Every renewal price checked against the new-business price before it goes out, every pricing change tested for fair value and proxy discrimination, and clear evidence for the senior manager who signs the attestation. People own every pricing decision.
At a glance
An illustrative scenario. The numbers are typical examples, not a client case study. The cloud services and UK rules were checked on 9 November 2026.
Since 1 January 2022, the FCA's pricing rules in ICOBS 6B have said a home or motor renewal price must be no higher than the equivalent new business price (ENBP): what a new customer with the same risk would pay through the same channel. A senior manager attests each year that the firm has complied. In practice the ENBP is hard to get right. It depends on the channel, the price comparison website, discounts and incentives, add-ons, payment method and mid-term changes, and many insurers run several rating engines and partner schemes. Checks are often sample-based, in spreadsheets, weeks after the renewal went out.
The ban on 'price walking' worked in part. The FCA's July 2025 evaluation found it reduced price walking and was linked to lower motor prices, with around £1.6bn of consumer savings estimated over ten years, but found no statistically significant effect on home prices. The risk has moved rather than gone. Pricing can now differ in other ways: by payment method, through add-ons, through new-business loadings, or through rating factors and enriched data that sort customers more finely each year. Under the Consumer Duty, firms have to show each product gives fair value, for every group of customers, not only on average.
Proxy discrimination is the hardest part to see. Postcode, occupation, vehicle and some external data can correlate with ethnicity and other protected characteristics. The FCA's research note of December 2025 found that, across the market, area risk explained the overwhelming difference in motor prices between areas with more and fewer residents from minority ethnic backgrounds. That is a market-level finding. Each insurer still has to be able to show the same for its own models, as they change, and machine learning models with hundreds of features make that harder, not easier.
Agree these, and measure the baseline, before anything is built. Example targets are typical, not promises.
Renewals issued above their equivalent new business price. Near misses are logged too.
ExampleZero, every near miss explained
The share of renewals checked against ENBP before issue, rather than by sample afterwards.
ExampleEvery renewal
How long between a pricing change causing a problem and the pricing team knowing about it.
ExampleFrom months to days
Price and margin by years with the firm, by channel. A rise with tenure is the early sign of price walking coming back in another form.
ExampleNo upward drift with tenure
Margin, claims paid against premium, cancellations and complaints for each group, including monthly payers and vulnerable customers.
ExampleNo group worse off without an accepted risk reason
New rating factors, data sources and models tested for proxy discrimination, with the result recorded and signed off.
ExampleEvery change, no exceptions
Cross-sector rules are on the UK regulation map.
Renewal prices for home and motor must not exceed the ENBP for the same channel, and a senior manager attests compliance each year by 31 March, alongside the firm's pricing data returns. The design checks every renewal against its ENBP by rule, keeps the evidence, and builds the attestation pack from the monitoring rather than from a year-end sample.
Firms must show the price is reasonable for the benefits, for each group of customers. Fair value is one of the FCA's 2026 insurance priorities. Monitoring reports value by segment, including long-tenure customers, monthly payers and customers with characteristics of vulnerability, and the results feed the fair value assessment.
The insurer as manufacturer reviews fair value across the distribution chain, including partner schemes and add-ons. Partner and add-on pricing is monitored with the same tests, and the reviews use the same data.
The FCA chose not to cap rates or require interest-free monthly payments, but expects every firm to review whether its premium finance gives fair value, and will keep monitoring. The design tracks the total cost of paying monthly by customer group, so a gap between monthly and annual payers is seen and explained.
Insurers may price on age, and on disability where it's reasonable, only with relevant information from a reliable source. Sex hasn't been allowed as a pricing factor since 21 December 2012, and there is no exception for race on new business. Indirect discrimination through proxies is still discrimination unless it can be justified. Every factor and data source is tested for proxy effects, the risk justification is written down, and legal advice is taken on anything borderline.
Quotes come from the rating engine, so they are automated decisions, and a sharply higher renewal can be a significant one. Since 5 February 2026, those need safeguards: telling people, letting them challenge, and a person who can step in. The stricter limit applies where special category data, such as declared medical conditions in motor, is used. Monitoring uses pseudonymised data and area-level statistics, never an individual's inferred ethnicity, and a DPIA covers it.
Home and motor pricing aren't in Annex III. Risk assessment and pricing for life and health insurance are, with high-risk duties from 2 December 2027 under the Digital Omnibus. A UK insurer writing only home and motor in the UK is outside it. A group that also writes life or health in the EU should build this monitoring to the same standard there.
Showing AWSAzuresovereign / on-prem services. Switch with the toggle; every diagram and table on this page follows it.
01Channels
02Intake
03Intelligence
04Orchestration
05Systems of record
06People
| Layer | AWS | Azure | Sovereign / on-prem |
|---|---|---|---|
| Quote and renewal eventsEvery quote, renewal and price change, with the rating factors used, as it happens. | Amazon Kinesis Data Streams | Azure Event Hubs | Apache Kafka |
| ENBP checkPlain, tested code owned by pricing, treated like any financial control. No model decides whether a price breaches ENBP. | AWS Lambda, or code beside the rating engine | Azure Functions, or code beside the rating engine | Containers on Kubernetes, or code beside the rating engine |
| Pricing analytics storeKeep several years of quotes and renewals, pseudonymised, so tenure patterns can be seen. | Amazon Redshift, with data in S3 | Microsoft Fabric | PostgreSQL, or your existing data warehouse |
| Area reference dataArea-level only. Scotland's census was in 2022, the rest of the UK's in 2021. Never infer an individual customer's ethnicity. | Census area statistics (ONS, NRS, NISRA) | Census area statistics (ONS, NRS, NISRA) | Census area statistics (ONS, NRS, NISRA) |
| Fairness, drift and proxy testsSageMaker Clarify and Model Monitor closed to new customers on 30 July 2026, so on AWS the same tests run as your own scheduled jobs. | Amazon SageMaker AI (scheduled jobs and pipelines), with open-source fairness libraries such as Fairlearn | Azure Machine Learning (model monitoring and the Responsible AI dashboard) | MLflow and Kubeflow on Kubernetes, with open-source fairness libraries such as Fairlearn |
| Explanations of price driversShows which rating factors moved a group's prices, so the pricing team can judge whether the reason is risk. | SageMaker AI, with open-source SHAP | Azure Machine Learning, with open-source SHAP | MLflow and Kubeflow on Kubernetes, with open-source SHAP |
| Writing up the findingsThe language model only writes up numbers the tests produced. Every figure in a draft links back to its query. | Amazon Bedrock | Microsoft Foundry | Open-weight models served with vLLM on your own GPUs, or IONOS AI Model Hub |
| Agent runtimeThe original Bedrock Agents closed to new customers on 30 July 2026; new builds use AgentCore. | Amazon Bedrock AgentCore | Foundry Agent Service | An open-source agent framework (e.g. LangGraph) on Kubernetes |
| Workflow | AWS Step Functions | Logic Apps or Durable Functions | Temporal, or n8n |
| Middleware and events | Amazon EventBridge, SQS | Event Grid, Service Bus | Apache Kafka or RabbitMQ |
| Dashboards and reporting | Amazon Quick (formerly QuickSight), or your existing BI tool | Power BI | Your existing BI tool, or Grafana |
| Human reviewAlerts are worked where pricing already tracks issues, with the decision and reason recorded. | In the pricing team's existing case or ticket tool | In the pricing team's existing case or ticket tool | In the pricing team's existing case or ticket tool |
| Guardrails | Bedrock Guardrails | Foundry guardrails (Azure AI Content Safety) | Open-source guardrails (e.g. NeMo Guardrails, Llama Guard) |
| Identity, keys, data protectionKeep the key that links pseudonymised records to customers away from the monitoring platform. | IAM Identity Center, KMS, Macie | Entra ID, Key Vault, Purview | Keycloak or your existing Active Directory; OpenBao or HashiCorp Vault; an HSM |
| Monitoring and auditLog every test run, threshold, alert and decision, with the model and data versions used. | CloudWatch, CloudTrail | Azure Monitor, Log Analytics | Prometheus, Grafana, OpenTelemetry |
| UK data residencyIONOS AI Model Hub runs in German data centres: EU, not UK, residency. | eu-west-2 (London) | UK South / UK West | Your own UK data centre, or a UK or EU provider |
On a phone, the table shows the platform you pick. Sovereign / on-prem means open, portable components you can run in your own data centre or with a UK or European provider; IONOS is shown as one example, and others exist. What sovereign cloud means →
Many organisations keep core systems and sensitive data on their own servers. This is how the same design works across both.
Stays on-prem
The rating engine, pricing models and policy admin system stay where they are, on your own servers or with their vendors. The ENBP check runs beside the rating engine, on your side, so every renewal is checked whatever happens to the cloud.
Uses cloud AI
The analytics store, the fairness and drift tests, the explanations and the monitoring analyst agent can run in AWS or Azure UK regions on pseudonymised data, or on your own platform for the sovereign option.
How they connect
Quote and renewal events stream over a private link (AWS Direct Connect or Azure ExpressRoute) from the rating engine. Only the fields the tests need are sent, and customer identities stay on your side.
If the link drops
Renewals carry on, and every one is still checked against its ENBP locally. Events queue on your side, and the monitoring catches up and back-fills when the link returns.
What each agent may do through its tools, what it may not, and where a person signs off. The limits are enforced by the platform, not the prompt.
Monitoring analyst agent
Where a person signs offThe pricing committee decides every pricing change, and the senior manager signs the annual attestation. Only the pricing team can change thresholds, under change control. The agent's permissions are enforced by the platform's identity and policy controls, not by instructions in a prompt.
The rating engine publishes every quote and renewal as an event, with the factors and the ENBP used. The monitoring reads these events and never calls back into the rating engine, so it can't slow down or change a quote.
The ENBP check sits in the renewal flow as a rule, before the price is issued. If the check can't run, the renewal waits for a person; it is never issued unchecked. Every renewal carries an ID, so a retry never creates a second notice.
Partner schemes, price comparison website feeds and premium finance data arrive through the API layer you already have. If you run an integration platform, the monitoring uses it like any other system.
The AI monitoring can be switched off without stopping the business. Renewals and the ENBP check carry on; only the alerts and reports pause.
Map how ENBP is calculated in every channel and scheme, measure today's checks, list the customer groups and data sources, and agree the fairness definitions.
Gate · must be true to passENBP logic documented, data access confirmed, DPIA started, compliance and legal involved.
Motor renewals, in parallel with today's checks, and the tests run back over the last 12 months to see what they would have found.
Gate · must be true to passENBP check matches or beats current checks, alert volumes workable, fairness definitions signed off by the pricing committee.
The ENBP check moves into the renewal flow and monitoring goes live for motor, with support and a rollback plan.
Gate · must be true to passMonitoring, support model, rollback plan and pricing analyst training in place.
Add home, partner schemes, add-ons and premium finance, then use the same tests for new-business pricing changes.
Gate · must be true to passEach new product passes the same pilot gate.
| Activity | Attesting senior manager | Head of pricing | Solutions architectI can fill this | Technical PMI can fill this | EngineersI can fill this | Compliance & legal | DPO | SecurityI can fill this |
|---|---|---|---|---|---|---|---|---|
| Business case and benefits | Accountable | Responsible | Consulted | Consulted | Informed | Consulted | Informed | Informed |
| Target architecture | Informed | Consulted | Accountable and Responsible | Consulted | Consulted | Consulted | Consulted | Consulted |
| Fairness definitions and thresholds | Accountable | Responsible | Consulted | Informed | Consulted | Consulted | Consulted | Informed |
| DPIA and use of protected characteristic data | Consulted | Consulted | Consulted | Informed | Informed | Consulted | Accountable and Responsible | Consulted |
| Build and integration | Informed | Consulted | Accountable | Responsible | Responsible | Informed | Consulted | Consulted |
| Back-testing on real renewals | Informed | Accountable | Responsible | Consulted | Responsible | Consulted | Informed | Informed |
| Go-live decision | Accountable | Responsible | Consulted | Responsible | Informed | Consulted | Consulted | Consulted |
| Run, monitor and improve | Consulted | Accountable | Consulted | Consulted | Responsible | Consulted | Consulted | Consulted |
R ResponsibleA AccountableC ConsultedI Informed· not involved
Option A
If there's no architect on the programme, I'd own the design end to end: the event flow from your rating engine, the ENBP control, the monitoring platform and the gates. I'd report to your programme board and work closely with your head of pricing and compliance.
Option B
If your partner or your own team already has an architect, I'd take on the event integration, the ENBP check and the monitoring pipeline under their lead, and give them a second senior opinion on data protection, security and how the evidence will stand up.
Goes wrong
ENBP is checked by sample after renewals go out, so a breach is found months later.
How the design avoids it
Every renewal checked by rule before it is issued, and a renewal that can't be checked waits for a person.
Goes wrong
Proxy tests run once when a model is built, then the data drifts.
How the design avoids it
Tests run on every release and every month, with drift alerts between.
Goes wrong
The team infers each customer's ethnicity to test fairness, and creates a new data protection problem.
How the design avoids it
Area-level statistics only, kept apart from customer identities, under a DPIA.
Goes wrong
Different fairness measures disagree, and nobody has said which one counts.
How the design avoids it
The pricing committee, compliance and legal agree the definitions in discovery and write them down.
Goes wrong
Alerts flood in, so people stop reading them.
How the design avoids it
Thresholds tuned in the pilot, materiality levels set, and every alert owned.
Changes already announced or under way that affect this design. More in the UK horizon scan.
Run the pilot in time to show the attesting senior manager what the monitoring would have found this year.
Report value by customer group now, including the total cost of paying monthly, so you can answer a supervisory request with data you already have.
Record the cloud and model providers in this design and decide which are material.
Keep the pricing decisions, tests, explanations and challenge route in a form you could show the ICO.
If it does, build that pricing into the same monitoring, with the documentation the Act asks for.
Good answers are specific. Vague answers to any of these are worth following up.
Whether you need someone to lead the design or to support the architect you already have, let's talk it through.
