people thatare doing the work now have the
ability to basically programwith conversation.
But going back into yourbackground, you were involved in
slack and and they had some biginitiatives around

(07:00):
conversations and data andthings like that.

Speaker 2 (07:02):
Take us, take us through that around
conversations and data andthings like that.
Take us through that, yeah.
So I think the insight thatit's not much of an insight,
right, like it's like way easierto talk to someone and be like,
hey man, hand me a burrito thango on to an e-commerce website
or download the Grubhub app orthe Groupon app or whatever is
going to deliver it to your UberEats.

(07:23):
So I don't think it reallytakes a genius to figure out
that people want to converse forrequests rather than looking in
knowledge repositories.
So Slack was an interesting onewhere I was very fortuitously
kind of a fairly earlyshareholder.
We had a company that we'd soldto them and Slack actually you

(07:44):
know the full name, the, theacronym or take us through it.
Yeah, I know we talked about itat lunch, I thought just for
the viewers, um, but slack isthe um searchable log of all
conversations and knowledge, um,so slack's entire premise was
that it was going to be somesort of central database or
brain and they've done basicallyall the things that you would

(08:07):
call.
You know that now, if you lookat the assistance api and you
look at tokenization and pinecone and vectorization and
vector databases and vectorsearch, slack was doing that
stuff in a more plot along sortof way, which was okay, let's
take all of the stuff that's inthis word document and let's
make that queryable as well asmetadata.

(08:27):
So if you're looking foremployee leave policy, if Slack
can find that that's mentioned,or stuff similar to that is
mentioned, into any documentusing direct keyword lookup plus
fuzzy logic, et cetera, etcetera.
That's why people use SlackBecause people found that people
wanted to chat back and forthand share info and the.

(08:49):
You know, if I WhatsApp you andI, you know, I can, you know,
work with some of my teams andWhatsApp and I WhatsApp you a
file, um, the rest of thecompany can't see it Right,
correct Um, if they come in andwant to sort of use it.
Um.
So Slack always had a visionand actually a roadmap that in
the s1 was pretty like heavilycommunicated um, where they

(09:09):
really wanted to become um kindof an ai assistant, like that
was really the intent of slackbot, which sits in here which
gives you all the alerts um forlike approvals and this and that
, and workflow automation wasbasically to become a knowledge
repository, that search for thatstuff automatically being like
oh, it looks like you've justjoined, just so you know this is
our medical policy, or whateverit is.

Speaker 1 (09:31):
Yeah, it's interesting because a company
may have things in theirdocuments if we're talking AI,
maybe things that they'reputting into retrieval libraries
and, to that point, those arethings that have been documented
and the large language modelshave documented things as well
but there's this invisible stuffthat happens in private
meetings and conversations andit's really interesting.

(09:54):
You mentioned the WhatsAppmessage.
Like, the rest of the companycan't see it because it's
invisible.
Sure, one-to-one, one-to-oneright, and there's opportunities
, I think, in business that areinvisible to leaders.

Speaker 2 (10:11):
Yeah.

Speaker 1 (10:11):
You see the invisible .

Speaker 2 (10:13):
Yeah.

Speaker 1 (10:15):
Take me through.
How do you spot these things?
Seeing the invisible Well yeah,because think about it, In
multifamily, it's thedevelopment process.
We're very patient,understanding like we have to
plan ahead.
We have to think.
We're really good atintegrations.
We have multiple disciplines,from plumbing to architects.

(10:38):
We have plans, we follow thoseplans, we have regulation and
we're patient with that capital.
We have to think about thecustomer who we're building for
all that stuff.
But when it comes to technology, we rush in some decisions and
that's why we're talking aboutwhat you know, when to not use
AI.
Yeah, you know, because youknow you have to unwind things

(10:59):
and stuff like that as businessrolls out.
But I'm curious to talk to youabout how, when you see a
multifamily portfolio, you'reseeing what others don't see in
many cases.
And what is that?
Knowing first principles,knowing what's possible?

Speaker 2 (11:19):
I think so.
I think the multifamilyindustry is interesting for a
number of reasons.
I mean, obviously it's the it'sresidential real estate is the
biggest industry on earth and Ithink it's enormously capital
intensive.
So when it's an industry isthat enormously capital

(11:40):
intensive and has that manystakeholders, enormously capital
intensive and has that manystakeholders, um, there gets to
be a lot of um, a lot ofcompetitiveness, a lot of loss
aversion, a lot of like focus on, um, uh, visibility throughout
the value chain, and I thinkthat rigidity and visibility and

(12:01):
sort of almost territorialhabits that's begin to sort of
come into play can be adisservice to our industry at
various points in time.
This is another thing we weretalking about in lunch.
You know, when you're talkingabout, like the PMS landscape
and building a potentially trulyopen source sort of PMS.
You know, you know the theadvent of Yardi, you know to, to
my knowledge, in terms of howit got to be so pervasive in the

(12:23):
industry, was it started as anaccounting shop?
So it was, from what Iunderstood used to be a CPA who
was working in the US and workedpredominantly with multifamily
owners and was doing well, thenbuilt software around automating
some of those processes and thebig institutional funds got

(12:47):
used to sort of the reportingframeworks that he was doing and
there you've got this sort ofthere you've got this sort of
like you know this accountingsoftware that people are really
relying on, which then snowballsinto it, turns out a you know
ERP and a CRM and yada, yada,yada Um, even if it's not, you
know, perhaps the absolute bestsolution for each one of those

(13:08):
things and has chosen to be very, very sort of closed off Um.
And I think that closeness andrigidity, which is a function of
sort of the capital stack andthe capital intensivity um of
the industry, you know,sometimes hurts a pragmatic
person who might walk into theindustry, who is sort of newer

(13:33):
and I think you know newerentrants and, more generally,
more pragmatic people, would seethings very differently from a
technological lens becausethey're not sort of caught into
the monolith.

Speaker 1 (13:47):
Yeah, you know, the more and more time I spend in
the AI community talking to somereally bright people, I notice
the collaborative approach whereI mean developers understand
that they build on top ofsometimes other developers' work
and that makes for a greatermarketplace network effect
product.
And you mentioned thatsometimes as leaders we end up

(14:12):
staying with what's familiar,what's safe and it makes sense.
I mean you know you're fightingfires, you know you're dealing
with what's urgent and important.
Um, and you know, back way backin the days of where the
internet came along and it wasprint, it was like you know,
your phone number was an asset,right, and so you had printed it

(14:33):
on certain things and so, like,changing a phone number would
be problematic because you'dhave to then do all this other
stuff that can feel like adistraction to the business,
right, so you're kind of stuckor you feel sticky in some way.
And to those providers that havebuilt those businesses, I mean
good for them for creating a.
I mean you know what awonderful founder story to be

(14:57):
able to build demand andretention and all that stuff.
But I think where we go next isreally interesting because you
know for an executive to trulyunlock value within the
organization, they need to seedata.
You can't do AI without data.
Correct Right, and so that's aconversation that's not solved

(15:19):
in one meeting, but it's where.

Speaker 2 (15:22):
Yeah, absolutely, and it's interesting because it
isn't you know.
So, when we were a group onback in the day I don't know
this for sure, but I heard itsaid colloquially so don't quote
me on it as for sure true, bynumber of records, because we

(15:46):
had um in chicago, six hundredwest chicago, we had three
thousand sales reps who werecold calling every spa and
restaurant and whatever.
We sped up a ton of data and aton of ton of records and we did
a.
I remember us doing a, amigration taking from like leads
to accounts and like changingthe disposition of how it's
working.
And we took down salesforce,like global, uh, took down their
servers, like it.
It was catastrophe, not justfor us but for Salesforce
globally.
But interestingly, you know, itisn't unusual for CRM providers

(16:10):
in all industries and Salesforceused to be like this to, once
they've got your data, say thisis ours.
So Salesforce at the time hadlike very limited ability to
export.
They had very limited abilityto do APIs, all this sort of
stuff.
They had rate limits that weresuper heavy.
They'd cite, you know, vagarieslike oh, compute is expensive,

(16:33):
blah, blah, blah, blah, blah.
But really it was like we wantto lock you in, yeah, we want to
lock you in and I think that'slike very much um for better or
for worse.
Like I don't want to, I don'twant to hurt anyone else's
businesses, but um seems to bethe case with with sort of the
large pms providers here.
Yeah, and it really has beenquite um cost prohibitive and

(16:57):
you know, I know that entratarecently introduced a fee as
well.
Um, but sometimes it's 20 to 50000 a year to be integrated
with the rd.
Sure, try to are like one ofthese groups.
And it's not even about thecost.
You know the the api.
The process of being approvedfor the api firstly requires you
to have a customer, a jointcustomer right, who is

(17:18):
aggressively advocating for youyeah, um, also could be a
competitive environment too,because some of the other thing
they they ask for your use case,they ask for your financial,
they ask for, like, a lot ofstuff that's pretty invasive, to
be blunt, and that stuff isreviewed by the corp dev teams.
Sure, that's known to be thecase because and then they say,

(17:39):
okay, well, we'd like to let youon the platform, but like that
may be influencing their ownsort of product roadmap.

Speaker 1 (17:45):
Sure, yeah, no, absolutely.
And and you know as we talkthrough those things, what it
really comes down to is, whenyou pull the lens back, you know
you're trying to create valuefor the end customer, the
resident, right, and so if wewant to unlock innovation I
believe in the industry then youknow we can get more efficient
with many things in the business.
That allows these real estateoperators to not only depend on

(18:08):
one source of income Right,because with limited units you
just can't.
I mean you could build more ifyou found land and all that
stuff, but on your site you have300 units.
Yeah, you have 300 rentalchecks that you could
potentially get.
Yeah, you have 300 rental checksthat you could potentially get,
and that's where, I think,where you're finding alpha in

(18:29):
other ways.
That gives relief to be able to, you know, fulfill obligations
that these real estate companieshave made to investors in ways
that could be significant, Ithink, material.

Speaker 2 (18:43):
Yeah, yeah, I mean, let's get into it.
You know.
So, with, with blue lake, whatwe're doing is we really we're a
alternative income partner for,um, multi-family operators.
Um, we're we're saying to themyou know, get a 10 000 foot
level.
Basically, people uh,multi-family investors are asset
driven, hard asset drivenpeople and they have their

(19:03):
assets, which are their land andtheir brick and their mortar
and their elevators andwhatever's in there.
But they have another assetthat they tend not to value,
which is data, customer data andinsights and the contact
information people areinteracting with their
properties.

(19:25):
Fortunately, and though realestate operators themselves
don't value this data, it's veryclear that the financial
markets do.
You'd only look at thevaluations of Facebook and
Google, who own no real estatebut own customer data.
Sure Right.
So should you have the abilityto just sort of like monetize
customer data?
That's obviously something thatputs an enormous sort of

(19:45):
multiple in your business.
Now, most real estate operatorsobviously don't come from that
background.
You know, it's a totallydifferent way of operating.
It's totally differenttechnological stack and skillset
required.
We've been partnering withmultifamily operators in the
most non-invasive ways possibleto help them to generate income

(20:10):
streams from customer data thatthey're housing at no cost to
them.

Speaker 1 (20:18):
They're actually more like vendors to us and we
revenue share with them.
Yeah, and in your network youhave access to some of the
brightest minds in AI andintelligence around this stuff.
So it's like you know, if youhave 2,500 units, you know
ideally that data scientist oryou know the engineer is likely
not going to want to come intothat organization and reshape it

(20:38):
and reimagine it, becausethey're going to be working for
teams like yours and doing bigreshaping industries and and
seeing greater upsides.
And that's why I think it'sinteresting what you're doing to
provide a lot of value, to sortof be this outside arm that
comes in and value engineersfinancial opportunities that
otherwise people don't even knowis possible yeah, certainly.

Speaker 2 (21:02):
I mean, the fact of the matter is, anybody who's
going through a moving event, umis going through a major major
life transition.
You know, if you're, if you'removing, you're moving because
you're some plans, someunplanned.
Some plans, some unplanned,unfortunately, yeah, um, and
hopefully positive, but somealso not so positive.
But you know, let's, let'sstart with the positive ones.
You know you're getting marriedor you're having a child or

(21:23):
you're.
You know you're starting a newexciting job and in Chicago,
from New York city, where rentis one third Um and uh, and you
know you're, you're keen togetting moving forward with life
.
But, like you know, whensomebody is going through a
moving event, um, people knowthat this is when all new
purchasing behavior is born,right, so I'm moving to downtown
Chicago from Manhattan.

(21:45):
I need to get an apartment, butnot only do I need to get an
apartment, I need to find a gym,I need to find a supermarket.
I need to.
I might get a new credit card.
I might do, you know, like,quite a lot of things.
I might find a romantic partner, I might want to be on tinder,
I might, because those things,once I've lived in that unit for
long enough, maybe I meetsomeone there.
Maybe now I need a bigger place, maybe I need to purchase a

(22:07):
place, maybe, you know, maybe Ineed a mortgage.
All these sorts of intents umcustomer potential, customer
purchasing intents that exist,sure, um in and around a moving
event, um, but the difficulty is, how do you become that trusted
guide or that trusted personthat takes them through?
So, um, blue Lake's use ofconversational AI is basically
by being um a conversationalfront end um, which is a

(22:32):
concierge Um, the core conceptof which is empathy, like an
extremely empathetic conciergeconnected to um what I'd call
SKUs or SKU databases, whichhave various different types of
products for different thingsthat people could need based on
what they're doing.
We roll up that income streamand we do it at very good

(22:53):
margins and we give a chunk toour operators.
I know you're about to ask howmuch?

Speaker 1 (23:00):
Well, no, no, I mean, the thing is, this is happening
already anyway.
Right, so it's just a matter ofyou mentioned the enterprise
value of, like the biggercompanies, like companies that
do get the data right Yep,there's, and we're talking about
AI too, and when not to use it.
In some cases, not having thedata makes it very hard to you

(23:22):
know, not use AI or leverage itin the in the right places, and
so I start to think about youknow, as you, as you move
through these opportunities forcompanies, you know what, what,
how, how should they thinkthrough them?
Because they they're not comingwith this knowledge to the
table, right, they're, they're,they're looking at their cap

(23:43):
rate, they're looking at theirbudget.
They're looking at their caprate, they're looking at their
budget, they're looking at theirday-to-day operations.
What are some of the questionsthat they should be asking when
going through something?

Speaker 2 (23:52):
Yeah.
So I think there's what we do,which is Blue Lake, which is
taking these lost cause, renterinquiries or move-ins that we're
not monetizing anyway, andattempting to monetize it.
I think that's a great use casefor AI.
The reason I think that's agreat use case for AI is it's
very, very high scale.
It's very repetitive,responsiveness is extremely

(24:15):
important and the stakes arepretty low, meaning if I get it
right it's worth something, butif I get it wrong, I mean the
thing was already worth zero, soit's not like something is
catastrophically gone wrong.
What concerns me the most iswhen we see multifamily
operators talking to us aboutusing AI in their core
operations and they're eitherpicking something that is not

(24:39):
repeatable enough to warrant itSure, it's not scalable enough
to warrant it or, and here's thescariest one if they put in ai
and it hallucinates, somethingcould go catastrophically wrong.
Yeah, um, I had a developer umapproach me recently.
Um, who is who?
Uh had a team that did you sortof stuff, taking B minus and

(25:05):
turning him to B plus oranything, and what he was
concerned about was codes,building codes and building code
stuff, so sort of trying tobuild a conversational agent
that you vectorize or put inpine cone or something and all
the codes and the rules anddocuments, and then said, hey,
like can we build this X?

(25:27):
And like get a yes or no or geta, you know, get feedback on it
and get citations.
And I almost had a meltdown.
I was like, yes, but like, howare they doing this so far?
He's like, well, they would,you know, it's a lot of
documents, but they know thesimilar terms and my, you know,
I have a team that does this andthey do keyword look, they

(25:48):
basically do control F and dokeyword lookups.
Sure, I was like, please stilldo the keyword lookup, right,
because LLMs are not, they'renot meant to return correct
answers, they're meant to givenan input of tokens, given an
output of tokens that looksimilar.

Speaker 1 (26:07):
Sure, Generate If you literally asked it.

Speaker 2 (26:11):
you know, am I allowed to?
You know, do I need to have anylow-income housing here?
It could say it would say no,you're not allowed to have any
low-income housing.
And then if you write to itback and say actually I am,
it'll say oh, yeah, you'recorrect, it's actually yes,
because it's not trying tofactually answer the question.
It's trying to give you a tokenof text.

(26:35):
That is logical given thepreceding token of text, and the
preceding token of text is ayes or no question.
It's going to give either yesor no, depending on what it
thinks you wanted to say.
You know, it's kind of right,kind of like um, I'm sorry for
for married people out there,like I I sometimes do this, you

(26:55):
know.

Speaker 1 (26:55):
Uh talk about she's like can we go to dinner?
Yeah, yeah, right, you knowwhat I'm saying.

Speaker 2 (26:58):
Short answer so it's sort of like you're not
listening but you're trying toyou're not really listening to
it or give an actual answer, butyou're trying to you're not
really listening to it or givingthe actual answer, but you're
trying to give a agreeablesounding response.
So that's kind of what, like,large language models are doing.
And, yes, with vectorization ofdocuments and knowledge
repositories, certainly, likethe risks have, like it's become

(27:19):
more accurate.
I would say it's definitelymoving in the direction of more
accurate.
I would say it's definitelymoving in the direction of more
accurate.
But you know, when we are in anenterprise multifamily context
and you're talking about ahundred million dollar physical
asset and you know somebodywants to save, you know, five
hours a week of time for like awell-trained staff member who

(27:40):
knows how to do this stuff andhas a lot of it says I don't
know.
This doesn't feel right, like Ifeel like yeah, we did talk
about low income in that meetingsix months ago when we were
looking at the plans.
The AI is not going to havethat knowledge.
And I think my major use casesare like when people are trying
to replace AI, use AI to replacebasically our search function,

(28:02):
a control F or a keyword lookup,I'd say vector databases are
not great for that.
Um, and god forbid, it's noteven relying on a vector
database and it's just doinglike some sort of google search
or bing or something like thaton the back end.
Yeah, um, then you really canbe in trouble.
So, um, I think ai is effectiveif you can constrain it to

(28:26):
really mimicking the human right, if you can say, hey, like I
don't want to run a vectorsearch, like you can actually if
you were using Lanczain orLanczmith appropriately, instead
of saying, hey, here's this300-page document, answer this
question.
You could say come up with the10 words that I need to use to

(28:49):
answer this question.
I could use a potentiallow-income or Section 8, and
then run a Control-F functionfor each of it and then respond
to me the chunks of text.
That's really what your adminperson is doing, right?
Sure, that's the safer way togo and I think people get.
They think these are sort ofgod models and they get, you
know, a little bit overzealousyeah, you need to start with

(29:12):
your process right, how, how,how is work getting done?

Speaker 1 (29:15):
I always say that there's a difference between
automated and autonomous.
Yes, and I think it's a littleuh ambitious to to be autonomous
in some of these situationsthat you're talking about,
because then you're you know,you know people begin and end
the process.
There's some level of review.
Then you can automate the stuffin the middle, yeah, but it's

(29:36):
still the.
The leader is still making adecision from.
You know it's more assistantthan artificial at that point,
but, um, yeah, it's veryinteresting.

Speaker 2 (29:46):
Yeah, and there are human in the loop, frameworks,
right so, where you can do that,and then it has to go to
someone for approval and thenthat person sort of says yes,
but I just get really.
I get really worried with thisstuff because so many people
think this is how they're goingto operate their business and it
like there's a reason that youhave that person who's been
reviewing building codes withyou for 20 years and costs

(30:07):
$100,000.
It's because they've been withyou for 20 years and they've
seen all these things andthere's all of this stuff that
they're intuiting around.
And offloading this stuff whenit's super high stakes if
something goes wrong is ahorrendous idea Right, and let's
talk about frameworks shall we.

Speaker 1 (30:26):
I mean, what are the frameworks we should be thinking
through?

Speaker 2 (30:31):
So I yeah, I think it's how to say it, it's just
that the use case has to becorrect.
So I think that first thingsfirst, don't try and skip what
the labor is, what the personwho's doing it today is doing,
don't try and leapfrog them, tryand mimic them, right.

(30:52):
So if the person is doing acontrol F search, program it to
do a control F search.

Speaker 1 (30:56):
I will tell you.
What's interesting is just thisis where I think AI can be a
benefit without even using AI.
The idea of using AI is youfirst need to look at your
current process.
Who's doing it, where's thedata, what tools they use?
You know all that stuff, right,and we did this.
We have this framework, we takepeople through.

(31:18):
It's like a sort of like a timeassessment thing, right?
And you can't do that until youunderstand, okay, well, who's
doing the work, how's it done,what steps are they taking?
And then there's the actionsthat you'd want to have happen.
And it was interesting thatjust by getting the just
journaling what was done for theweek on these things and what

(31:39):
tools were used allowed us torealize like, wow, we've got
four people we're paying highdollar amounts sitting in a
60-minute meeting every week,and so originally we just said,
well, let's just stop doing thatmeeting.
So, without even using AI, tothe fact of getting ready to use
AI, we found money.

(32:01):
We didn't even use AI yet,right?
And so when you talk aboutmimicking the work, like there's
value, because sometimes thesethings creep into the
organization, people love tosurround themselves with work
and Busy work, yeah.
Yeah exactly, and the CEOs don'talways know that.
Sarah, three departments downwhen they say do the thing, that

(32:24):
they've got to go into fourdifferent systems and pull out
and download and upload and dothe thing.
They don't have visibility overthe work.
That's what I love about thisis that because we went from a
leadership responsibility whereyou could see work like you had
an office and you knew when theywere there and you saw them
working and they were in theboardroom with you and you saw

(32:46):
them on calls we visually seethe work.
But now, with even remote workand people in different areas,
we don't often see where workgets done.
But when you get these thingsinto tables and into databases,
into systems, we have a wholenother visibility of how work
gets done.
And I think if executives knewthey could be trained to find

(33:08):
value within the organization.

Speaker 2 (33:11):
Completely.
Yeah, I mean, if your need touse AI forces you to first do
this pre-work, that's a winenough.
You know what I'm saying.
Right, versus you to first dothis pre-work, that's a win
enough.
You know what I'm saying?
And this sort of like whatyou're describing is to what
extent some extent.
Recently it's called firstprinciples, thinking, just
data-driven, operating ingeneral.

(33:31):
I think it's probably morevaluable.
And then AI is sort of like atool set right, sure, Kind of.
At the end of it, that like mayor may not be relevant, but
probably 98% of the win is justadopting a culture that's
looking for efficiency andthat's really breaking things

(33:51):
away from roles and into sort ofaccountabilities right.
And sort of saying like we do weneed a head of marketing or do
we need leads to apartmentbuildings?
Right, Like do we need a headof marketing or do we need leads
to apartment buildings?
Because a head of marketing mayresult in leads to apartment
buildings, they also may not.
It might be, you know, aperformance marketing person
who's not a head of marketing.
It might not even be aperformance marketing person, it

(34:13):
might be an agency.

Speaker 1 (34:14):
Yeah, and we have our biases coming into these
conversations because we grew upand this is the other problem
that you may be bumping intomore than I is as people are
considering these things,they're quite successful.

Speaker 2 (34:26):
Yeah.

Speaker 1 (34:27):
And they've been cashing million-dollar checks
for a long, long time, and theydid that in a world that AI
didn't exist in some cases Imean obviously in later years
here and so it's tough to makechange when you have so much
certainty around how the wealthwas already created, Like what's

(34:47):
the incentive?

Speaker 2 (34:49):
right, yeah, the incentive is that Jerome Powell
just continues to raise ratesyou know that we're all going to
get pretty creative.
Yes, Out of necessity, rightOut of necessity.
I mean, necessity is the motherof invention, right.

(35:16):
Flow while being very liquid,yeah, um, and so, yeah,
necessity is the mother ofinvention, right.
So, like the second, you um,things get tough.

Speaker 1 (35:25):
You gotta go to the mat and you gotta find a way
yeah, I, I love the theconversation around uh well,
self-driving, autonomousvehicles, uh, obviously very
data-driven.
I recently uh went and touredone, or actually went on a ride
down here in Phoenix, sort of aWaymo.
Yeah, the Waymo, have you beenin one?

Speaker 2 (35:48):
No, not yet.
I'm going to do that later.
Yeah, you will.

Speaker 1 (35:50):
You will.
It's pretty interesting, and sohere's the deal.
At first, I'll tell you a quickstory.
We were in a sort of weupgraded to uh in a ride share
uh, current ride share, uh,vendor or product.
We summoned our ride andhalfway through the ride, um, we
realized, uh, it got scary,like literally, like you know,

(36:16):
aggressive driving, okay, andthis was a, like a five-star
driver, and I was like I wasn'texpecting that, oh, this was
with a five-star driver.
And I was like I wasn'texpecting that, oh, this was
with a human.
This was me in the ride, yeah,so with the, exactly.
And what happened was thatdriver got a new ride and wanted
to finish my ride to be able totime and accept the other one,

(36:38):
sure, so the interests weren'taligned to our safety, and so
it's funny because we werehaving brunch.
And then we're like, oh, let'stry the Waymo, right?
So we got the.
I'm like, well, we'll fix thatproblem.
First principle is thinkingright.
Elon says you know, engineersoften optimize something that
shouldn't exist.
Well, in this case it was thedriver.
So I'm like we can do this.

(37:00):
We got the app summoned, aWaymo came and immediately when
I hit the app, it said and thiswas brilliant marketing, it said
.
The most experienced driver ison the way and I thought you
know what.
This thing is probably drivenmore than I have.
Yeah, you know what I mean.

(37:21):
Probably millions, millions ofmiles or something, right,
because it's just always drivingand it's probably paying
attention more than me and stufflike that, and I know that
there's a lot involved in that.
But I start to think about whenyou look at what feels safe,
because we're familiar, likethere's a steering wheel.

(37:42):
I would say the Waymo Nex wouldnot even have the steering
wheel, right, because why wouldyou need that if right?
So, but right now it's there.
And I always tell our teams andpeople I'm sharing time with,
like, in the state oftransformation we talk about
things as they used to be, forus to even have context for what

(38:03):
they are.
So, cordless phone, you knowright now, self-driving car it
was horseless carriage, I meaneverything, motion pictures for
cinema and all that.
And so at first it was apicture, then it moved and now,
well, what is wait, it's a video.
No, we don't know, we're noteven to video.
It's like it's a motion.
Then it moved and now, well,what is wait, it's a video?
No, we don't know.
We're not even to video.
It's like it's a motion picture.

(38:23):
Yeah, and in the states thatyou're moving people through, I
got to imagine that there's alevel of context to what it was
and what it can be.
Yeah, do you feel liketransformation is as important
as AI?

Speaker 2 (38:44):
More, yeah, yeah, I mean you feel like
transformation is as importantas ai.
More, yeah, yeah, I mean likeit's, like it's always, it's
crawl, walk, run right like youhave to.
You have to start small, youhave to get it right, you have
to build the trust.
You gotta have the steeringwheel there for some period of
time when you start seeing thatthe thing's not going to run
into the road.
You know that's when you startdoing it.
So I, I think it's, I think theit's great, that it's exciting
and everyone's like ai, this, ai, that, whatever, but it it

(39:05):
really is, I think, a mentality,and you know that mark
suckerberg's been saying it andall these people like look,
facebook cut 10, 20 000employees.
What happened to revenue?
Nothing.
Earnings exploded right, likeyou know, in Elon's words, like
removing things that most peopleare optimizing, things that

(39:27):
shouldn't exist.
Probably many of the rolesdon't need to exist.
But that's not an AI statement,that's a cultural statement.
The fact of the matter is thatthe majority of the world is
doing something that I wouldstate whether that's a head of
leasing, whether that's aleasing agent, whether that's

(39:48):
even asset manager to someextent are doing something that
can be summarized as aconversational agent who looks
at a database, right.
So if I'm ahead of leasing, Isee a thousand leads coming in.
I was trying to see how manywent to tour, trying to see how
many applied.
And I call the agent who has alow tour to close rate and say

(40:11):
what are you doing, you know, orhow, why didn't you get this
many people to tour?
And I talked to him, right,that's a conversational agent
with a database, had to ask themanagement.
I say and rent rolls good, rentrolls not good.
Renewals are good.
Renewals are not good whencompared to the 10 other
properties.
Because I'm running aspreadsheet.
You know here's 50 of my assetsand this one's got extremely

(40:34):
high NOI and this one has lowerNOI and this one the least
velocity is this and that one'sthe least velocity, that this
one's doing comparably worse.
And then I reach out to theperson and I say, hey, you know
Green Gables apartments.
You know you guys aren't doingso great, look at these guys.
You know in another state, butthey're outperforming in least
velocity, they're outperformingin occupancy, they're

(40:55):
outperforming in a.
Y, what's going on, and I tryand have a conversation about
that.
I think most of those thingsare ripe for at least first step
automation, right In terms oflike-.

Speaker 1 (41:09):
We don't get the defensive behavior.
Sorry, you don't get thedefensive answer.
You get the right answer, youget the right answer.

Speaker 2 (41:16):
So I think they're ripe for at least first stage
automation where there's noreason every dashboard,
dashboard, every spreadsheetthat everyone's looking at
shouldn't be, instead of themhaving to log in to look at it.
On the flip side of an alertbased system that says you know,
anytime one of my properties isdoing super well, shoot me a
note.
Right, and it just sort of, andyou know the llm can shoot like

(41:40):
sure, this one's good.
We recommend you reach out toCatherine, who's over there and
and here or whatever you know.
Would you like me to do it, yesor no?

Speaker 1 (41:48):
Yeah, here are some good questions.

Speaker 2 (41:55):
Here's some good questions, and as long as you're
doing taking any majorlyconsequential decision like that
, you know we're talking about apositive scenario.
Let's talk about a negativescenario, where katherine's
actually an underperformer andthe algorithm says let her go
and reaches out and terminatesher.
That's a catastrophe, right,right, sure, um, because maybe

(42:15):
there's a bunch of extenuatingcircumstances.
Forget that.
Maybe you don't want thealgorithm you know involved in
it to like that extent let'slean into the human in the loop.

Speaker 1 (42:23):
Uh framework, what, how do you, how would you share
um?

Speaker 2 (42:29):
reflect on that.
We can show you.
We can show you in our systems.
But I mean, you're familiarwith um lang chain lang smith,
right?
So lang chain lang smith, a lotof these things have um what
we're talking about.
Are you familiar with agenticframeworks?
I'm not.

Speaker 1 (42:43):
No, okay are we talking about?

Speaker 2 (42:44):
embeddings of no um agentic frameworks are born out
of processor vision.
You're familiar processorvision, no, okay, so the the
reason that large languagemodels um, are so exciting isn't
because it's a chatbot that youcan shoot.
The shit with which I mean, ofcourse it's cute and whatever

(43:05):
you know, you can make jokes toit and say, hey, give me this
song and the style of jay-z andwhatever, and like all that
stuff is great, but it's thatyou can use it to become a
decision engine and the reasonthat you can use it to like an
actual end-to-end, close toautonomous agent, and the core

(43:26):
reason for that is because ofsort of chain of thought
prompting.
You feel the chain of thoughtMm-hmm.

Speaker 1 (43:31):
So basically, the way you, it's just like a thread
for yeah, where how it came to.
So, like you and I rememberwhat we said to each other yes
In how it came to this.
So, like you and I rememberwhat we said to each other yes,
in the previous context and wecoax each other.

Speaker 2 (43:43):
So, for example, let me give you an example where it
says you know what is the what'stwo times 10, right?
So there's two ways of makingthe model work.
When it says you can say what'stwo times 10?
And it'll say 20.
So if it says two times 10 is20, what that's doing is that's

(44:05):
just doing token extension.
Saying two times 10, 20 lookslike plausible.
It might give you, might justas well give you 30 or 40 or 50.
But it says I would like tocalculate two times 10.
Um, please give me the steps toget there and I'll say two plus
two is two times Two is four,and two plus two plus two is six

(44:31):
and two plus eight.
So this is what we call sort ofchain of thought, which is like
getting it to explain how it isderiving and it's changing, it's
getting to what it's.
What it's saying is ultimatelythe answer and ultimately, all
you know, this is some of thestuff that well, now is exposed

(44:52):
in the media but like wasn'topenly talked about before.
You know, people look atopening eyes.
It's sort of a tech company andit is very much a tech company,
but like that tech was trainedby tens of thousands of humans,
principally in kenya, actuallyin nigeria, some markets where I
actually have some offshorestaff as well who are paid like
two bucks an hour to like ratethe appropriateness of answer.

(45:14):
So it's sort of like two timestwo, two, you know, two times
ten is 20, correct, yes or no?
Now what they've done is nowwe've started moving from two
times 10 is 20 to two times 10is 20, explain your thoughts and
says well, two plus two is four, two plus two plus two is six,
and like kind of do thebreakdown and it says yes, I

(45:35):
agree with the reasoning processyou took to come to this.
So that second type of thingwhere we've gone from input
output to showing the process ofreasoning and then supervising
that process is called processsupervision.
So process supervision is aboutchanging reward models from

(46:02):
being token input to tokenoutput, to being like what is
the process through which youtook this token input and
ultimately came to that tokenoutput?
And many times that reasoningis taking place in the
background and you're not seeing, you're not and OpenAI is not
showing you those steps.
Sure, right, like, you can coaxit to show you the steps as
well as a user if you want, butlike, in actuality it's doing it
without you those steps.
Sure, right, like, you can coaxit to show you the steps as
well.
As a user if you want, but like, in actuality it's doing it

(46:24):
without using those steps.
So we're seeing sort of liketool selection, like this sort
of stuff happening as well aswhere it's like, hey, like,
what's the?
When was Nelson Mandela born?
You know, was Nelson Mandela agood guy?
How old is he?

(46:46):
So, like he'll realize, nelsonmedela is a good guy, sort of a
subjective thing.
If lm tokenization okay, yeah,and generally people speak about
him positively, he's a good guy, great, and it says and how old
is he?
He's no longer around,obviously, but you know it'll go
like okay, well, now that is afactual answer.
For that I have to go throughtool selection, do I?
And then I have an api for bing, because bing is the, the
integrated api at the momentsearch api, go through this was

(47:07):
the age.
And then, okay, now that I havethat, now I'm going to take you
, take the birth date and deriveit by today's date and say you
know, he would have been thismany years old, but
unfortunately deceased in thecase of nesmantella, um, so it
it's about cultivating rulesengines over periods of time.
So like, if I take that and Itake a step back from that and

(47:28):
let's come back to the personwho's sort of doing the
multifamily zoning stuff, right,like the developers looking
doing zoning so it's sort oflike, yeah, okay, it's coming
out that this one says nolow-income housing allowed, but
is that consistent with whatwe've seen?
Let me second-guess myself.
So it's like, okay, yeah,that's the answer, but let's

(47:54):
triple-check that answer.
Okay, well, let's look at allcomparable projects we have.
Do we have many where nolow-income housing is allowed?
Do we have other buildings inthe area as low-income housing
allowed there or not?
And that sort of like secondthird layer, like the model
questioning itself.
Inference is really wherethings are going to go.
But if you talk about that andyou know there's a, there's

(48:16):
somebody that I know somebody'scome to my talks much more
successful founder than me, guynamed uh mike merchants and
who's in canada.
His company called ada.
It's now valued at two billion.
Um, he talks a lot about how to.
He's been doing similar to usconversational ai in his case in
customer service, for theyhandle like all of air asia, for
example.

(48:36):
But he was saying that one ofthe things that mike says which
I think is great is that youreally have to treat your ai as
an employee so you know when weget back to the person who's
looking at, you know thosethings for code, you know
building codes, you know.
You trust her blindly becauseshe's been working for you for

(48:56):
15 years.

Speaker 1 (48:58):
Yeah, you didn't trust her blindly 15 years ago
right, because you've had manycorrections, adjustments,
investments, corrections,adjustments, knowledge
repository.

Speaker 2 (49:06):
Good things, she's had process supervision yeah,
she's had.
Like you are doing a good jobof this.
This is how you reason throughthis.
If something looks a littleunusual, you should double check
, triple check.
You should check with thisknowledge repository.
You you should look at fourother comparables, you should
use your sort of judgment andthe question becomes like how do
we inculcate judgment into AIagents if an AI agent is

(49:30):
ultimately going to becomebasically your employee or is
sort of like offsetting the workof what was historically done
by an employee and that is goingto be an arduous and continuous
investment and the precursor toall of that?
You rightly stated is culturalchange.
So if you take that culturalview that you're looking for

(49:51):
efficiency in your business, youknow that's a precursor to any
of this.
Right, when that startshappening, uh, you may find that
you never even get to the aipart.
Because if you found a way to,yeah, you train people to do it
and you find the ai isunnecessary, yeah, um, but I
think it's really a function oflike.
Do we want to find efficiencies?
Do we want to break things downinto first principles and think

(50:14):
does this role need to exist inthis way?
Right, um, do I really need ahead of like zoning, you know,
uh, zoning checks, is that likea, a role on an org chart that
needs to exist?
Or do I just need the zoneschecked Right?
Same thing Do I really need ahead of marketing?
Or do I need leads to myproperties?

(50:35):
Cause, at least my propertiesmay come from a head of
marketing.
They may not.
They may come from, you know, avariety of different ways.
Or maybe I don't even needleads to my properties, who
knows, right?

Speaker 1 (50:45):
Yeah, no, no.
Those are really great points.
So when I pull back as youreflect on this transformation,
I think about like, well, whatis a company?
A company, if you think aboutit going back historically, had
physical resources, buildings,things of that nature, and those

(51:06):
that grew in scale, had theability to bring together
capital resources, financialresources, loans, debt, all that
to really grow and scale.
Then, as the physical resourcesleveraging financial resources
grew we'll take a very easyexample Blockbuster, right.

(51:32):
More stores in more cities,more revenue, right.
So now you now have more whatEmployees, which brought the
third piece, which was the humanresources.
So you had physical resources,financial resources and now
human resources, and that waspretty much every company that
existed.
And now, as I speak to realestate owners and operators,

(51:53):
they're good at the financialresources.
They can bring JV dealstogether better than anybody in
the world, right, they can seevalue, they can see a piece of
dirt, have a vision for thefuture, make it happen, take the
risks.
Great at people like buildingteams, leading people, all that
stuff right, great at that pieceof things.

(52:14):
So they've got the physicalresources, they've got the
financial resources and they'vegot the human resources.
And then this fourth part isthis technology as leverage.
What I hope to do is inspirepeople to realize, as good as a
CEO knows how to do debt deals,cap all the things that pull

(52:34):
leverage on a real estate dealthat they need to speak to
someone like you, right, oranybody that's really joining
forces with bringing technologyinto a company not even AI, just
tech and look at it like it'snot something you give to IT,
like this is part of building ahealthy company.

(52:55):
That most companies aretechnology companies.
They just don't know.
And that's what goes back to mypoint seeing the invisible.
You see that fourth piece.
We were talking about oil andrefining, and if you could just
imagine all this value that'sunlocked in your business as a
CEO, where do you point peopleother than calling you up and

(53:16):
working with you on these typesof things?
But where do you point peopleto accelerate the learning, to
understand it's as important asunderstanding debt, equity,
financial resources and thesetypes of things that make up a
company.

Speaker 2 (53:30):
The number one thing is I think most people who are
at senior levels, like CXOlevels and multifamily operators
, specifically institutionalones, come from finance
backgrounds.
And how I learned to this stuffand I, you know, don't have a
formally coding background,econometrics background, um is
from building financial models,um.

(53:52):
So when I worked at city groupand I worked in investment banks
, we'd have to build largefinancial models where we would
do things like, say, we weremodeling multifamily.
You know, here's this asset andour rent assumption is three
bucks a square foot.
But my MD is going to come inand ask me well, what if it's
only two?
And I need to be able to changeone cell and it needs to flow

(54:13):
through and I had to be able totest sensitivities.
So most people who are, who canbuild a financial model that's
basic in Excel, are coding.
They don't realize they'recoding, but they're coding it's
a different interface.
It's a different interface andit's a simpler interface and
it's algebra and whatever.
If you can build a discountedcash flow model, you can use

(54:35):
zapier.
Are you familiar with zapier?
oh yeah, yeah yeah, zapier islike one of those most simple
like if this, then that rightconditional automation software
is on earth and there are somany low code solutions out
there to do things.

(54:56):
Once again, you know, in mybusiness I do one repetitive
thing at extremely high scale,right, 30,000 renter inquiries a
day coming through my you knowmy drive-through and I'm trying
to service them and find themapartments and do things like

(55:17):
this.
So I'm the equivalent of sortof like McDonald's, right?
So you know, mcdonald's is 100%a management science play.
How the heck are they makingmoney selling $1 McChicken?
Sure, well, they're doing itbecause they built such scale
and such efficiency, becausethey built a supply chain and a

(55:38):
machine that was purpose builtfor one repetitive action taking
place again and again, andagain and again.
That was purpose built for onerepetitive action taking place
again and again, and again andagain.
And they put thought into everyinch of that place.
Like the burger buns, you know,should they be together?
Do they reach to the left orright?
Should there be sesame seeds?
Should there not be sesameseeds?
If there's sesame seeds or notsesame seeds, what are the

(55:59):
consequences?
Right, if it's sesame seeds?
Should the bun be this way.
Should it be this way?
Should they be put together orshould they be on the separate?
sides of it.
You know we've got.
We got to get the perfectFrench fry.
How do we get the perfectFrench fry?
You know this is a labor oflove.
You know what they've doneright and they're.
You know they must've hadinconsistencies in the French

(56:22):
fries back in the day becausethere was, you know, coming in
and people would cut thepotatoes different sizes.
Now they're cut the perfectsize.
Now they put them in the thing.
I'm sure at one point somebodywas putting up and down the.
You know the fryer.
Yeah, that no longer happens.
It's completely One third of asecond.
Yeah, it's completely automatedbecause it needs to be exactly

(56:50):
precise, it needs to beconsistently done and they've
still got humans and they'redoing it and it's.
It's a beautiful thing, it's asort of symphony, you know, or
that.
So I think the, the it makessense to automate.
You know burger production.
If you're mcdonald's and yousell 30, 000, you know burgers a
day or whatever.
You're selling millions,probably globally a day.
But if you're, you know there'sa burger joint that's up in
Canada and Canadian called theWorks, and the Works charges $27

(57:11):
per burger and they pridethemselves on bison meat and
avocado and they'll put twofried eggs in there and they'll
put you're supposed to waitexactly and you can customize
thekinds're supposed to wait, like
exactly, and you can customize.
You know the kinds of likechili flakes you want, and this
and that, and sauces and thisand that, and like you know, if
I start building that with anautomated machine and I took

(57:32):
away all the options, all thevalues eroded, sure.
So I think, like a lot of theum, a lot of the tasks that,
like a multi-family operator, aninstitutional level, is dealing
with, he very well could useZapier and things like that to
automate some of them.
But, like I'd say, you know,buying a hundred million dollar
apartment building, it's a lotcloser to the you know the $27

(57:54):
burger than the $1 JuniorMcChicken.
Right, there's a few that don'tdo it.
So it's sort of like you got tofind the Junior McChickens.
So it's sort of like you got tofind the Junior McChickens.
Sure, what are the JuniorMcChickens in your business?
What is the basic repetitivehigh volume, low margin.
I don't really care about it,but they probably don't make
very much money on JuniorMcChickens.

(58:15):
It's fine, people come in andthey get them, and that's volume
, you're right, and they're ableto amortize a fixed cost of
running that operation andhopefully they'll buy some
higher margin product whilethey're there.
Yeah, right.

Speaker 1 (58:25):
That's interesting.

Speaker 2 (58:26):
So I'd say, find your junior McChickens and like
that's what you should beautomating.
But like the, the majority ofthe stuff in your business
probably isn't a juniorMcChicken, yeah, so don't go
chasing the automation.

Speaker 1 (58:38):
Yeah, you know it's funny, you, you mentioned Zapier
and I'm going to show you whileyou're in town here.
We kind of took that model andmade it for multifamily, because
there's no network effect in amarketplace like that, because
it's mostly for freelancers orpeople that are familiar with
how to connect APIs and thingslike that.

(58:58):
But if you take a marketplacelike that where you can connect
system A to system B and startwith everything outside the PMS
right, you start to think aboutan email signature and if your
turnover is 40%, you know thatis the thing that occurs.
And on and on and on, and youknow you're going to be
candidate experience resumescoming in at volume and things

(59:21):
like that, resumes coming in atvolume and things like that and
those are the types of thingsthat are the low-hanging fruit
or the crawl, walk, run stuffthat you can build a workflow
around.
Yes, you know, and so it'sinteresting you say that
because— that's your juniorchicken yeah, it's like a
low-stakes, high-volume thing,right.
And what we've found is thatpeople actually, when hired,

(59:46):
don't want to do that type ofwork, right.
So now you're, in a way,improving the employee
experience and repurposing theminto things that lead to either
more revenue or other parts ofthe business Higher value,
higher value, yeah, exactly,higher value things that you
need humans for, exactly,exactly.

Speaker 2 (01:00:03):
My mantra is aggressively automate,
aggressively humanize.
You have to be doing both,right.
Right, if we have people in theorganization and we're
automating away their work, it'snot so that they don't do any
work, right, it's so that theyapply the human layer, you know,
to that work.
So it might be saying, hey,give this customer this

(01:00:24):
apartment, but then weultimately expect, when that
message is delivered, for it tobe delivered with great care and
empathy and attention to detailand contextualization.

Speaker 1 (01:00:37):
We have some really great conversations in our
Innovation Council and, forthose of you listening or
watching Multifamily InnovationCouncil, we watching multifamily
innovation council.
We talk to multifamily ownersand operators on a weekly basis
around the challenges they wantto solve the priorities from
fraud, centralizing the business, property automation, business
automation, all these types ofthings and that's kind of how we

(01:00:58):
end up in conversations likewhat we're having is like we
first identify what's going tomake the business better.
That's the way we look atinnovation, not the new tech.
It's like what's going to makethe business better either by
driving more profitability,saving money, more revenue,
whatever.
That is yeah.
And so when, when we have theseconversations, we're like, well

(01:01:19):
, okay, well, who's solvingthese things?
That's kind of how we ended upbringing you in to have these
conversations.
But we have some really good,healthy debates because
everybody has this window bywhich they see the world,
because of A how they wereinfluenced growing up and then
also how they worked.
They get a good job, they do agood job, they get promoted,

(01:01:43):
rewarded, they go to the nextlevel.
Now they're in charge of others, they tell others how to do
their job and it sort ofinfluences this.
In some cases it could bebureaucracy, it could be waste
in the organization, just fromfamiliarity.
And so that's why these firstprinciples thinking kind of
unlocks some of those things.
Well, what if we could do that?

(01:02:05):
What if we didn't?
You know what if we could leasewithout depending on a human
we're not making the debate, youshould have one or not.
But if you didn't have todepend on it, what then?
And so these debates are fun,exciting, and I would love to
have you come in and we couldfacilitate some fun stuff there.
But we have some that arecompletely nobody on site and

(01:02:27):
some that are like how do I getto a new model, centralized,
specialized, whatever they wantto call it, like testing
assumptions about how work isdone or can be done.

Speaker 2 (01:02:40):
Yeah, you got me thinking and there's something I
want to share.
So obviously, as you know, Iwas an early shareholder in
Slack very, very fortuitous, andthat created some wealth.
Eventually, I was trying to bea real adult.
I just had a kid, you know.

Speaker 1 (01:02:56):
Growing up still.

Speaker 2 (01:02:58):
So I just had a kid and that you know, having a kid
and having to just sober up andlook at the world and how real,
responsible adults you know makemoney, it's what led me towards
multifamily, um, but I've hadthe good fortune that I I keep,
um, quite a bit of portfolioallocation.
So I'm a partner to venturefund in canada, um, and I keep

(01:03:18):
portfolio allocation towardsventure and tech and one of the
big reasons I get to learn stuff, um, and I'm a fortune being a
JP Morgan private bank client.
You normally have to have waymore assets than I have to be
there, but I don't know.
They were nice to me.
They seemed to be taking a beton me, a young guy who may get
there in the future.
So they're like, okay, this isa smaller portfolio, we'll take

(01:03:39):
it, but through that I got tomeet Tiger Global.
Have you heard of Tiger Global?
No, tell me more.
So, tiger Global is a NewYork-based hedge fund originally
that became one of the mostprolific late-stage venture
capital investors now in theworld.
So they go neck and neck withSoftBank, for example, and they

(01:04:00):
had a webinar and they weregracious.
I got to go to it, as did allthe JP Morgan clients and I got
to hear from Chase Coleman, theTiger Global founders, who were
on that call.
They said some reallyinteresting things.
So I think you were coming backto sort of like what is a
company?
And I'm sort of like I have amaster's in finance, right.

(01:04:22):
So for me I'm a complete, Ilike to call myself.
To my teams I say I'm a simplecapitalist, meaning I don't
understand rules or org charts,any stuff.
I just understand like where'sthe money?
Tell me where the money is.
So a company has only one jobreally, which is return on
equity maximization, right, like?

(01:04:42):
We have equity in the company.
We invested, working, you knowcapital, and now we're going to
deploy those.
In that argument, how do wedeploy it efficiently to
maximize value of the company?
You can do that by buildingmoats or things that you know
building revenue, building freecash flow, building moats around
that revenue and free cash flow, or perception of moats.
Um, and I think there was a.

(01:05:02):
There was a while, you know,when we look back 20 years ago
in my dad's era of investing,where you know the biggest
companies in the world mostvaluable were the ones that had
oil and now they're thecompanies that have data.
Um, and how did that happen?
And why did that happen?
Right, like, why do facebookand google and you know these
things work?

(01:05:23):
It isn't because, like,facebook is a cool app and
everyone's excited about the app, large scale, long.
Only, you know, a hundredbillion trillion dollar
institutional investors andsovereign wealth funds don't
invest in Facebook because theapp is cool.
Those people only and only andonly care about one thing, which

(01:05:44):
is free class for a generation.
Right, and you know, for all theflack around meta and Oculus,
and it hasn't gone so well, Ithink Zuckerberg was putting 10
billion plus a year into this ARbet that as yet, hasn't had any
great positive results from arevenue perspective.

(01:06:07):
Um, and the street couldn't sayanything to him.
You know why?
That is because he produces 100billion of free cash per year,
so he can launch 10 billion, buthe can put 10 billion bucks and
light it on fire and the streetcan't say shit.
So, um, when you're talking tochase coleman, these guys, the

(01:06:27):
thing that they were sayingwhich is really interesting
about, like, the businesses theyinvest in and the thesis
because they're hedge fund guys.
So it's like, what the heck arethese hedge fund guys, these
serious, traditional late stagepublic markets investors, who
are supposed to be like talkinglike Warren Buffett, what the
heck are they doing investing inFlipkart, which is the Amazon

(01:06:49):
of India, all these sort ofspeculative looking companies?
And they had a really clearanswer.
And the really clear answer wasthey said look, I love Tesla, I
have a Tesla, it's like I'vegot one in the garage.
We invest in technologycompanies.
Tesla is not a technologycompany, tesla is a car company.

(01:07:09):
Do they use technology in allthe wheels?
Sure they do.
What defines a technologycompany?
What defines a technologycompany is a company that maybe
spends $10 million a year togenerate $2 million a year, and
then it grows its revenue from$2 million to $5 million and

(01:07:32):
they're spending maybe $11million a year.
And then they grow theirrevenue from $5 million to $50
million and they're spending $12million a year.
Because what happens is, whenyou build a true technology
company, you're able to supportmarginal revenue at no
incremental cost, right?

(01:07:53):
So if I get to you know $20million revenue, I want to go to
200 or a billion or 2 billionor 10 billion or 20 billion.
You can basically do it withoutadding any headcount.
And there's giant private techcompanies that people don't talk
about.
Do you know what the revenue ofCraigslist is?

Speaker 1 (01:08:12):
And the employee count very low employee count.

Speaker 2 (01:08:14):
It's like 25 employees and it does like five
billion of free cash flow and itdoesn't even look.

Speaker 1 (01:08:21):
And how many people come in and say let's redesign
the site right?
Yeah, exactly.

Speaker 2 (01:08:25):
And it's just people paying that stupid $2 or $5 or
whatever for a Craigslist ad andit is a monster, it's an
absolute cash monstrosity.
So Tiger Global were saying weget it Because their investor
set are endowments and pensionplans and, like major insurance

(01:08:47):
companies and major, they'rethinking three generations, in
some three, four generations.
And then how do we get thistraditional investor set to
understand why we're doing this?
And they're like we're doing itnot because the tech's cool,
we're doing it because cash freecashflow is the goal.
Right, and all technology youknow should be oriented around

(01:09:09):
free cash flow.
Like, if you're deploying youknow technology in your business
, there needs to be a very clearbusiness case that this grows
revenue, reduces costs, boostsincome or protects maybe future
reduction of you know.
Sure, you know people can dopreventative things too.
Is what I'm saying like, makesure that this thing stays up

(01:09:31):
and doesn't break?
Yeah, but beyond that, like,business cases have to be really
, really, really clear and Ithink there's too many things
that happen because of with, youknow, in all organizations and
I'm very much to blame for thisand it happens in my
organizations as well butwithout clear, concise, coherent
business cases and financecases to support them right?

(01:09:53):
It's really simple.
Everything is a tiny littleprofit and loss.
If I got that leasing software,what is it costing me and what
is it making me?
Period Sure Right.

Speaker 1 (01:10:03):
Yeah, or what is it saving me?
Right?
And the nice thing about andI'll leave this because we're
coming up on the end of time andI'd love to obviously we'll
have you back for more stuff inthe future.
But the P&L management is whatthese operators get, right.
I mean, their job is thatthey're in the P&L all the time,
and so when you look attechnology, that's just another

(01:10:24):
business case where you'retalking about that free cash.
That's.
The purpose of this is toincrease the revenue, cut the
cost, which increases the netincome, and so there's some.
You know, if you're in financeand you're running P&Ls, you're

(01:10:44):
well positioned to really thinkthrough this tech thing.
But, like to your point, goingback to these wealthiest
companies on the planet, aredata companies, right?
So we have to ask where is ourdata?
Who owns it?
What are we doing with it?
How are we unlocking it?
You know all that stuff.

Speaker 2 (01:10:57):
Yeah, and those are the partnerships we're working
on right now is sort of sayinglike, hey, how do we get that
data out of your system and turnit into a cash source?
And I think it's like.
I think everyone at a very highlevel knows that their data is
probably worth something.
Um, but knowing that your datais worth something and being
able to monetize that data arevery different things, right,
Like I grew up in Saudi ArabiaI've lived in 14 countries, by

(01:11:20):
the way, a lot, a lot of theworld.
Um, I've lived in 14 countries,by the way, I've lived in a lot
of the world, a lot of emergingcountries, Africa and all these
places.
But one of the places I grew upwas Saudi Arabia for some time,
and Saudi Arabia I mean in thedesert.
They joke like literally, andI'm sure it does happen you can
like just stick a sand in the,you know, stick in the sand and
out, spritz oil.
Like that's great, Like oil,sure, like theoretically

(01:11:42):
valuable, but like if it's justspraying in your face.

Speaker 1 (01:11:45):
Yeah, right it isn't super valuable yeah uh.

Speaker 2 (01:11:48):
So you need to.
You know, take it, capture it,structure it, store it, build a
supply chain around you know,refine it, make sure it's usable
, it's the right grade, get itto you know.
Uh, golfer, sorry, pet, petroCanada or something, some retail
across the world who's like,built a station?
Yeah, so now the oil is worthsomething?

(01:12:10):
Yeah, so I think it's it's.
It's interesting and what wefound, you know, in our
partnerships, is that themajority of real estate
operators have some concept thattheir data is worth something.
But just because they thinkit's worth something doesn't
mean that they have thewherewithal or necessary, you
know, to be in all of thebusinesses that would make

(01:12:30):
something from it, right?

Speaker 1 (01:12:32):
And that's where I think when I mentioned my
analogy to the developmentprocess, they trust the
structural engineer.
They know like I need thisbuilding to go vertical, but I
also know like I'm not going togo out and be a structural
engineer.

Speaker 2 (01:12:43):
Yeah, it's hard.

Speaker 1 (01:12:45):
So that's where you guys come in to help that
refinement.
We've already talked about howit unlocks the value, turn that
data into cash, but really youneed somebody that understands
that space and keeps up withthat space, because if they
can't find people to bemaintenance technicians or
leasing agents or even VPs ofproperty management, they're

(01:13:08):
certainly not going to have thechallenge of bringing in all the
teams and disciplines andunderstanding to try and really
unlock that type of value.

Speaker 2 (01:13:19):
Look, you can ask my wife if she looks at me and
other people.
And I say to you, and we allsay to each other running a
business is hard, Running yourcore business is hard enough.
And if you have the wherewithalto run your core business and
five non-core businesses, pleaseteach me how, Because I don't
know how to do it.

(01:13:40):
It's hard enough to bite off.
You know one problem to solve.

Speaker 1 (01:13:43):
Right, and there's all types of research on how
that doesn't work for people,even in any kind of you can't.
If you don't focus on it,nothing gets done.
Well, listen, we're coming upon the end of our time.
What any final thoughts youwant to leave our listeners with
?

Speaker 2 (01:14:01):
Nothing specific, just thanks for having me Great,
invigorating conversation.
I'm glad I flew out here.

Speaker 1 (01:14:08):
Wonderful.
Well, it's great having you,and we'll be tracking all your
success.

Speaker 2 (01:14:11):
Yeah.

Speaker 1 (01:14:12):
Well, our attempts, that's it.
Attempt, that's a good point.
We'll leave that Attempt.
Something today that would befun to do.
All right, we'll see you in the.