Article 16. Algorithmic System Accuracy Reviews – Choosing the Right Approach

Spoken (by a human) version of this article.

• Outcome-focused accuracy reviews directly verify results, offering more robust assurance than process-focused methods.
• This approach can catch translation errors, unintended consequences, and edge cases that process reviews might miss.
• While more time-consuming and complex, outcome-focused reviews provide deeper insights into system reliability and accuracy.

This article explains why verifying outcomes is preferred over tracing through processes, and how it works.

About this podcast

A podcast for Financial Services leaders, where we discuss fairness and accuracy in the use of data, algorithms, and AI.

Hosted by Yusuf Moolla.
Produced by Risk Insights (riskinsights.com.au).

Episode Transcript

(00:00):
This article was publishedin November, 2024.
It's titled AlgorithmicSystem Accuracy Reviews,
Choosing the Right Approach.
Now the background to thisis that we've done a range
of accuracy reviews aroundvarious types of systems.
Things like home loans termdeposits broker commissions,
et cetera, where the approachthat we used is a bit different

(00:22):
to the traditional processfocused approach So, over the
last few years, that's beenour go to approach for these
as opposed to the traditionalprocess, and so I wanted to
talk a little bit about thathow it actually works better.
So here we go.
Even small errors in algorithmicsystems can lead to customer

(00:43):
harm, financial losses, failureto meet contractual obligations,
or regulatory issues.
Accuracy reviews help us findand resolve errors that may
exist, providing assuranceabout the overall level of
reliability of the system.
When reviewing the accuracyof algorithmic systems, the
method used can significantlyinfluence the outcome.

(01:05):
how robust the review is,and what it can reveal
about reliability andpotential improvements.
There are severalapproaches that we can use.
A common one focuseson processes.
A relatively newer methodfocuses on outcomes.
Opting for an approach thatfocuses on the outcomes
rather than the processescan lead to more insightful

(01:27):
and actionable results.
This outcome focused strategyoffers reassurance by
verifying whether your systemsmeet their intended goals.
such as upholding customerpromises or meeting
contractual obligations.
It contrasts with traditionalprimarily examine processes
which discrepancies.

(01:49):
Now a major drawback withthe outcome focused method
is that it typically takesmore time and effort.
It is usually more difficultand more expensive.
It often involves fullpopulation testing.
It won't be appropriatefor all situations.
This article explainswhy verifying outcomes is
preferred over tracing throughprocesses and how it works.

(02:13):
Why depart fromtraditional methods?
Focusing on outcomes meansevaluating whether your
algorithmic systems areproducing the intended results,
such as accurate fees or precisebroker commission payments.
This contrasts with traditionalmethods, which primarily
examine the design andimplementation of processes.

(02:33):
While a process focuscan be useful in certain
situations, such as identifyinginefficiencies or ensuring
regulatory compliance, itmay not be the best method
for accuracy reviews.
Here's why focusing on outcomescan be more beneficial.
We've got seven benefitsor explanations of

(02:55):
what differences are.
So the first is directverification of results.
By concentrating onoutcomes, you directly
verify whether the systemdelivers what it promises.
This helps identifydiscrepancies between
expected and actual results.
Allowing for timely corrections.
Second is overcomingtranslation errors.

(03:17):
Translation errors oftenoccur when business
requirements are convertedinto technical specifications,
leading to discrepancies.
By focusing on outcomes, you canbypass these translation layers
and verify that results alignwith expectations, catching
any misalignments early.

(03:37):
Three, addressing quickfixes and process changes.
Processes often evolve throughquick fixes and adjustments.
which can introduceinconsistencies.
An outcome focused approachcan identify errors that
are masked by outdated rulesor undocumented changes.
Four, detectingunintended consequences.

(04:00):
Process focused methods maymiss critical discrepancies
that arise from complexinteractions within the system.
For example, a seemingly minorchange in one part of the
process could have unforeseeneffects on the final output.
That may not be evident whenexamining the process alone.
Revealing edge cases.
Process focused methodsoften concentrate on typical

(04:22):
scenarios, potentially missingrare but critical edge cases.
Outcome focused analysiscan highlight these outlier
situations where the systemmay produce unexpected
or incorrect results.
Identifying data qualityissues, process reviews
may overlook problemswith input data quality.
An outcome focused approachcan reveal discrepancies caused

(04:45):
by incorrect or outdated data.
that may not be apparentwhen simply examining
the process flow.
And finally, slowlychanging dimensions.
The outcome focused approachis designed to deal with
changes in master data duringthe period under review.
This can be handled by aprocess focused review,
but only if it involvessignificant data testing.

(05:06):
So, there are several goodreasons for adopting an
outcome focused approach.
A step by step explanationabout how it works.
My help in understanding howthese benefits manifest, how the
outcome focused approach works.
This method involvesseven key steps, one
of which is iterative.
One, scope definition, to definethe review objective in scope.

(05:31):
Two, base understanding.
Examine relevant documentsthat outline expectations,
for example productdisclosure statements or
third party contracts.
Overall process flows and systemarchitecture or data flows.
3.
Data collection.
Gather relevant data, inputs asclose to possible to the source,

(05:52):
and target the final outputs.
4.
Reperformance.
Reperform the calculationsbased on agreed contractual
terms to produce themodeled or expected results.
Then compare those withthe system's outputs,
the actual results.
5.
refinement.

(06:12):
If there are differencesbetween the expected and
actual, check whether thisis a problem with the model
result or the actual result.
For example, a valid exceptionapplies, or the model result has
missed a key step or made an ora data fix was applied in the
actual process because of a bugor issue in the source system.

(06:33):
Six, root cause identification.
Once the specific inaccuracieshave been confirmed,
identify the sourcesof those inaccuracies.
And then seven, reporting.
Document the result, how thereview was performed, the issues
and how they will be resolved.
Other approaches, like theprocess focused approach,

(06:53):
also start with definingscope and end in a report,
but the other five stepswill look quite different.
Choose wisely.
While this approach, that isthe outcome focused approach,
can take significantlymore time and effort, it
is certainly possible.
Newer analytic software andtechniques enable datasets,

(07:14):
reducing manual effort.
It is not the norm.
Most reviews follow atraditional approach.
But don't let anyone tellyou that it is impossible.
It can be done, even if you havean exceptionally large dataset.
We do this routinely acrosshundreds of millions of records.
By choosing an outcomefocused approach, you can

(07:36):
improve the chances that thereview will pick up errors.
This method is more robust forreviewing algorithmic accuracy
than traditional audit methods,Addressing common challenges
and giving you better assurance.
That's the end of the article.
Thanks for listening.