Rules first. AI second. Every number reproducible.

Step 1

The data and taxonomy

FOS Watch reads the public FOS decision database. We don't host or republish the decisions themselves — for specific decision text, FOS Watch links the user to the FOS's own database. The structured insight we provide is built on top of that public record.

FOS Watch analyses decisions published each week and captures data on firm, product, and the following dimensions:

Product

mortgages, credit cards, motor insurance, and so on.

Reason

the complaint itself: affordability, mis-sale, claim handling, etc.

Outcome

upheld, not upheld, as classified by the FOS (partial uphold is also captured where appropriate).

Rules

the regulatory rules and codes the ombudsman cited (DISP, ICOBS, CONC, Consumer Duty outcomes, the CRM Code, and others).

In addition we capture:

Raised by

consumer direct, or via a professional representative.

Escalated by

consumer-driven or firm-driven escalation to ombudsman stage.

Firm defence

what the firm argued.

Evidence

what documents or proof the firm or consumer provided, or the ombudsman requested.

Directions

what the ombudsman directed the firm to do.

APP subtype

for authorised push payment cases: purchase scam, investment scam, romance scam, and so on. This can be extended on client request to capture additional complaint types in more granular detail.

The taxonomy is plain-English, designed by former ombudsman casework leaders building on feedback provided by analysts, risk and compliance leads and casehandlers from Tier 1 banks and insurers. It is versioned — every change is dated and recorded, so a dashboard read on a given date can be reproduced months later.

Step 3

The aggregations: arithmetic you can audit

The numbers on a FOS Watch dashboard — uphold rates, peer comparisons, theme volumes, month-on-month movement — are produced by structured queries over the tagged decisions, running on a relational database. A dashboard tile showing "consumer hire agreements: 27.8% uphold rate over the last 12 months" is a single query. Two consequences follow:

• The same query run twice produces the same number. No model-output drift between runs.
• Aggregates appear on the same view as a sample of the most recent underlying cases, so the user can check that the decisions feeding the analysis match their search intent. Each case shows its FOS decision reference, firm, decision date, and outcome, and links to the original on the FOS site.

FOS publishes decisions six to eight weeks after they are issued. The most recent months are therefore always incomplete, and partial reads can be volatile — particularly on smaller case populations where a 0% or 100% uphold rate would be misleading. The Insight Hub suppresses incomplete months on monthly trends. The fortnightly Risk, Compliance & MI Briefings follow the same rule. The Case Update Digest works on a sliding fortnightly window and so isn't subject to it; we note that on the digest itself.

Step 5

Audit and traceability

For a Tier 1 firm, the question "how do I know the numbers are real?" is a governance question, not a product question. FOS Watch is built so the answer is straightforward:

• Every number traces to a query.
• Every query runs over a versioned taxonomy.
• Every taxonomy version is dated.
• Every tag was produced by a rule that a human signed off.
• Each underlying decision retains a link to the original on the FOS site.

If your internal audit team or second-line function wants to validate a specific figure, we provide the calculation, the taxonomy version in force, and the underlying decisions. We expect to be asked. The methodology isn't proprietary in a way that obstructs review — the value is in doing the work, not in keeping it opaque.