Regulatory Reporting
A return is a number with a firm’s name on it. What gets tested is where the number came from.
Nobody buys a report because they want one. These are the submissions a firm has to produce, to a deadline, accurately, with somebody’s name against them — IFRS 9 expected credit loss, financial crime returns, APP scams performance data, and the periodic returns that go out whether or not anything happened. We build them as controls: reconciled, traceable to source, and reproducible after the fact.
Figure Lineage
ReproducibleSource extract
Reconciled to ledger
Exceptions counted
Figures signed
Reproducible later
Every figure on the return traces back to the system that produced it — dated, reconciled, and re-runnable after the people who filed it have moved on.
The shift
The return became public
Some of this data is no longer read only by a supervisor. The PSR publishes APP scams performance for directed firms and ranks them against one another, so a firm mis-stating its own reimbursement rate is now mis-stating it in front of its customers and its competitors.
The gap
Nobody can rebuild last quarter
The calculation is rarely the hard part. The hard part is that the figure came out of a spreadsheet that took an extract that somebody filtered, and the person who knew which filter has moved teams. The number is defensible only as long as they are still answering email.
The standard
A return you can re-run
The test we apply: regenerate the submission from source months later and land on the same figures, or be able to say precisely why not — new data, or a corrected method. Those are very different answers and a firm should never have to guess which one it is giving.
What we report on
Four kinds of submission, one problem underneath
Different regimes, different signers, different deadlines — and in every case the challenge lands on provenance rather than arithmetic.
IFRS 9 expected credit loss
Staging and the significant-increase-in-credit-risk test, forward-looking scenarios and their weights, post-model adjustments held as a named, dated list rather than a plug. The disclosure is the visible part; what gets challenged is the movement between periods and whether anybody can explain it.
Financial crime reporting
The annual financial crime return under SUP 16.23 — REP-CRIM — is an aggregate of numbers that live in the transaction monitoring and case systems. It is the one return where our fraud and AML work and our reporting work meet: the same alerts counted for detection are counted again for the return, and the two counts should agree.
APP scams performance data
The PSR collects and publishes reimbursement and scam rates for directed firms under PS23/1 and Specific Direction 18, and ranks them against each other. A published table makes reporting accuracy a commercial exposure and not only a compliance one.
Prudential and periodic returns
The scheduled submissions a firm files whether or not anything happened that period. The work is rarely the calculation; it is the reconciliation to the ledger, the treatment of late-arriving data, and knowing which of last quarter’s figures moved and why.
Unknown is not zero
An empty field and a zero are the same character on a return, and they mean opposite things.
A blank read as a number is nought. So a field nobody populated — a control nobody assessed, a category no system emits, a month the extract missed — arrives at the regulator as a firm stating positively that the count was nil. Not an omission a supervisor might query, but an assertion the firm has signed.
That single conversion is behind more restatements than any modelling error, and it is invisible in the finished return: the totals foot, the file validates, and the submission is accepted. So our returns keep “not recorded” distinguishable from “recorded as none” the whole way through, and the count of the first kind is on the face of the pack rather than in a footnote.
What makes a return defensible
Six properties, handed over with the return itself
A submission a firm cannot reproduce is a figure it can only defend by remembering. Each of these is a property of the return, not of the team that produced it — which is the point, because the team changes.
01
Lineage to source
Every figure traceable back to the system that produced it, in one step a reader can follow — not through a spreadsheet nobody owns.
02
Reconciliation, evidenced
The return agrees to the ledger and the case system, and the agreement is recorded as a result rather than asserted as a habit.
03
Exceptions counted, not dropped
Rows that could not be classified are reported as a count, not quietly excluded until the total looks reasonable.
04
A named signer
Who signed, on what date, against which version of the figures — because approval that does not bind to a version goes stale silently.
05
Restatement history
What changed since the last submission, and whether the change was new data or a corrected method. Two very different answers.
06
Reproducible on demand
The return can be regenerated from source months later and land on the same numbers, or the difference can be explained.
How engagements are shaped
Start with one return, traced end to end
The useful first piece of work is narrow on purpose: a single submission you already file, walked back to source. It is the cheapest way to find out whether the rest of the estate has the same problem.
Start here
One return, traced
Fixed fee, one return
- Pick the return you would least enjoy reproducing
- Every figure walked back to its source
- Blanks separated from zeroes, and counted
- Written verdict on what could not be traced
Then
Rebuild
Scoped on the return, not the headcount
- The return produced as a control, end to end
- Reconciliations recorded as results
- Restatement history from the first run
- Handed over so the firm files it without us
Ongoing
Cycle support
On the reporting calendar
- Support through each submission window
- Movement between periods explained, not smoothed
- Changes to the return tracked against the rule that drove them
- Independent review stays with the firm or a third party
What we cannot be
We cannot produce a firm’s return and also assure it.
The same line that governs the model work governs this: independence is the control, and a check performed by the party that produced the figure does not satisfy it. Building a return and reviewing it are separate engagements, and they do not both come from us.
Where the figures rest on a model, the doors meet and the boundary is worth stating plainly. Models & controls owns the estimate and the evidence that it works; AI & automation owns the pipeline that runs it every period. This page owns what leaves the building.
Starting
Name the one return you would not enjoy being asked to reproduce
That is the conversation worth having, and it takes about twenty minutes. No production data and no system access is needed to have it. Every enquiry is treated in confidence.