The Feedback Loop

Demonstrating Good Outcomes

Engagement

All Sectors

FG22/5 Chapter 10 requires firms to identify where they are not delivering good outcomes and act on the root causes. The FCA's December 2024 Complaints and Root Cause Analysis review found firms generally collected complaints management information and ran some root cause analysis, but the loop consistently broke at two points — discussion at decision-making forums did not always translate into action, and where action was taken, firms rarely measured whether it had improved outcomes. The challenge is to design the closing move from signal to verified change, not merely the analysis.
The structural move is to design the firm's complaint, query, and claims signals as a continuous learning loop — aggregating signals across channels, root-cause-analysing them, translating insight into product or process change with a named owner, and re-measuring to verify that the change improved the outcome:
Aggregate the signal across channels and cohorts
Complaint volumes are the visible tip; the richer signal sits across queries, claims experience, contact-centre transcripts, FOS decisions, social media feedback, and exit reasons — and across customer cohorts, including those with characteristics of vulnerability that the December 2024 review found firms often failed to disaggregate. The first move is a cross-channel aggregation that preserves the texture of each signal while letting themes surface. Smaller firms with low complaint volumes apply the same discipline using FOS decisions, FCA letters, and industry news as the December 2024 review specifically endorsed. The design test: can the firm show its root cause work draws on more than the complaints log, and that themes for vulnerability cohorts are visible separately from the firm-wide picture?
Root cause to a level that names the change
Root cause analysis must reach the level at which a specific product, process, or communication change can be named — not stop at categorisation. The 'five whys' technique cited in the December 2024 review is one example; fishbone diagrams, fault-tree analysis, Pareto analysis, and journey-stage attribution are equally valid. What matters is that the analysis traces the chain through product design, distribution, communication, and servicing, and surfaces the change owner. Where the cause is distributed across functions, the closing action must be co-owned. The design test: for any theme the firm has analysed in the last twelve months, can it produce the named change, the function accountable, and the deadline by which the change will be in place?
Re-measure to verify the change worked
The loop is closed only when the change has been measured against the outcome it was meant to improve — not against an output proxy. The December 2024 review's most consistent finding was that firms rarely measured whether interventions had delivered the intended impact, and where they did, the measure was often training-completion or process-adherence rather than the customer outcome. Re-measurement requires defining, before the change is made, the metric that will move if the change works — complaint volume on the specific theme, query rate at the affected journey step, claims acceptance variance, vulnerability-cohort outcome — and the threshold at which the loop is closed or re-opened. The design test: for each closed loop in the last cycle, can the firm produce the pre-change baseline, the post-change measurement, and the governance decision that confirmed closure or re-opening?
- Themes surfaced through the firm's signal aggregation are routinely cut by vulnerability characteristic and other relevant cohort lens, and themes specific to those cohorts are visible separately from the firm-wide picture rather than absorbed into a single category.
- Every action arising from root cause analysis carries a named change owner, a deadline, and a defined re-measurement metric tied to the customer outcome the change was meant to improve — not training-completion, process-adherence, or another output proxy.
- The firm can produce, for any closed loop in the last twelve months, the pre-change baseline, the post-change measurement, and the governance decision recording closure or re-opening — with closure refused when the metric has not moved.
- Loops surfaced through complaints, queries, claims, and contact-centre data are aggregated to a single owner who tracks them end-to-end, so the closing step does not fall between customer operations, product, and MI — the structural gap the December 2024 review repeatedly identified.
- A retail bank's complaint MI showed a rising theme of customer dissatisfaction with switching between front-book and back-book savings rates after the FCA's September 2024 cash savings update. Initial analysis attributed the volume to confusing rate-change letters, and the firm rewrote the letter — a self-contained intervention with no measurement defined. Complaint volumes did not fall. Re-opening the loop, the firm applied a five-whys trace through the journey and found that the cause was not the letter but the absence of a prompt at the point at which the introductory rate ended; customers learned from a statement, not a notification, and felt the firm had not warned them. The fix moved to the journey itself — a notification ten days before reversion with a one-click switch path — and the firm defined re-measurement against complaint volume on the specific theme, switch-rate take-up, and vulnerability-cohort outcomes. Volumes fell by half over two quarters, and the loop was closed in the next governance cycle with the baseline, post-change measurement, and a documented decision recorded in the minutes. The earlier letter rewrite, which had not been measured, was logged as an open loop the firm had wrongly closed.
- An insurer running a redress communication exercise drew on the FCA's March 2026 motor finance redress communications research, which had shown that small changes in letter wording and structure materially shifted comprehension, intent to opt in, and sentiment. Rather than treating the firm-led letter as a fixed compliance artefact, the insurer treated it as a closed-loop intervention. Phase one shipped a Plain English version with comprehension and opt-in tracked at cohort level. Query patterns from the contact centre were aggregated weekly and root-caused: a recurring confusion about the deadline, and a smaller but persistent suspicion that the letter was a scam. Two changes followed — a clearer deadline panel and a verification reference to a named scheme webpage — each with a defined re-measurement against comprehension, opt-in rate, and call-centre query volume on the affected sub-themes. The vulnerability-cohort cut showed low-financial-capability customers needing additional treatment, and a routed callback offer was added for that cohort with its own re-measurement. The board's annual reckoning (Pattern 25) recorded the closed loops and the open ones; one loop was re-opened when the second-cycle measurement showed no improvement on scam-suspicion sentiment, and the change was redesigned.
Common failure modes
The most common failure mode is treating completion of the root cause analysis as the goal, with the action and the measurement deferred — the December 2024 review's headline finding. A second is the DISP-compliant illusion: the firm meets DISP 1.3.3R on management controls and root cause analysis as set out in PRIN 2A.9.9R, the loop looks closed on paper, but no product or process has actually changed and no outcome has been measured. A third is single-source aggregation. Treating the complaints log as the signal, without query patterns, claims experience, FOS decisions, contact-centre transcripts, or social-media feedback, produces a partial picture; smaller firms in particular were called out by the December 2024 review for relying on too narrow a base. A fourth is over-reliance on one root cause technique; the five-whys is one example, not the method. A fifth is measurement against the wrong metric: training-completion or process-adherence does not evidence that customers are experiencing something different, and the FCA's review specifically warned firms not to substitute output proxies for outcome change. A sixth is closing the loop on average outcomes while a vulnerability or other cohort continues to experience the same harm — the disaggregation gap the review named explicitly.

Related Patterns

The Outcome Dashboard

The Early Warning System

Reading the Room

The Foreseeable Harm Test

Evidence & Supporting Facts

FCA Complaints and Root Cause Analysis Review (December 2024, 40-firm sample) found that while most firms collected complaints MI and ran root cause analysis, the consistent gaps were taking action on insight, measuring impact, and disaggregating data for different customer cohorts including vulnerability.

FG22/5 Chapter 10 requires firms to set up controls and key processes to identify where they are not delivering good outcomes and to have a strategy to understand and tackle root causes — together with PRIN 2A.9.9R on appropriate action where failings are found.

The December 2024 review cited the 'five whys' technique as one good practice example for tracing causation, and named good practice in firms whose RCA insights reached front-line agents and the risk team — not only the complaints function.

FCA Motor Finance Redress Communications Research (March 2026) demonstrated that small changes in letter content, tone, and structure materially shifted consumer comprehension, opt-in intent, and sentiment — empirical evidence that comms-level closed-loop testing produces measurable outcome change.

AI Automation Lens

AI offers most leverage at three points in the loop. First, signal aggregation: NLP applied to call transcripts, chat logs, complaint free-text, FOS decisions, and social-media feedback can surface emerging themes weeks before they would appear in conventional MI categorisation, and can disaggregate by vulnerability indicators visible in the conversation rather than only the structured fields the December 2024 review found firms relied on. Second, root cause acceleration: large language models can run a draft five-whys or journey-stage attribution against a theme using the firm's process documentation and journey maps, surfacing candidate causes for human verification rather than starting from a blank page — useful particularly where causation crosses product, distribution, communication, and servicing. Third, re-measurement: pre-defined metrics can be tracked automatically, with the loop's closure or re-opening flagged by the system when the threshold is met or missed, and the governance pack populated with the baseline, post-change, and decision artefacts. The discipline the AI cannot supply is the closing decision itself — that the change is sufficient, that the measurement is the right one, that the loop is genuinely closed — which remains a judgement act for the named owner and the governance forum. The risk to manage is automation theatre: the dashboard says the loop is closed because the metric moved, while the customer's experience has not.