Practice · XIV.

AI Vendors. Enterprise AI contract advisory.

The OpenAI Enterprise commitment, the Anthropic enterprise contract, the Azure OpenAI consumption design, the Bedrock model and provisioned-throughput commitment, the Vertex AI design, the data residency and training-data clauses, the model deprecation protection, and the intellectual-property indemnification. Enterprise AI contracts close on a use-case-by-use-case commitment and a documented set of model and data clauses, not on the headline price-per-million-tokens. Independent buyer-side advisory across the model providers, the hyperscaler-hosted AI services, and the dedicated enterprise AI platforms.

Speak with a Senior Advisor Engagement models

Where the practice intervenes

Six points on the enterprise AI commercial cycle.

Use case

Workload mapping

Before the commitment is sized, the AI use cases are mapped at business-unit level. Token volume, latency requirement, model tier, data residency requirement and the regulatory regime are documented. The position paper governs the commercial conversation with every model provider and hyperscaler in scope.

Commitment

ii.

Volume design

Commitment volume is sized against documented use cases. Provisioned throughput, capacity reservation, prepaid token packs, and pay-as-you-go burst are mixed against forecast workload. Multi-year ramp, overrun protection, underrun forgiveness and exit-to-PAYG terms are documented at signature.

Data residency

iii.

Data clauses

Data residency, regional model availability, training-data exclusion, customer-data retention and the audit-trail entitlement are documented in the contract. EU AI Act, the wider data protection regime and sector-specific obligations are mapped to specific contract clauses.

Model deprecation

iv.

Lifecycle protection

Model deprecation, model version pinning, deprecation notice period, backward-compatibility commitment and the migration credit on forced deprecation are documented. The model lifecycle protection is filed before the use case is built on the model.

IP / indemnification

IP clauses

Intellectual-property indemnification scope, the customer-content licence, the training-data ownership question, the output ownership clause and the third-party IP indemnification cap are documented. The IP position is approved by counsel before the commercial commitment is signed.

Multi-vendor

vi.

Closing contract terms

The commitment closes against a documented multi-vendor architecture, with model portability mechanics, exit-credit treatment on forced deprecation, residency entitlement, and the renewal anchor terms. The closing contract is reviewed clause by clause. The senior advisor sits opposite the publisher's account team for the full negotiation.

Named senior advisor

The senior advisor named on the engagement letter leads every AI vendor conversation. No leverage model, no rotation of junior consultants, no anonymous correspondence with the publisher. The advisor signs every position document.

II.

Counter-signed positions

Every written counter-position to every AI vendor is signed by the senior advisor and counter-signed by a second partner before transmission. Two named signatures on every position document for the full five-year retention period.

III.

Independent on every clause

Admodum holds no commercial relationship with any AI vendor. No reseller margin, no partner status, no implementation subcontract from OpenAI, Anthropic, AWS, Microsoft, Google or any model provider. We do not work both sides.

IV.

Model deprecation is a clause

Model deprecation, version pinning, deprecation notice period, backward-compatibility commitment and migration credit on forced deprecation are documented in the closing contract. The clause is filed before the use case is built on the model.

Five-year retention

Every position paper, counter-position, and closing memorandum is retained inside the firm for five years from engagement close. The buyer can request source workings at any point.

VI.

Reset the baseline every cycle

Every AI commitment renewal closes with a written baseline reset. The closing position, the contractual amendments accepted, and the live obligations carried forward are tied into the next renewal preparation cycle through the Renewal Programme.

The opening proposal carried a single-vendor commitment at the top model tier across forty business units, with no model deprecation protection and no exit-credit mechanic. After a use-case reconstruction and a multi-vendor architecture redesign, the closing commitment landed at half the proposed run-rate with documented portability and an IP indemnification ceiling.

Chief Technology Officer

Financial Services Group · Enterprise AI Commitment Q1 2026

Operating principles

How the practice is run.

Workload by use case.

AI commitments are sized against documented use cases at business-unit level. Token volume, latency requirement, model tier, data residency and regulatory regime are documented before the commercial conversation opens.

Deprecation is a clause.

Data is a clause.

Data residency, regional model availability, training-data exclusion, customer-data retention and audit-trail entitlement are documented in writing. EU AI Act, GDPR and sector-specific obligations are mapped to specific contract clauses.

Portability at signature.

Model portability, exit-credit on forced deprecation, multi-vendor architecture and the IP indemnification cap are documented at signature. Portability cannot be negotiated at exit because the buyer has lost the leverage.

Questions

Asked before engagement.

What matters most in an enterprise AI vendor contract?

Commitment sizing against measured throughput, IP indemnification scope, data-use and training-exclusion terms, model-deprecation protection and exit portability. The commercial mechanics are young and the clauses vary materially between OpenAI, Anthropic, Google and Microsoft paper.

How do we size a provisioned-throughput commitment?

On pilot telemetry against named use cases. Pilot volume rarely projects linearly to production, and production rarely projects to renewal; the commitment should absorb growth without locking the buyer to today's model catalogue.

Should we commit to one AI vendor?

Keep portability. Benchmark equivalent workloads across at least two providers, contract for model and price review points, and avoid commitments that outlive the model generation they were priced against.

How are engagements priced?

Three frameworks: fixed fee for scoped deliverables, contingency tied to verified savings, and an annual retainer for continuous coverage across every vendor. The model is agreed on the first call, which is always complimentary, and independence is warranted in writing on every engagement letter.