Most enterprise AI programs should not use one commercial model for the entire engagement. Use fixed price where scope, dependencies and acceptance are stable; capped time and materials during discovery and technical uncertainty; milestone payments when capabilities can be demonstrated; and outcome-linked compensation only when baseline, attribution and measurement are auditable. A phased hybrid is usually the defensible default.
A contract cannot make uncertainty disappear. Fixed pricing against immature requirements returns as contingency, change requests or a thin solution. T&M without a budget, backlog and continuation gates shifts too much efficiency risk to the buyer. Outcome pricing without attribution creates arguments over what actually produced the gain.
The AI Commercial Model Fit-8
Score eight conditions from zero to four: zero is unknown; one is a hypothesis; two is partial evidence; three is validated evidence; four is stable and auditable. Do not let the total choose the contract automatically. Use it to see which risks can be priced, shared or retained.
- Definable outcome — decision, user, unit of value, minimum quality and unacceptable failures.
- Stable scope — deliverables, interfaces, volumes, exclusions and change rules.
- Ready data — access, quality, permission, labeling, retention and responsibility for remediation.
- Controllable dependencies — systems, model vendors, APIs, security approvals and client staffing.
- Measurable acceptance — baseline, test set, tolerances, human review and reproducible evidence.
- Known operations — SLOs, monitoring, incidents, usage cost, support and ownership.
- Change velocity — expected learning and changes to models, prompts, workflows or policies.
- Auditable attribution — ability to separate supplier impact from seasonality, media, pricing, staffing and other factors.
Where each model fits
- Fixed price: high confidence in scope, dependencies and acceptance; suitable for closed deliverables or a repeatable stage.
- Capped T&M: low confidence in scope or data; useful for discovery, unknown integration and evaluation with transparency and gates.
- Capability milestones: useful when the buyer can observe baselines, integrations, evaluation, release, operations and transfer.
- Fixed base plus variable: useful when minimum capacity must be funded and a limited component can depend on measurable performance.
- Outcome-based: uncommon; requires a reliable baseline, attribution period, control of levers, data rules, floors, caps and audit rights.
FAR Part 16 is a useful design reference rather than a rule for private deals: firm fixed price shifts maximum cost risk to the contractor, while T&M requires special control because charges grow with labor and materials. UK guidance adds the practical point: pricing must align with risk allocation, and fixed price depends on specification clarity.
Four gates for outcome-linked fees
- The supplier controls a material portion of the levers that affect the metric.
- The baseline and measurement window are agreed before work begins.
- Quality, safety and experience operate as guardrails.
- The formula addresses seasonality, pricing, media, mix, downtime and buyer decisions.
If any gate fails, use capability milestones or a fixed base with a limited bonus. A service agent should not be rewarded only for deflection if that encourages premature closure. A sales agent should not receive credit for all influenced revenue when campaigns, inventory and pricing changed. Outcomes need counter-metrics and auditability.
A four-phase hybrid structure
- Phase 1 — short capped paid discovery: decision, data, risks, baseline, architecture and evaluation plan.
- Phase 2 — validation slice: milestone price for one complete flow, including a failure and recovery.
- Phase 3 — iterative delivery: T&M with named team, rate card, sprint budget, prioritized backlog and go/no-go checkpoints.
- Phase 4 — production: base fee for operations and SLOs, plus limited incentives tied to quality, adoption, unit cost or an auditable outcome.
UK agile-contracting guidance notes that payment-by-results can retain a fixed component to fund supplier capacity. For AI, that reduces pressure to optimize manipulable metrics and lets variable compensation depend on a balanced set of outcomes, quality, risk and knowledge transfer.
What to normalize across proposals
- Total price by phase and usage scenario, not only day rates.
- Assumptions and exclusions; ownership of data, access, experts and environments.
- Acceptance criteria and evidence required for every payment.
- Ceiling, burn rate, estimate to complete and backlog-change authority.
- Model, cloud, license and third-party costs with explicit markups.
- Warranty, remediation, support, transition, documentation and exit.
- Variable formula, guardrails, audit, floor, cap and treatment of external factors.
World Bank rated-criteria guidance supports considering non-price qualities and lifecycle cost alongside price. That matters in AI because a low bid may omit evaluation, operations, security, adoption or exit. Compare cost per accepted capability and per operating outcome, not price alone.
Commercial red flags
- Fixed pricing before access to critical data or systems.
- T&M without backlog, ceiling, burn report or stop decision.
- A bonus tied to one metric with no quality or risk counter-metrics.
- Production payment before objective acceptance.
- Model and cloud costs excluded without usage scenarios.
- Supplier-controlled change classification.
- Evaluations, configuration or documentation withheld as commercial leverage.
Connect price, proposal and contract
Use https://makinai.co/insights/en/how-much-ai-consulting-services-cost to build lifecycle cost, https://makinai.co/insights/en/how-to-evaluate-ai-consulting-proposals-scorecard to compare value and https://makinai.co/insights/en/what-to-include-ai-services-contract-sow to express acceptance, data, IP, operations and exit. Explore https://makinai.co/services/en/ai-strategy-transformation-consulting.
When to involve MAKINAI
MAKINAI can structure discovery, normalize proposals and design a hybrid commercial model that allocates risk without rewarding fragile metrics. The final decision should state what is known, who controls each dependency, which evidence releases payment and how the buyer can stop or change providers.