The TrueVision transition: what the move to trended data changes across your scoring stack

Many UK lenders already have the TrueVision data feed available, and will have benchmarked it against the scores on their own portfolios. A benchmark tells you how much better the new score separates your loan portfolio into good and bad customers. It does not tell you which cut-off, which provisioning baseline, or which rating system has to move.

TrueVision is best understood as an umbrella term rather than just a single score or new bureau data. It covers a family of scores and the data block that delivers new trended variables into a lender’s decisioning systems. That block replaces the Bureau Summary Block (BSB) and carries trended variables under new variable names. A trended variable tracks a value over time rather than as a single snapshot. Rebuilding the scoring stack on this new data touches everything downstream.

From Gauge’s snapshot to TrueVision’s film strip

For originations, most UK lenders currently use TransUnion’s Gauge 2 Account Origination (AO) scorecard. TrueVision AO replaces it as the third-generation origination scorecard. It sits inside a wider suite that also covers account management, a UK mortgage-specific score, collections, financial stress indicators, and thousands of trended attributes. Together the suite draws on more than 2,000 variables and between 30 and 72 months of raw bureau history.

A TrueVision score is not a rescaled Gauge score; it is a different measurement. Gauge is point-in-time; TrueVision is trended. A point-in-time score captures the credit record on the day; a trended score reads the same record as a sequence: whether balances are building or clearing, utilisation rising or falling, the file overall improving or deteriorating.

The value of reading a credit record as a sequence rather than on the day has been measured outside the bureau setting. One study combined a commercial bank’s own customer transaction records with credit bureau data covering January 2005 to April 2009, and estimated savings of 6 to 25% of total losses on conservative assumptions about the cost of cutting credit lines.¹ Adding behavioural history to a bureau file is the same move TrueVision makes, delivered as a standard feed rather than as a modelling project.

TransUnion is not moving alone. Experian launched Trended Data in the UK in November 2018 as a dedicated Trended Data Block inside its Delphi suite (now in Generation 11), with trended attributes bolted onto the score across rolling 6-, 12-, 24- and 36-month windows. Equifax has an equivalent. TrueVision itself has been available since 2019. Now TransUnion is moving lenders off the legacy Gauge and BSB pathway rather than leaving both live.

How a trended attribute is built

Trended attributes let a score separate two borrowers who score the same today: one whose credit history has steadily improved, one whose credit history has been deteriorating. TrueVision is described as drawing on 30 to 72 months of raw bureau data transformed into "trended attributes and algorithms", without naming either the data attributes or the transformations. From building trended attributes in-house, credit risk teams know the calculations sit in one of three families:

1. Windowed statistics: the maximum, minimum, average, or standard deviation of a variable over a defined lookback (last 3, 12, or 36 months). Ratios across time: value now against value six or twelve months ago. Doubled utilisation in six months looks different from steady utilisation, even where today’s number is identical.
2. Slopes and event counts: the slope of a trend line through monthly observations, or months since the last significant event (missed payment, new limit).

One score suite, many models

The application scorecard is usually the first place a lender pulls bureau data. Once the score is in the stack, its predictive power pulls it into other models and decision points:

• Application scorecards: convert a new customer’s bureau and application data into a single score, deciding whether to accept, decline, or price.
• Behavioural scorecards: the equivalent for existing customers, driving retention, credit-line changes, and pricing reviews.
• Score-triggered decisions: policy rules, cut-offs, and pricing bands, keyed off specific score values.
• Provisioning and rating-system inputs: a feature in IFRS 9 expected credit loss (ECL) models or, for internal ratings-based (IRB) firms, in the rating systems that drive capital.

Application scores and cut-offs

A lender’s application decision can be built using bureau data in more than one way: the external bureau score used directly with lender-set cut-offs; the bureau score used as one input inside an application scorecard alongside internal data; or the scorecard built directly on the raw data block incorporating internal data as well.

Whichever path a lender takes, the biggest gain from TrueVision at application is what the scorecard can see. Historically a bureau reported on the applicant’s file as it was at the point of request; reconstructing history meant commissioning retros (backdated bureau pulls for the applicant population) and hand-engineering richer features on top, which was expensive and rarely done at scale. TrueVision delivers those trended variables as a standard feed, so the model sees the applicant’s trajectory rather than a single frame. A large share of TrueVision’s value lands here.

Replace the predictor data and the score changes; a score from one system cannot feed a downstream process built for the other. Three questions follow:

1. Where do the cut-offs sit under the new score? Score historical application data under both scorecards and find the new cut-off that matches the old acceptance rate.
2. Is the "good" population still the one the scorecard was written for? A cut-off hitting a target bad rate on the old score may not hit it on the new one, either because the new score is more discriminating or because it ranks a certain population (e.g. near prime) differently.
3. Do the risk-appetite strategies still hold? Every override, manual referral, and policy carve-out keyed off a legacy band needs review; some translate cleanly, others encode old-distribution assumptions that no longer hold.

How the cut-off is reset depends on the objective: hold volume constant, hold the bad rate constant, or start from a target risk outcome. Each calls for a different sequence of percentile-matching, shadow-testing, and calibration. Our next article walks through this.

IFRS 9: same loan, different scale

IFRS 9 requires firms to compare current credit risk against origination-date risk on each loan. This is the significant increase in credit risk (SICR) trigger, which moves loans from Stage 1 to Stage 2 and raises provisions. The standard sets no quantitative test, so supervisors have supplied benchmarks: the European Banking Authority (EBA) monitors IFRS 9 implementation. Its exercise compares the probability of default at origination with the probability at the reporting date and treats a threefold increase, meaning an increase of 200% of the initial PD, as the level that should have moved an exposure out of Stage 1, then measures the share of exposures institutions still hold there despite it.²

The IFRS 9 comparison assumes both origination-date risk and reporting-date risk are measured on the same scale. A book that migrates to TrueVision during the life of its loans breaks this assumption. Without a bridge, distribution shifts can push accounts across the staging threshold and drive a provision spike unrelated to borrower deterioration, or mask real deterioration behind a favourable scale shift. IFRS 9 does not prescribe an approach during a bureau data change, leaving it to the lender to design and defend.

In practice there are three options:

1. Restate origination baselines: rescore the back-book under the new data (where historical inputs exist) so origination and reporting sit on the same scale.
2. Keep existing vintages on the legacy baselines until the loans run off, and apply the new data to new lending only. Only viable if the legacy data remains available for those vintages; if it is being discontinued, this option collapses.
3. Run both measures in parallel through a transition period and treat the difference as an itemised post-model adjustment (PMA), until the recalibration is fully documented.

A note for IRB banks

Banks on the IRB approach carry one procedural point and one caveat. Trended variables are transparent raw-data calculations, so they raise no models-within-models issue. A bureau score is itself a model, and IRB firms typically would not consume the external score directly, in line with the push by the Prudential Regulation Authority (PRA) to unwind embedded models.

Where the rating-system change is material, the bank will need supervisory approval; where it is not, a notification.

The current framework makes the test unusually concrete. The EBA’s March 2026 final draft regulatory technical standards (RTS) amending Delegated Regulation (EU) No 529/2014 puts the inclusion or removal of risk drivers in the notification tier, except where a change moves "the rank ordering or the distribution of obligors or exposures across grades or pools, in a fundamental manner, the measure and level of which will have been defined by the institution", with two quantitative backstops: a reduction of 1.5% or more in total risk-weighted exposure amount, assessed separately for each affected rating system, or 15% or more within the rating system being changed.³ A bureau score swap is the case that exception describes, so the classification turns on an empirical question about your own book rather than on a reading of the rules.

Note the jurisdiction: that is the EU instrument, and a UK bank works to the onshored version and to the PRA’s own approach. Ex-ante notifications go in at least two months before implementation, during which the supervisor can reclassify the change as material, so picking the classification up early keeps the conversation on the bank’s timetable.

Where this leaves the transition

A snapshot shows one frame; a trended score shows the film that got the borrower there. That extra history picks up risk a snapshot cannot: it sharpens acceptance decisions and steadies performance through the cycle.

How the migration actually lands on a lender’s stack is where the next article picks up. A lender’s decisioning consumes bureau information in one of the three ways set out above. The next article takes the first: the bureau score fed straight into the calibration layer that produces the model’s output.

Frequently asked questions