Financial risk assessments pilot – update on progress
Our Director of Major Policy Projects and Evaluation Helen Rhodes provides an update on the ongoing pilot of financial risk assessments following completion of the first phase.
Posted 10 February 2025 by Helen Rhodes
Last summer, we updated on how we planned to approach our pilot of financial risk assessments. Financial risk assessments are a proposed way of identifying high-spending remote gambling customers who may be in financial difficulties, in order to help support them. This is not just another way of saying “affordability checks” - we do not have any requirements for affordability checks and are not proposing any. Financial risk assessments would be a much more targeted way of identifying potentially financially vulnerable customers.
Over time, this way of identifying vulnerable people could be used to support these customers and prevent some of the very serious cases we have seen where customers have been able to gamble large amounts without any checks or support being in place. But we have always said that we will proceed cautiously, and this is why we are conducting a pilot of the assessments.
We are testing whether - and how - financial risk assessments could be introduced in a way that supports high-spending customers in financial difficulties but also supports a frictionless customer journey for the vast majority of customers. You can read more about what we are seeking to achieve and how we have approached setting potential thresholds for assessments in our consultation response.
Financial risk assessments would not affect a customer’s credit score if they were introduced in the future.
NatCen is working as our evaluation partner to support this work.
How to read this blog
We want to be open to stakeholders and provide updates on the preliminary findings of the pilot, but we emphasise that these findings are preliminary.
Findings from later stages may be different because they will look at more recent data for current customers or we may need to make adjustments as we continue to assess the data provided by the pilot participants to ensure that the data is accurate and reported consistently. We therefore urge caution on how these findings are used, and we urge stakeholders not to assume that conclusions are being reached following the first stage of a three-stage pilot.
The pilot’s overall approach
The pilot is not a ‘live test’ and no consumers have been affected. The pilot involves a group of the largest remote operators and commenced in September 2024. It is taking place in 3 stages and these stages will run to April 2025 with reporting and assessment of the full findings, alongside other data and evidence, to follow.
The pilot is testing four set success criteria:
Frictionless part 1: What proportion of those high-spending customers checked could get a frictionless financial risk assessment if they were introduced?
Frictionless part 2: How quickly could credit reference agencies return a financial risk assessment?
Data relevance and accuracy: Is using credit reference data meaningful for understanding of an individual customer’s current or imminent overall financial risk and financial vulnerability?
Implementation issues: How could the data be presented to operators to help understand the level of financial risk or vulnerabilities associated with individual customers? How could operators build financial risk assessments into their overall customer interaction processes?
At each stage, we are testing at least one of the success criteria. We expect each stage will give us different results. For example, it is likely that a smaller proportion of accounts would be able to receive a frictionless assessment when using historical data. Our final decisions will also be informed by other evidence and data.
Where are we now?
The first of the three stages is complete. Stage 1 looked at a cohort of inactive customers, looking back at historical data, and has primarily informed the first success criterion on the proportions of customers that might be able to receive a frictionless check. It also gave us some insights on data quality and understanding (success criterion 3) and implementation issues (success criterion 4). Stage 1 did not look at the speed of an assessment (success criterion 2).
Stage 1 tested customer accounts which had - during a set historical period - met high-spending thresholds. The account details were shared with one or more credit reference agencies and they provided a financial risk assessment at the point the threshold was met. This means the pilot is testing what financial risk indicators were present when the account met the high spending threshold. As a result, the credit reference agencies are replicating the data returns to operators as close to automated or live implementation as possible.
Stage 1 can be considered a pilot of the pilot - we wanted to test how pilot participants prepared data for the credit reference agencies and how the data was returned to the pilot participants. We wanted to test the pilot reporting tools to see if we were getting the right data back to the Commission. We wanted to refine and improve our systems for Stages 2 and 3 of the pilot, and identify issues that need further exploration.
We have now moved on to Stage 2 which is testing more recent data and allowing us to refine some of the aspects tested in Stage 1.
The emerging findings from both Stage 1 and Stage 2 are informing our approaches to the final stage of this pilot - where we will look at current data.
What did Stage 1 tell us?
Stage 1 involved more than 530,000 assessments across three credit reference agencies for approximately 300,000 accounts for the relevant year. This is not equivalent to how many accounts might be checked in a year if financial risk assessments were to be introduced, it was just the number of assessments that were undertaken for the number of accounts which met the test criteria for this Stage of the pilot across the largest operators. You can read more about how thresholds could be set in our consultation response.
Approximately 95 percent of these assessments met the first success criteria of a frictionless assessment. This means that the data shared by the operators was successfully matched by the credit reference agencies, which would allow a financial risk assessment to be returned to the operator in a frictionless manner.
Approximately 5 percent of the assessments were unmatched or the data provided by operators was invalid. This 5 percent was not able to receive a frictionless financial risk assessment. We emphasise that these findings are for Stage 1 only and comparisons should not yet be made to the estimates of the 2023 White Paper. Findings from later stages of the pilot may be different because they will look at more recent data or current customers, and in addition the separate data collection exercise will be relevant. However, for information the White Paper estimated 3 percent of customers would receive an assessment and roughly 80 percent of those would be able to receive a frictionless assessment. While direct comparisons cannot yet be made, the proportions for stage 1 of the pilot outperformed this assumption.
More detail on the figures
As referenced above, 95 percent of the approximately 530,000 assessments resulted in a frictionless assessment and 5 percent did not. Breaking these figures down a little further - just under 92 percent related to accounts where a full assessment was available and just over 3 percent were “thin files” or no adverse information. Thin files are where the customer can be matched, there is no positive credit history but the lack of negative indicators means they are considered lower risk in terms of current financial vulnerability.
Making up the 5 percent of assessments that could not be conducted in a frictionless manner, less than 1 percent were due to data formatting issues, invalid data or duplications in the data provided to credit reference agencies by operators. Later stages will explore how these issues could be addressed or minimised. Just over 4 percent of the assessments were unmatched, where the credit reference agency was unable to identify the customer, and no information was available. We will further explore which type of accounts are unmatched and whether using more recent data would affect this figure.
It is important to note that that this is not 5 percent of accounts. Our consultation proposal originally estimated that only 3 percent of active accounts would receive an assessment if introduced - these would be the highest spending accounts. Only a very small proportion of active accounts would receive an assessment and be unable do so in a frictionless manner - the White Paper estimate would mean approximately 0.6 percent of active accounts would receive an assessment but not be able to do so in a frictionless manner. It is not possible to do direct comparisons from Stage 1 to percentages of active accounts. However, the proportion of assessments that could not be conducted in a frictionless manner in Stage 1 was lower than the White Paper estimate. This is encouraging about the ability of assessments to be conducted in a frictionless manner, but we must wait for full pilot and data collection findings to get a true estimate.
What else did Stage 1 tell us?
In considering how many customers could receive a potential financial risk assessment, operator data quality can play a role in reducing friction. Operators can take steps to reduce duplicate accounts and rectify incorrect data fields to improve data linkage rates. Credit reference agencies have systems in place to support this data cleansing in a live environment, should this proceed to implementation.
In relation to data quality, we identified a number of issues to be explored further. Credit reference agencies have unique systems and ways of presenting the findings back to the pilot participants. It is expected that the credit reference agencies would have their unique systems. However, this caused some issues for pilot participants in assessing the findings of Stage 1. For example, a green RAG rating means different things across credit reference agencies, and the more detailed score and explanation of the risk indicators must be considered alongside the simple RAG rating. More can be done in the next stages or in any final implementation to support operator understanding of these different systems and indeed to allow credit reference agencies to make refinements to their models to reduce unnecessary variation or confusion, before implementation should this be introduced.
For some permitted data, the Commission will propose common definitions such as time periods to ensure commonality across credit reference agencies where that is genuinely needed. We will test whether this reduces the variation across credit reference agencies that some pilot participants found difficult to interpret.
In relation to implementation issues, pilot participants are as yet uncertain about the exact actions that might be proportionate when they consider both the financial risk assessment and the information they already hold and act on for customer interaction. We have created a working group of pilot participants to focus on these issues and their work could be used to inform guidance to operators to help them consider how to embed financial risk assessments into their wider customer interaction processes. Of course, financial risk assessments are not designed to be acted on in isolation, as that would fail to balance the financial risk alongside everything else that is known about the customer. As part of Stage 1, the participants shared anonymised case studies to help provide early insight into how the financial risk assessment could inform decision making.
What does all this mean?
Given that the Stage 1 findings are preliminary only, we cannot say what the pilot findings so far might mean for future decisions. However, the pilot exercise is proving to be worthwhile.
The interactive and collaborative approach is proving to be an incredibly useful exercise in testing how financial risk assessments might work in practice and explore practical implementation issues before final decisions are made. Taking a staged approach to the pilot has also been helpful. For example, for later stages of the pilot some changes to data reporting are being made to ensure accuracy/consistency from pilot participants and reduce the need for recalculations and resubmissions.
We will continue to share key updates as the pilot progresses.