Cross-functional · Playbook 61

Supplier risk monitoring

Critical suppliers watched continuously across official registers and licensed sources, every alert summarised with its sources cited, and a named person deciding what happens next.

  • Retrieval (RAG)
  • Generative AI
  • Human in the loop
  • Modern Slavery Act 2015
  • Procurement Act 2023
  • ECCTA 2023
  • UK sanctions (OFSI, OTSI)
  • UK GDPR

At a glance

The problem
Suppliers are checked once at onboarding, then rarely again, so changes in ownership, sanctions, finances or conduct are found late
Typical volume
Around 2,500 active suppliers, 300 critical or high-risk, and a few hundred register and news changes a week
Who's involved
Procurement, category managers, legal and compliance, finance, the modern slavery lead, information security
Pilot
8 weeks on 100 critical suppliers, after a 3-week discovery
Platforms
AWS, Azure, or a sovereign or on-prem stack

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.

The business problem

Most organisations check a supplier carefully once, when it's onboarded: company details, financial health, insurance, certificates, a questionnaire. After that, the file sits still while the supplier changes. Directors come and go, ownership moves, a subsidiary is added to a sanctions list, accounts are filed late, a factory is linked to labour abuses in the press, a certificate expires.

The information is out there, but it's spread across Companies House, the UK Sanctions List, the government's debarment list, the modern slavery statement registry, your risk data provider's feeds, and the supplier's own documents. A category manager with 200 suppliers can't read all of it. So reviews happen on an annual cycle, if at all, and the first sign of trouble is often a failed delivery or a question from a customer.

The stakes have risen. Large organisations can now be prosecuted for failing to prevent fraud by people acting for them, with reasonable procedures, including due diligence, as the only defence. Public sector buyers assess suppliers against the Procurement Act's exclusion grounds. The government plans to make modern slavery statements mandatory in content and backed by penalties. Any monitoring has to be accurate, too: an alert that gets a supplier's name wrong, or repeats an allegation as fact, creates its own legal risk.

What AI changes

  • Critical and high-risk suppliers are watched continuously against official registers and the licensed sources you choose, matched by company number rather than by name.
  • When something changes, the AI reads the source and writes a short summary of what it says, with a link and a quote for every point. It reports what a source says; it never states an allegation as fact or adds its own.
  • Alerts are sorted by type and severity against your risk policy, and routed to the supplier's named owner, with sanctions matches going straight to compliance.
  • Periodic reviews start from a drafted pack: what changed since the last review, what's missing or expired, and the questions to ask the supplier.
  • Modern slavery evidence for high-risk categories is gathered in one place, ready for the annual statement, which people write and the board approves.
  • The AI flags; it never suspends, debars, blocks or terminates a supplier. A named person decides, and the decision and its reasons are recorded.

How success is measured

Agree these, and measure the baseline, before anything is built. Example targets are typical, not promises.

Monitoring coverage

The share of critical and high-risk suppliers monitored continuously, matched to a verified company number.

ExampleFrom annual checks to 100% of critical suppliers

Time to owner review

Hours from a change appearing in a source to the supplier's owner reviewing the alert.

ExampleUnder 2 working days; sanctions the same day

Alert precision

The share of alerts the owner judges relevant. Low precision means people stop reading them.

ExampleAbove 60%, and rising

Citation accuracy

In a weekly sample, the share of summary points fully supported by the cited source.

Example100% of sampled points; every miss investigated

Review effort

Time to complete a periodic review of a critical supplier.

ExampleDown by half

Overdue evidence

Critical suppliers with expired certificates, insurance or missing modern slavery evidence.

ExampleDown by three quarters

UK regulation and controls

Cross-sector rules are on the UK regulation map.

  • Home OfficeModern Slavery Act 2015, section 54 (transparency in supply chains)

    Commercial organisations with a turnover of £36 million or more publish a slavery and human trafficking statement every financial year, approved by the board and signed by a director, and can submit it to the government's modern slavery statement registry. The AI gathers and summarises evidence for high-risk suppliers and categories; people write the statement and the board approves it. The government plans to make the content mandatory and add penalties (see On the horizon).

  • Cabinet OfficeProcurement Act 2023 (in force 24 February 2025)

    If you sell to the public sector, authorities assess you, and the associated persons you rely on, against the mandatory and discretionary exclusion grounds in Schedules 6 and 7, check the debarment list, and can ask about your subcontractors. Your core supplier information sits on the Central Digital Platform. If you're a contracting authority, monitoring feeds your own checks. Either way, an AI alert is evidence for a person to weigh, never an exclusion decision.

  • Home Office and SFOFailure to prevent fraud (Economic Crime and Corporate Transparency Act 2023, in force 1 September 2025)

    Large organisations can be liable for fraud committed by associated persons, including suppliers performing services for them, intended to benefit the organisation. The defence is reasonable procedures, and due diligence is one of the six principles in the Home Office guidance. Continuous monitoring, with recorded decisions, is part of that evidence.

  • OFSI and OTSIUK financial and trade sanctions

    Paying a designated person can breach financial sanctions, and OFSI can impose civil penalties on a strict liability basis. Since 28 January 2026 the UK Sanctions List is the only source for UK designations, so screening uses it, not the old OFSI consolidated list. OTSI enforces trade sanctions, which matter for what you import. Possible matches go to compliance the same day; only compliance clears or confirms a match.

  • ICOUK GDPR and the Data Protection Act 2018

    Directors, owners and supplier contacts are people. Adverse media about them can be criminal offence data, which needs a condition under Schedule 1 of the DPA 2018. That means a DPIA, screening individuals only where your policy needs it, processing in UK regions, a retention period, a privacy notice that covers screening, and summaries you'd be comfortable disclosing in a subject access request.

  • CourtsDefamation and accuracy of what the AI writes

    A summary that misattributes an allegation, or states one as fact, can harm a business or a person who did nothing wrong. Summaries only restate what a cited source says, use 'reported' and 'alleged' where the source does, never generate allegations, and are checked against the source by a grounding check before anyone sees them.

The solution

Showing AWSAzuresovereign / on-prem services. Switch with the toggle; every diagram and table on this page follows it.

Supplier risk monitoring: reference architecture
  1. 01Sources

    • Companies House, UK Sanctions List, debarment list, modern slavery registry
    • Your risk data provider (licensed news, financial and ESG data)
    • Supplier documents: statements, certificates, questionnaires
  2. 02Intake

    • Lambda + EventBridge SchedulerAzure Functions (timer trigger)Kubernetes CronJobs, or n8nScheduled source connectors
    • Amazon S3Blob StorageCeph, or IONOS S3 Object StorageDocument and evidence store
    • Aurora PostgreSQLAzure Database for PostgreSQLPostgreSQLSupplier entity register
  3. 03Intelligence

    • Bedrock Knowledge BasesFoundry IQ (Azure AI Search)OpenSearch, or PostgreSQL with pgvectorSource index (retrieval)
    • Amazon BedrockMicrosoft FoundryOpen-weight models served with vLLMCited summaries
    • Agent
      Bedrock AgentCoreFoundry Agent ServiceLangGraph on Kubernetes, with open-weight models via vLLMSupplier risk research agent
  4. 04Orchestration

    • Step FunctionsLogic Apps / Durable FunctionsTemporalAlert and review workflow
    • EventBridge + SQSEvent Grid + Service BusApache Kafka or RabbitMQEvents and queues
  5. 05Systems of record

    • Supplier master in your ERP
    • Your procurement and contract system
    • Your risk register or GRC tool
  6. 06People

    • Person
      Supplier owner: reviews the alert
    • Person
      Compliance and legal: sanctions, fraud, modern slavery
    • Person
      Supplier risk committee: decides serious cases
  • Cloud or platform service (AWSAzuresovereign / on-prem)
  • AI agent
  • A person decides
  • Your existing systems
  • Outside your organisation

Components, side by side

Supplier risk monitoring: components on AWS, Azure and a sovereign or on-prem stack
LayerAWSAzureSovereign / on-prem
Source connectorsOfficial registers through their published APIs and downloads; news and financial data only from sources you license. No open web scraping.AWS Lambda with EventBridge SchedulerAzure Functions, Logic AppsKubernetes CronJobs, or n8n
Supplier entity registerEach supplier tied to its company number, parent and subsidiaries. Matching on numbers, not names, prevents most false alerts.Amazon Aurora PostgreSQLAzure Database for PostgreSQLPostgreSQL
Reading supplier documentsModern slavery statements, certificates and questionnaire answers, with expiry dates extracted.Amazon TextractAzure Content Understanding (includes Document Intelligence)Open-source parsing (e.g. Docling) plus an open-weight vision model
Source index (retrieval)Every chunk keeps its source, date and link, so every summary point can cite where it came from.Bedrock Knowledge BasesFoundry IQ, built on Azure AI SearchOpenSearch, or PostgreSQL with pgvector
Summaries and draftingWrites from retrieved sources only. Both clouds offer a choice of models, including Claude.Amazon BedrockMicrosoft FoundryOpen-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 AgentCoreFoundry Agent ServiceAn open-source agent framework (e.g. LangGraph) on Kubernetes
WorkflowRoutes alerts by type and severity, and escalates any not reviewed within the agreed time.AWS Step FunctionsLogic Apps or Durable FunctionsTemporal, or n8n
Middleware and eventsAmazon EventBridge, SQSEvent Grid, Service BusApache Kafka or RabbitMQ
Link to supplier master and procurementRead-only for the AI. Status changes are made by people in the supplier master.API Gateway, plus your existing integration platformAPI Management, plus your existing integration platformKong or your existing API gateway
Human reviewOwners review alerts where they already manage suppliers, and record the decision and reason there.In your procurement or GRC toolIn your procurement or GRC toolIn your procurement or GRC tool
Guardrails and groundingBlocks summary points the cited source doesn't support, before anyone reads them.Bedrock Guardrails, including contextual grounding checksFoundry guardrails (Azure AI Content Safety), including groundedness detectionOpen-source guardrails (e.g. NeMo Guardrails, Llama Guard) plus a source-match check
Identity, keys, data protectionAlerts about individuals visible only to the roles that need them.IAM Identity Center, KMS, MacieEntra ID, Key Vault, PurviewKeycloak or your existing Active Directory; OpenBao or HashiCorp Vault; an HSM
Monitoring and auditLog every alert, summary, source, review and decision.CloudWatch, CloudTrailAzure Monitor, Log AnalyticsPrometheus, Grafana, OpenTelemetry
UK data residencyIONOS AI Model Hub runs in German data centres: EU, not UK, residency.eu-west-2 (London)UK South / UK WestYour 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 →

Hybrid and on-prem

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 supplier master, contracts, the risk register and any investigation notes stay where they are. Sensitive cases, such as a whistleblowing report about a supplier, never enter the AI index.

Uses cloud AI

Source connectors, the index, cited summaries and the research agent can run in a UK cloud region, or on your own GPUs with open-weight models if supplier data must stay inside your estate.

How they connect

A private link (AWS Direct Connect or Azure ExpressRoute) and the API layer. Supplier identifiers go out; alerts and summaries come back as events onto your queue.

If the link drops

Purchasing and payments are unaffected. Sanctions screening at onboarding and before payment runs keeps working in your own systems, and monitoring catches up when the link returns.

The agents

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.

Supplier risk research agent

May do

  • Look up a supplier's Companies House record, filings, officers and owners
  • Screen the supplier and its owners against the UK Sanctions List and the debarment list
  • Search the licensed sources and the supplier's own documents in the index
  • Draft a summary of what changed, with a citation for every point
  • Draft a periodic review pack and questions for the supplier
  • Create an alert in the procurement or GRC tool and route it to the supplier's owner

May not do

  • Suspend, block, debar or terminate a supplier, or change its status
  • Clear or confirm a sanctions match
  • Contact the supplier or anyone else
  • Search the open web or use sources you haven't approved
  • State an allegation as fact, or include anything a source doesn't say

Where a person signs offThe supplier's named owner reviews every alert and records a decision. Sanctions matches go to compliance, who alone clear or confirm them. Suspending, exiting or reporting a supplier is decided by the supplier risk committee, with legal advice where needed. The agent's permissions are enforced by the platform's identity and policy controls and by read-only access to the supplier master, not by instructions in a prompt.

Middleware and integration

The supplier master and procurement system are reached only through an API or your existing integration layer, and the AI only reads. Every alert carries the source, the date it was retrieved and an ID, so the same change never raises two alerts.

Each source change becomes an event. The register lookups, matching and summaries follow from it. If a source is down or late, its connector retries and the gap is shown on the monitoring dashboard, so nobody assumes a quiet week means no change.

Monitoring never gates buying. If the AI is paused, orders and payments carry on, onboarding checks run as before, and the owners fall back to their review calendar.

Process and people

  • A supplier risk policy that says which suppliers are monitored, against which sources, and which alert types go to whom.
  • A named owner for every critical supplier, with a target time to review each alert type.
  • A clear escalation route: owner, then compliance and legal, then a supplier risk committee for suspension, exit or reporting decisions.
  • A weekly sample check of summaries against their sources, and a fast way for owners to mark a summary wrong.
  • A named owner for each source, prompt and risk rule, with change control, and a yearly review of which sources you license.
  • Owners trained before go-live on what an alert is and isn't: a lead to check, not a finding.

Delivery plan

  1. 3 weeks

    Discovery

    Agree the risk policy and alert types, list critical suppliers, confirm their company numbers, and choose and license sources.

  2. Gate · must be true to passRisk policy approved, sources licensed, DPIA started, legal and compliance involved.

  3. 6–8 weeks

    Pilot

    Monitor 100 critical suppliers in shadow mode: alerts go to a review team, not to owners, and every summary is checked against its source.

  4. Gate · must be true to passCitation accuracy at target, alert precision acceptable, no individual's data used beyond the policy, and legal sign-off on summary wording.

  5. 4–6 weeks

    Production

    Alerts go live to owners for the pilot suppliers, with workflow, escalation and monitoring.

  6. Gate · must be true to passOwners trained, escalation route tested, monitoring and rollback plan in place.

  7. Ongoing

    Scale

    Add high-risk suppliers, then subcontractors on public sector contracts, then modern slavery evidence packs by category.

  8. Gate · must be true to passEach new supplier group or source passes the same shadow-mode gate.

Roles and RACI

Who is responsible, accountable, consulted and informed, by activity
ActivityChief procurement officerProduct ownerSolutions architectI can fill thisTechnical PMI can fill thisEngineersI can fill thisLegal & complianceDPOSecurityI can fill this
Business case and risk policyAccountableResponsibleConsultedConsultedInformedConsultedConsultedInformed
Source selection and licensingAccountableResponsibleConsultedConsultedInformedConsultedConsultedConsulted
Target architecture and integrationInformedConsultedAccountable and ResponsibleConsultedConsultedConsultedConsultedConsulted
DPIA and summary wording rulesInformedConsultedConsultedInformedInformedAccountableResponsibleConsulted
Build and integrationInformedConsultedAccountableResponsibleResponsibleInformedInformedConsulted
Accuracy testing on real alertsInformedAccountableResponsibleConsultedResponsibleConsultedInformedInformed
Go-live decisionAccountableResponsibleConsultedResponsibleInformedConsultedConsultedConsulted

Business case and risk policy

Chief procurement officer
A · Accountable
Product owner
R · Responsible
Solutions architect
C · Consulted
Technical PM
C · Consulted
Engineers
I · Informed
Legal & compliance
C · Consulted
DPO
C · Consulted
Security
I · Informed

Source selection and licensing

Chief procurement officer
A · Accountable
Product owner
R · Responsible
Solutions architect
C · Consulted
Technical PM
C · Consulted
Engineers
I · Informed
Legal & compliance
C · Consulted
DPO
C · Consulted
Security
C · Consulted

Target architecture and integration

Chief procurement officer
I · Informed
Product owner
C · Consulted
Solutions architect
A/R · Accountable and Responsible
Technical PM
C · Consulted
Engineers
C · Consulted
Legal & compliance
C · Consulted
DPO
C · Consulted
Security
C · Consulted

DPIA and summary wording rules

Chief procurement officer
I · Informed
Product owner
C · Consulted
Solutions architect
C · Consulted
Technical PM
I · Informed
Engineers
I · Informed
Legal & compliance
A · Accountable
DPO
R · Responsible
Security
C · Consulted

Build and integration

Chief procurement officer
I · Informed
Product owner
C · Consulted
Solutions architect
A · Accountable
Technical PM
R · Responsible
Engineers
R · Responsible
Legal & compliance
I · Informed
DPO
I · Informed
Security
C · Consulted

Accuracy testing on real alerts

Chief procurement officer
I · Informed
Product owner
A · Accountable
Solutions architect
R · Responsible
Technical PM
C · Consulted
Engineers
R · Responsible
Legal & compliance
C · Consulted
DPO
I · Informed
Security
I · Informed

Go-live decision

Chief procurement officer
A · Accountable
Product owner
R · Responsible
Solutions architect
C · Consulted
Technical PM
R · Responsible
Engineers
I · Informed
Legal & compliance
C · Consulted
DPO
C · Consulted
Security
C · Consulted

R ResponsibleA AccountableC ConsultedI Informed· not involved

Where I'd fit

Option A

Lead the architecture

If there's no architect on the programme, I'd own the design end to end: the sources and how they're matched, the cited summaries and grounding checks, the integration with your supplier master and procurement system, and the gates. I'd work closely with your procurement and compliance leads, and report to your programme board.

Option B

Support your architect

If your partner or your own team already has an architect, I'd take on the source connectors, the retrieval and the research agent under their lead, and give them a second senior opinion on accuracy, data protection and how the alerts reach the right people.

Where this goes wrong

  1. Goes wrong

    The AI matches news to suppliers by name, and a business with a similar name is flagged for something it didn't do.

    How the design avoids it

    Match on company number and known subsidiaries. Name-only hits are marked unconfirmed and never reach an owner as a finding.

  2. Goes wrong

    A summary turns 'was reported to be investigated' into 'committed fraud'.

    How the design avoids it

    Summaries only restate cited sources, keep the source's own hedging, and pass a grounding check. A weekly sample is checked by a person.

  3. Goes wrong

    Owners get hundreds of low-value alerts, and the important one is missed.

    How the design avoids it

    Alert types and thresholds set by the risk policy, severity ranking, and precision tracked and tuned each month.

  4. Goes wrong

    An alert is treated as a decision, and a supplier is dropped without a fair look.

    How the design avoids it

    The AI only flags. A named person decides, with the supplier given a chance to respond where appropriate, and the reasons recorded.

  5. Goes wrong

    The open web is used as a source, so rumours and stale pages feed the summaries.

    How the design avoids it

    Official registers and licensed sources only, each with an owner, and every alert showing its source and date.

On the horizon

Changes already announced or under way that affect this design. More in the UK horizon scan.

  1. Going through Parliament
    The Immigration and Asylum Bill (introduced 30 June 2026, at Commons report stage in November 2026) would amend section 54 of the Modern Slavery Act: prescribed statement content, an accuracy statement from the signatory, submission to the government electronically, financial penalties for the first time, and reporting extended to public bodies over a threshold.

    Organise modern slavery evidence by topic now: risk assessment, policies, due diligence, training and effectiveness. Then the statement can follow whatever content the final Act prescribes.

  2. November 2026 onwards
    Companies House identity verification: mandatory for new directors and people with significant control since 18 November 2025, with existing ones verifying at their company's next confirmation statement during a 12-month transition that ends in November 2026.

    Companies House data should become more reliable over time, but it's still a register, not proof. Keep it as one source among several, and record when each check was made.

  3. Late 2026–2028
    The Cyber Security and Resilience Bill: back in the Commons for Lords amendments. Medium and large managed service providers come into regulation, with duties set by secondary legislation by about 2028.

    Add supplier cyber posture as an alert type for IT and managed service suppliers, and plan for new contract terms. See the supplier cyber assurance playbook.

Questions to ask your partner or vendor

Good answers are specific. Vague answers to any of these are worth following up.

Copy them into your next supplier meeting

  1. Which sources will the AI use, who licenses them, and does it ever search the open web?
  2. How are suppliers matched to news and registers: by name, or by company number?
  3. How do we check that every point in a summary is supported by its source?
  4. Can the AI change a supplier's status in any way, and where is that blocked?
  5. What data about individuals does it collect, and how is that justified under UK GDPR?
  6. What does it cost to run per supplier per year, including the data licences?

Planning something like this?

Whether you need someone to lead the design or to support the architect you already have, let's talk it through.

Sam Naseem
Sam NaseemBirmingham, UK · UK & Europe
Available now