Strategy & Metrics (SM)

Vendors Domain - HAIAMM v3.5

Generated from the canonical cell records in cells/. Do not edit by hand; edit cells/SM-Vendors-L*.cell.json and re-run tools/render.py.

---

Practice Overview

Business Function: Governance

Objective: Stand up vendor AI assurance, inventory every AI tool and service the organization consumes, and baseline the metrics that prove shadow AI adoption is decreasing.

Description: A program at this level knows what AI its people are using. One inventory covers the five vendor archetypes, being consumer generative AI, AI-embedded SaaS, AI coding assistants, AI API and foundation-model vendors, and AI agent or automation platforms. This is the domain where shadow AI is the primary L1 outcome rather than a secondary concern, because adoption happens on a corporate card in minutes and because AI features arrive silently inside SaaS products the organization already approved. Discovery therefore runs across expense records, identity provider app catalogues, network egress, endpoint management and SaaS admin consoles, and an amnesty window does more for coverage than any control.

---

Maturity Level 1, Foundational

Objective: Stand up vendor AI assurance, inventory every AI tool and service the organization consumes, and baseline the metrics that prove shadow AI adoption is decreasing.

A program at this level knows what AI its people are using. One inventory covers the five vendor archetypes, being consumer generative AI, AI-embedded SaaS, AI coding assistants, AI API and foundation-model vendors, and AI agent or automation platforms. This is the domain where shadow AI is the primary L1 outcome rather than a secondary concern, because adoption happens on a corporate card in minutes and because AI features arrive silently inside SaaS products the organization already approved. Discovery therefore runs across expense records, identity provider app catalogues, network egress, endpoint management and SaaS admin consoles, and an amnesty window does more for coverage than any control.

Dependencies

• Required: None.
• Alignment (not blocking): Existing third-party risk management and procurement intake, so AI vendor assessment extends rather than duplicates the vendor lifecycle; Software asset management for seat and licence data
• Supports / unblocks: PC-Vendors-L1; TA-Vendors-L1; SR-Vendors-L1; IM-Vendors-L1; ML-Vendors-L1; EG-Vendors-L1; Required by: DR-Vendors-L1, EH-Vendors-L1, IR-Vendors-L1, SA-Vendors-L1, SM-Vendors-L2, ST-Vendors-L1

Desired Outcomes

• Every AI vendor in use is inventoried against one of the five archetypes and attributed to a business owner.
• AI features silently enabled inside already-approved SaaS products are discovered rather than assumed absent.
• Shadow AI adoption is visible across expense, identity, network and endpoint signals, and is trending down.
• Data-sharing posture with each AI vendor is known, including whether inputs are used for training.
• Employees have a sanctioned path that is faster than the unsanctioned one.

Activities

A) Charter vendor AI assurance and open an amnesty window

Publish a charter within the first 30 days sponsored by the CISO jointly with the Chief Procurement Officer or CIO, co-signed by Privacy counsel. Name the five in-scope archetypes explicitly, because the most commonly missed category is the AI feature that appeared inside an already-approved SaaS product without a new contract, a new invoice or any procurement event at all. Establish the working group across Security, Procurement, Legal and Privacy, IT, Data Governance and one business-unit representative, and define decision rights over who sanctions an AI vendor, who blocks one, who grants exceptions and who owns offboarding. Open an amnesty window of at least 60 days, publicized through all-hands and team channels, under which anyone can disclose an AI tool they are using with no disciplinary consequence. Amnesty is the single highest-yield discovery mechanism in this domain and it only works once, so pair it with a sanctioned path that is genuinely faster than signing up with a credit card, or disclosure will not repeat.

B) Discover across five signals and inventory by archetype

Build one AI vendor inventory recording per vendor the archetype, business owner, sanction status, data classes shared, whether inputs are used for provider training and how that is contractually established, sub-processor chain, data residency and transfer basis, model provider behind the service, seat count and utilisation, contract and renewal dates, and links to the TA snapshot and PC compliance mapping. Reconcile monthly against five discovery signals: expense and procurement records including corporate card transactions, which catch the self-service adoptions that never reached procurement; identity provider application catalogues and single-sign-on grants, which catch tools users authenticated into; network egress to known AI provider domains, which catches everything with a browser or API footprint; endpoint management and browser extension audits, which catch locally installed assistants; and the admin consoles of major approved SaaS products, which is the only way to see AI features that were switched on inside a product the organization already owns. Treat any vendor appearing in a signal but not the inventory as shadow AI until attributed.

C) Baseline the scoreboard and open the vendor spend ledger

Instrument and report quarterly: AI vendor inventory coverage against discovery reconciliation, the shadow-AI ratio as the domain's primary L1 outcome, acceptable-use attestation across the workforce, the share of sanctioned vendors with a documented training-data position, and known data-exposure events to unsanctioned AI tools. Open the vendor consumption ledger recording cost per active seat and cost per token or unit of service, attributed to the business owner, together with committed-spend terms, overage rates, price-change notice periods and exit cost. This domain is the only one where consumption is contractually governed rather than merely observed, which makes the ledger a negotiating instrument as well as a control: concentration risk becomes a number in dollars, and lock-in becomes a measured switching cost rather than an asserted worry. Track seat utilisation alongside seat cost, since provisioned-and-never-used seats are simultaneously wasted spend and unnecessary data exposure.

Outcome Metrics (L1)

Metric	Baseline	L1 Target	Source
AI vendor inventory coverage against discovery reconciliation	first five-signal sweep	≥90% within 12 months	Inventory reconciled against expense, IdP, egress, endpoint and SaaS-admin signals; where no central identity provider covers the tool, per-vendor admin console user lists, recorded at E1
Shadow AI ratio (unsanctioned AI vendors in use ÷ total AI vendors in use)	first sweep, taken at the close of the amnesty window	≤15% and falling for two consecutive quarters	Inventory sanction-status field
Workforce with an acknowledged AI Acceptable Use Policy	first attestation run	≥95% of workforce	HR / LMS attestation records
Sanctioned AI vendors with a documented, contractually established training-data position	first contract review pass	100% of vendors receiving confidential or regulated data	Contract register joined to inventory data-class field
Known data-exposure events to unsanctioned AI tools	first quarter of DLP and incident telemetry	falling quarter over quarter, with zero involving regulated data	DLP alerts and incident tracker

Process Metrics (leading)

• Five-signal discovery sweep runs monthly with its last-run timestamp published.
• SaaS admin-console review for newly enabled AI features runs at least quarterly per major product.
• Intake service-level agreement: new AI vendor requests triaged within 5 business days.
• Renewal calendar reviewed monthly so reassessment precedes auto-renewal rather than following it.

Effectiveness Metrics (business value)

• Time from AI vendor request to sanctioned availability falls, which is what makes the sanctioned path win.
• Vendor consolidation: the number of distinct vendors serving the same archetype falls without capability loss.
• Documented cases where discovery caught confidential data flowing to an unsanctioned tool before it became reportable.

Consumption Metrics (denominated, never absolute spend)

Metric	Denominator	Economic failure it detects	Source
Cost per active seat, by vendor and archetype	per active seat	licence waste where seats are provisioned and never used, which is wasted spend and standing data exposure at the same time	SaaS management platform joined to identity provider activity data
Cost per unit of service against committed-spend terms and overage rates	per governed workload	overage exposure and unbudgeted price changes on renewal, and concentration that is only visible once consumption is priced per vendor	Vendor invoices joined to the contract register
Assured Spend Ratio for vendor AI: consumption through sanctioned vendors	per dollar of vendor AI consumption	a healthy vendor count masking that most actual AI consumption flows through tools the program has not sanctioned	Expense and invoice data joined to inventory sanction status

Level of Effort (incremental over the level below)

• Stand-up: 150 hours · Ongoing: 80 hours/quarter
• Program Manager / Practice Lead — primary, 26 days/year
• Procurement / Sourcing — regular, 15 days/year
• TPRM Analyst — regular, 18 days/year
• Legal / Privacy Counsel — regular, 10 days/year
• Executive Sponsor — low, 3 days/year
• Material new tooling: SaaS management platform or equivalent covering expense, IdP and endpoint signals, Egress monitoring against a maintained AI-provider domain list

Success Criteria

• A charter is published and jointly sponsored, naming all five archetypes and four decision rights, with a publicized amnesty window of at least 60 days completed.
• One AI vendor inventory reaches 90% coverage or better against five-signal discovery within 12 months.
• The shadow AI ratio is baselined at the close of amnesty and has fallen for two consecutive quarters.
• Every vendor receiving confidential or regulated data has a contractually established training-data position on file.
• A vendor consumption ledger is open recording cost per active seat and per unit of service, with seat utilisation tracked beside seat cost.

---

Maturity Level 2, Comprehensive

Objective: Tier AI vendors by data exposure and dependency, calibrate assessment depth per tier, and govern concentration, committed spend and exit cost as measured positions.

At this level vendor assurance intensity follows the exposure rather than the contract value. A published rubric derives a tier from data classes shared, criticality of the dependency, autonomy of the service, sub-processor depth and switching cost, and a treatment matrix states what each tier receives in assessment depth, contractual terms, monitoring and reassessment cadence. Concentration and lock-in stop being adjectives: this is the one domain where consumption is contractually governed, so exposure can be priced. A Critical-tier vendor that cannot be exited inside a quarter at a tolerable cost is a governed risk position with a named owner, not a background fact.

Dependencies

• Required: SM-Vendors-L1, PC-Vendors-L1, TA-Vendors-L1
• Alignment (not blocking): Enterprise third-party risk tiering, so AI vendor tiers compose with the existing supplier scheme
• Supports / unblocks: PC-Vendors-L2; TA-Vendors-L2; SR-Vendors-L2; SA-Vendors-L2; DR-Vendors-L2; IR-Vendors-L2; ST-Vendors-L2; EH-Vendors-L2; IM-Vendors-L2; ML-Vendors-L2; Required by: EG-Vendors-L2, SM-Vendors-L3

Desired Outcomes

• Every AI vendor carries a tier derived from data exposure, dependency criticality, service autonomy, sub-processor depth and switching cost.
• Assessment depth, contractual terms and reassessment cadence are differentiated per tier and enforced.
• Concentration exposure is priced in dollars per vendor and carries an explicit accept, mitigate or migrate decision.
• Switching cost and exit time are measured for every Critical-tier vendor rather than assumed tolerable.
• Committed-spend and overage terms are governed, so consumption growth does not arrive as an unbudgeted invoice.

Activities

A) Publish the vendor risk-tier rubric

Derive four tiers from five auditable dimensions. Data exposure is the floor: a vendor receiving regulated or customer-confidential content sits at Critical or High regardless of spend, because contract value and risk exposure are only weakly correlated in this domain and a cheap tool can carry the whole customer base through it. Dependency criticality elevates where the vendor sits on a revenue path or a customer-facing flow. Service autonomy elevates where the vendor's AI acts rather than advises, which is what distinguishes an agent platform from a drafting assistant. Sub-processor depth elevates where the vendor's own model provider and its sub-processors extend the chain beyond the point where the organization has visibility or contractual reach. Switching cost elevates where exit would take longer than a quarter or require re-engineering, because a dependency that cannot be exited is a dependency that cannot be governed by the threat of exit. Publish the rubric and re-derive when data classes or service scope change.

B) Calibrate assessment depth and contract terms per tier

Publish a tier-treatment matrix covering per tier: assessment depth from questionnaire to evidence-backed review with technical validation; required contractual terms including training-data prohibition, sub-processor notification and approval rights, audit and evidence rights, incident-notification windows, model-change notification, data-residency commitments and deletion-on-termination obligations; monitoring obligations including provider status and advisory tracking; reassessment cadence and the events that force reassessment ahead of schedule; and exit requirements including data-export format and a tested exit runbook for Critical tier. Add the consumption row: Critical- and High-tier vendors carry governed committed-spend terms with a defined overage rate, a price-change notice period and a documented exit cost. Treat a vendor's model change, a sub-processor addition or a material pricing change as a material change triggering reassessment inside the tier window, since each alters the risk profile that the original assessment described.

C) Price concentration and exit, and report per tier

Produce a standing concentration and dependency statement, refreshed quarterly and signed by the sponsor, that prices exposure rather than describing it: the share of AI consumption carried by each vendor and each underlying model provider, the single points of failure that follow, the measured exit time and switching cost for every Critical-tier vendor, and an explicit accept, mitigate or migrate decision for each. Measuring exit cost requires actually specifying the exit, which is why most organizations never do it, and why the number is worth having. Report the scoreboard per tier quarterly covering inventory and tier state, shadow AI ratio per tier with unsanctioned Critical-tier vendors targeted at zero, contractual-term conformance per tier, reassessment currency, seat utilisation by tier, and the Assured Spend Ratio for vendor consumption. Maintain a tier-movement log recording the triggering dimension and review it with the sponsor.

Outcome Metrics (L2)

Metric	Baseline	L2 Target	Source
AI vendors carrying a current tier assignment	L1 inventory coverage	100% of vendors in use, with tier age under 12 months	Inventory tier field
Critical- and High-tier vendors with all required contractual terms in force	first contract conformance audit	≥95% of Critical, ≥85% of High	Contract register conformance review joined to inventory tier
Unsanctioned Critical-tier AI vendors in use	L1 shadow ratio re-cut per tier	0, sustained	Inventory tier and sanction-status fields
Critical-tier vendors with a measured exit time and documented switching cost	first exit assessment pass	100%, refreshed annually	Concentration and dependency statement
Reassessment currency: Critical- and High-tier vendors reassessed within their tier cadence	first cadence audit	≥95%	TPRM assessment records joined to inventory tier
Seat utilisation at Critical and High tier	first utilisation report	≥80% of provisioned seats active in the last 30 days	SaaS management platform joined to identity provider activity; where none is in use, expense records reconciled against identity grants, recorded at E1

Process Metrics (leading)

• Tier re-derivation triggers on data-class or service-scope change, and each run is logged.
• Vendor model-change and sub-processor notifications are triaged within 10 business days of receipt.
• Renewal calendar drives reassessment to complete before auto-renewal, not after.
• Concentration statement refreshed quarterly and signed by the sponsor.

Effectiveness Metrics (business value)

• Vendor consolidation within archetypes reduces the count of Critical-tier relationships without capability loss.
• Negotiated outcomes improve as concentration and exit cost become quantified positions rather than assertions.
• Reclaimed licence waste from unused seats, reported quarterly.
• Assessment effort concentrates on Critical and High tiers relative to L1.

Consumption Metrics (denominated, never absolute spend)

Metric	Denominator	Economic failure it detects	Source
Cost per active seat by tier, with utilisation	per active seat	licence waste concentrated at high tiers, where an unused seat carries standing access to the vendors handling the most sensitive data	SaaS management platform joined to identity provider activity and inventory tier
Overage as a share of committed spend, by vendor	per governed workload	consumption growth arriving as an unbudgeted invoice, and pricing changes taking effect without the notice period being enforced	Vendor invoices reconciled against the contract register commitment terms
Concentration: share of vendor AI consumption carried by the largest vendor and the largest underlying model provider	per dollar of vendor AI consumption	dependency accumulated by convenience rather than decided, where the underlying model provider concentration is often higher than the vendor concentration suggests	Invoice data joined to inventory model-provider field
Measured exit cost as a multiple of the vendor's annualised contract value	per Critical-tier vendor relationship	lock-in that is asserted as a worry but never priced, so the threat of exit carries no weight in negotiation and migration is never budgeted	Concentration and dependency statement exit assessment

Level of Effort (incremental over the level below)

• Stand-up: 140 hours · Ongoing: 110 hours/quarter
• Program Manager / Practice Lead — primary, 22 days/year
• TPRM Analyst — regular, 24 days/year
• Procurement / Sourcing — regular, 14 days/year
• Legal / Privacy Counsel — regular, 12 days/year
• Executive Sponsor — low, 3 days/year
• Material new tooling: Contract register with structured AI-specific clause tracking, Seat utilisation reporting joined to identity provider activity

Success Criteria

• A vendor tier rubric is published and applied to 100% of vendors in use, re-derived on data-class or service-scope change.
• A tier-treatment matrix is published covering assessment depth, contractual terms, monitoring, reassessment cadence and exit requirements.
• Unsanctioned Critical-tier AI vendors stand at zero, and Critical-tier contractual-term conformance is at 95% or better.
• Every Critical-tier vendor has a measured exit time and documented switching cost, refreshed annually.
• A sponsor-signed concentration and dependency statement prices vendor and model-provider exposure with an explicit decision per single point of failure.

---

Maturity Level 3, Industry-Leading

Objective: Drive vendor discovery and tiering from live identity, expense and egress feeds, benchmark vendor unit economics, and contribute AI vendor assurance intelligence back.

At this level vendor discovery is continuous rather than campaign-based. Identity-provider grants, expense feeds, egress telemetry and SaaS admin APIs drive the inventory, so a new AI tool surfaces in hours rather than at the next sweep. Vendor unit economics are benchmarked, which turns negotiation from assertion into comparison, and the program contributes assessment criteria and observed vendor-layer techniques back to the community. Return is computed from the vendor ledger: reclaimed licence waste, avoided overage and consolidation savings, all read from invoices rather than modelled.

Dependencies

• Required: SM-Vendors-L2, ML-Vendors-L2, EG-Vendors-L2
• Alignment (not blocking): FinOps at Run maturity for the SaaS and AI-vendor scope
• Supports / unblocks: Every other Vendors-domain practice at L3, which inherits continuous discovery and tier data

Desired Outcomes

• New AI vendor adoption surfaces within hours through continuous identity, expense and egress feeds rather than at the next campaign.
• Tier assignment updates automatically when a vendor's data classes, service scope or sub-processor chain changes.
• Vendor unit economics are benchmarked against named peers, giving negotiation a comparison rather than an assertion.
• The organization is a net contributor of AI vendor assessment criteria and observed vendor-layer techniques.
• Vendor-domain return is computed from invoice and ledger arithmetic and reconciled with Finance.

Activities

A) Make discovery continuous across identity, expense and egress

Replace the monthly sweep with continuous feeds. Stream identity-provider application grants and first-authentication events, so a new single-sign-on grant to an AI service raises a discovery event the same day; stream expense and corporate-card feeds filtered against a maintained AI-vendor merchant list; stream network egress against a maintained AI-provider domain list that is itself updated from threat-intelligence and community sources rather than by hand; and poll the admin APIs of major approved SaaS products for newly enabled AI features, which remains the only reliable way to see AI that arrives without a transaction. Wire tier derivation to re-run automatically when a vendor's recorded data classes, service scope or sub-processor chain changes, and to open downstream reassessment obligations on tier increase. Publish a data-quality objective covering time from first adoption signal to inventory record, targeting 48 hours, and reserve human curation for ambiguous merchant records, shared-account adoptions and genuinely novel vendor archetypes.

B) Benchmark vendor unit economics and assessment depth

Publish a semi-annual brief comparing the program against named peers on at least five metrics: vendor inventory coverage and discovery latency, shadow AI ratio, contractual-term conformance at Critical tier, cost per active seat by archetype, and seat utilisation. Vendor unit economics benchmark unusually well because the products are the same across organizations, so a materially higher cost per active seat for the same archetype is a negotiating position rather than an observation, and a materially lower one usually indicates either better terms worth sharing or unused capacity worth reclaiming. Draw comparison data from peer chief information security officer communities, sector information-sharing bodies, the FinOps Foundation SaaS and AI scopes, and Shared Assessments or equivalent third-party risk communities. Name the cohort, defend its selection, and record where the program trails. Feed the deltas directly into the renewal calendar so that comparisons arrive before negotiations rather than after them.

C) Contribute assessment intelligence and compute the vendor return

Contribute at least four substantive artifacts per year: AI vendor assessment criteria and questionnaire modules published through Shared Assessments, CSA AI Safety Initiative or sector bodies; observed vendor-layer techniques such as silently enabled AI features, training-data term changes on renewal, and sub-processor chain extensions submitted to MITRE ATLAS or community advisories; anonymized contractual-term benchmarks that help peers negotiate; and OWASP LLM working-group input on third-party and supply-chain risks. Count only externally accepted artifacts. Compute the vendor-domain return from invoice and ledger arithmetic: reclaimed licence waste from deprovisioned unused seats, which is usually the largest and most defensible component; avoided overage where committed-spend governance caught consumption growth before invoicing; consolidation savings where duplicate archetype coverage was retired; and avoided price increases where notice periods were enforced. Reconcile with Finance and Procurement annually and publish alongside the domain Assurance Cost Ratio.

Outcome Metrics (L3)

Metric	Baseline	L3 Target	Source
Time from first adoption signal to an inventory record	L2 monthly sweep interval	≤48 hours at the 90th percentile	Discovery event pipeline telemetry
Vendor inventory entries created without human curation	first quarter of automation telemetry	≥75% auto-curated	Curation telemetry from the discovery pipeline
Tier re-derivation triggered automatically on vendor scope or sub-processor change	L2 manual re-tier cadence	100% within 5 business days of the notification	Tier derivation event log
Peer-comparable metrics in the semi-annual brief	0	≥5 metrics with a named and defended peer cohort	Published benchmarking brief
Externally accepted vendor assurance contributions per year	0	≥4 accepted, excluding talks and press	Contribution log with external acceptance references

Process Metrics (leading)

• Discovery feed freshness and error rate monitored, with alerting on staleness breach.
• AI-vendor merchant and provider-domain lists refreshed from community sources at least monthly.
• Benchmark deltas delivered to the renewal calendar at least 60 days before each Critical-tier renewal.
• Contribution pipeline carries at least two artifacts in draft or external review.

Effectiveness Metrics (business value)

• Negotiated savings and term improvements attributable to benchmark data, reported per renewal cycle.
• Reclaimed licence waste from continuous utilisation review, reported quarterly.
• Computed vendor return reconciles with Finance and Procurement annually without restatement.
• Peer adoption of published assessment criteria.

Consumption Metrics (denominated, never absolute spend)

Metric	Denominator	Economic failure it detects	Source
Cost per active seat by archetype, benchmarked against the peer cohort	per active seat	paying materially above the peer rate for an identical product, which is invisible internally because the invoice looks the same every month	Vendor ledger compared against the semi-annual benchmarking brief
Reclaimed licence waste from deprovisioned unused seats	per active seat recovered	seats accumulating through joiner-mover-leaver drift, carrying both cost and standing access to vendor-held data	SaaS management platform deprovisioning records joined to the vendor ledger
Avoided overage under committed-spend governance	per governed workload	consumption growth crossing a commitment threshold unnoticed and arriving as an unbudgeted invoice at a punitive overage rate	Invoice reconciliation against contract-register commitment terms

Level of Effort (incremental over the level below)

• Stand-up: 180 hours · Ongoing: 70 hours/quarter
• Program Manager / Practice Lead — primary, 18 days/year
• TPRM Analyst — regular, 20 days/year
• Procurement / Sourcing — regular, 14 days/year
• DevOps / Platform Engineer — low, 6 days/year
• Compliance / Audit — low, 4 days/year
• Material new tooling: Continuous discovery feeds from identity provider, expense system and egress telemetry, SaaS admin API polling for newly enabled AI features

Success Criteria

• Discovery is continuous across identity, expense, egress and SaaS admin feeds, with adoption-to-inventory latency at or below 48 hours at the 90th percentile.
• Tier re-derivation triggers automatically on vendor scope or sub-processor change and opens downstream reassessment within 5 business days.
• A semi-annual benchmarking brief covers at least five peer-comparable metrics, delivered to the renewal calendar ahead of Critical-tier negotiations.
• At least four externally accepted vendor assurance contributions per year.
• Vendor-domain return is computed from invoice and ledger arithmetic and reconciled with Finance and Procurement annually.

---

Key Success Indicators

Level 1:

• A charter is published and jointly sponsored, naming all five archetypes and four decision rights, with a publicized amnesty window of at least 60 days completed.
• One AI vendor inventory reaches 90% coverage or better against five-signal discovery within 12 months.
• The shadow AI ratio is baselined at the close of amnesty and has fallen for two consecutive quarters.
• Every vendor receiving confidential or regulated data has a contractually established training-data position on file.
• A vendor consumption ledger is open recording cost per active seat and per unit of service, with seat utilisation tracked beside seat cost.

Level 2:

• A vendor tier rubric is published and applied to 100% of vendors in use, re-derived on data-class or service-scope change.
• A tier-treatment matrix is published covering assessment depth, contractual terms, monitoring, reassessment cadence and exit requirements.
• Unsanctioned Critical-tier AI vendors stand at zero, and Critical-tier contractual-term conformance is at 95% or better.
• Every Critical-tier vendor has a measured exit time and documented switching cost, refreshed annually.
• A sponsor-signed concentration and dependency statement prices vendor and model-provider exposure with an explicit decision per single point of failure.

Level 3:

• Discovery is continuous across identity, expense, egress and SaaS admin feeds, with adoption-to-inventory latency at or below 48 hours at the 90th percentile.
• Tier re-derivation triggers automatically on vendor scope or sub-processor change and opens downstream reassessment within 5 business days.
• A semi-annual benchmarking brief covers at least five peer-comparable metrics, delivered to the renewal calendar ahead of Critical-tier negotiations.
• At least four externally accepted vendor assurance contributions per year.
• Vendor-domain return is computed from invoice and ledger arithmetic and reconciled with Finance and Procurement annually.

---

Common Pitfalls

Level 1:

• ❌ Inventorying only vendors that went through procurement, which misses every self-service adoption and every AI feature enabled inside an existing SaaS contract.
• ❌ Running an amnesty window without first building a faster sanctioned path, so disclosure happens once and never repeats.
• ❌ Accepting a vendor's marketing statement that it does not train on customer data instead of establishing the position contractually.
• ❌ Tracking seat cost without utilisation, so provisioned-and-unused seats persist as both waste and standing data exposure.
• ❌ Reviewing vendors at renewal only, so an AI capability added mid-contract is never assessed.

Level 2:

• ❌ Tiering by contract value, which in this domain correlates weakly with exposure: a low-cost tool can carry the entire customer base through it.
• ❌ Assessing the vendor while ignoring the model provider behind it, so sub-processor concentration is far higher than the vendor list suggests.
• ❌ Writing exit requirements into contracts without ever testing an export, so the exit runbook is discovered to be fiction during the incident that requires it.
• ❌ Treating a vendor's model change as routine maintenance rather than a material change to the system that was assessed.
• ❌ Describing concentration and lock-in qualitatively, which gives negotiation nothing to work with and migration no budget line.

Level 3:

• ❌ Automating expense and identity discovery while leaving SaaS admin consoles on a manual cycle, which misses the category that generates no transaction at all.
• ❌ Maintaining the AI-provider domain list by hand, so it decays and egress discovery quietly stops finding anything new.
• ❌ Producing benchmark data after renewals rather than before them, which makes the comparison interesting instead of useful.
• ❌ Counting reclaimed seats as savings without confirming deprovisioning actually removed the access as well as the charge.
• ❌ Publishing contractual benchmarks without legal review, which can breach confidentiality terms in the very contracts being compared.

---

Practice Maturity Questions

Level 1:

1. Is there a published vendor AI assurance charter jointly sponsored by the CISO and the procurement or IT leader, naming all five vendor archetypes, four decision rights and a completed amnesty window of at least 60 days paired with a sanctioned path faster than self-service adoption?
2. Does one AI vendor inventory reconcile monthly against expense and procurement records, identity provider catalogues, network egress, endpoint and browser-extension audits, and the admin consoles of major approved SaaS products, reaching 90% coverage or better within 12 months?
3. Do all vendors receiving confidential or regulated data have a contractually established training-data position, and is a consumption ledger open recording cost per active seat with utilisation, committed-spend terms, overage rates and exit cost?

Level 2:

1. Is every AI vendor assigned a tier from a published rubric covering data exposure, dependency criticality, service autonomy, sub-processor depth and switching cost, re-derived when data classes or service scope change?
2. Is there a published tier-treatment matrix covering assessment depth, required contractual terms including training-data prohibition and model-change notification, monitoring, reassessment cadence and tested exit runbooks, with at least 95% Critical-tier contractual conformance?
3. Does a sponsor-signed concentration and dependency statement price the share of consumption carried by each vendor and each underlying model provider, with measured exit time and switching cost for every Critical-tier vendor and an explicit accept, mitigate or migrate decision?

Level 3:

1. Is discovery continuous across identity-provider grants, expense feeds, egress telemetry and SaaS admin APIs, with adoption-to-inventory latency at or below 48 hours at the 90th percentile and at least 75% of entries auto-curated?
2. Does the semi-annual benchmarking brief compare at least five peer-comparable metrics including cost per active seat and seat utilisation by archetype, with the cohort named and the deltas delivered to the renewal calendar before Critical-tier negotiations?
3. Does the program contribute at least four externally accepted vendor assurance artifacts per year, and is the vendor return computed from invoice arithmetic covering reclaimed licence waste, avoided overage and consolidation savings, reconciled with Finance and Procurement?

---

Lightweight Questions (E0 rung)

Answerable yes/no by one person in under 30 seconds, with nothing open.

Level 1:

• Do you have one list of every AI tool your organization pays for?
• Would you know if an approved SaaS product switched on an AI feature?
• Can you say whether any vendor trains on the data you send them?

Level 2:

• Does every AI vendor carry a risk tier?
• Do you know what it would cost to leave your largest AI vendor?
• Are you paying for AI seats nobody has used this month?

Level 3:

• Would a new AI tool adopted today show up in your inventory this week?
• Do you know whether you pay above or below peer rates for your AI tools?
• Have you reclaimed unused AI seats this quarter?

---

Tracking

Level 1 — Jira Epic, labels haiamm-sm-vendors-l1, haiamm-governance

• Done when: Vendor inventory coverage at or above 90% with the shadow AI ratio at or below 0.15 and falling across two consecutive quarterly reports, and every confidential-data vendor carrying a contractual training-data position.
• Git artifacts: vendors/ai-vendor-inventory.json, vendors/contract-register.json, governance/vendor-ai-charter.md, finops/vendor-ledger.json
• Git signal: ai-vendor-inventory.json is regenerated monthly by the five-signal discovery job, and contract-register.json records a training-data position field for every vendor receiving confidential or regulated data.
• CI check: vendor-discovery-reconcile

Level 2 — Jira Epic, labels haiamm-sm-vendors-l2, haiamm-governance

• Done when: Vendor tier coverage at 100% with zero unsanctioned Critical-tier vendors and Critical-tier contractual conformance at or above 95%, sustained across two consecutive quarters.
• Git artifacts: vendors/tier-rubric.md, vendors/tier-treatment-matrix.md, vendors/contract-register.json, governance/concentration-statement.md, vendors/exit-runbooks/
• Git signal: concentration-statement.md carries a sponsor signature block and a commit within every quarterly interval, and each Critical-tier vendor has an exit runbook file with a recorded test date.
• CI check: vendor-contract-conformance

Level 3 — Jira Epic, labels haiamm-sm-vendors-l3, haiamm-governance

• Done when: Adoption-to-inventory latency at or below 48 hours at the 90th percentile with auto-curation at or above 75%, sustained two quarters, and the benchmarking brief current within its interval.
• Git artifacts: vendors/discovery-pipeline/, vendors/ai-provider-domains.json, governance/benchmarking-brief.md, governance/contribution-log.json, finops/vendor-return.json
• Git signal: ai-provider-domains.json is refreshed from community sources monthly rather than hand-edited, and vendor-return.json is regenerated from invoice data.
• CI check: vendor-discovery-latency

---

Mappings

Level	HAI TTP	ATLAS	OWASP LLM 2025	Compliance
L1	—	AML.T0010	LLM02, LLM03	EU AI Act Art. 25; GDPR Art. 28; GDPR Art. 44-49; ISO/IEC 27001 A.5.19; ISO/IEC 27001 A.5.21; ISO/IEC 42001 8.4; SOC 2 CC9.2
L2	EA	AML.T0010, AML.T0029	LLM02, LLM03, LLM10	DORA ICT third-party risk (where in scope); EU AI Act Art. 25; GDPR Art. 28; GDPR Art. 44-49; ISO/IEC 27001 A.5.19; ISO/IEC 27001 A.5.21; ISO/IEC 42001 8.4; SOC 2 CC9.2
L3	EA	AML.T0010, AML.T0029	LLM03, LLM10	EU AI Act Art. 25; GDPR Art. 28; ISO/IEC 27001 A.5.19; ISO/IEC 27001 A.5.21; ISO/IEC 42001 9.1; SOC 2 CC9.2

---

Document Version: HAIAMM v3.5 (generated) / Practice: Strategy & Metrics (SM) / Domain: Vendors