Evaluate the people who will do the work, not only the firm, résumé deck or sales team. Before award, require names, roles, allocation, start dates, location, delivery evidence and decision rights. Then test the proposed team in a paid session using a representative problem. If the provider cannot commit critical people, the proposal is not yet executable.
The World Bank's rated-criteria guidance explicitly includes key personnel, capacity, methodology and risk as non-price factors. NIST's AI RMF calls for documented responsibilities, clear communication lines, trained partners and teams with multiple disciplines. For AI services, every promise should map to an accountable owner across product, data, engineering, evaluation, security, operations and adoption.
The AI Delivery Team Proof-7
Score each proof from zero to four: zero is missing; one is a sales assertion; two is a checkable description; three is relevant evidence; four is evidence plus a demonstration in your context. Require at least 21 of 28, no zero and passage of five mandatory gates. Lock weights before proposals arrive so a famous résumé cannot hide operating gaps.
- Accountability — named owners for outcomes, technology, data, risk and decisions.
- Relevant evidence — comparable delivery by context, complexity and lifecycle stage without demanding confidential client disclosure.
- Real allocation — percentage, dates, duration, time zones, language and competing commitments.
- Cross-functional coverage — product, domain, data, engineering, UX, evaluation, security, legal and operations as risk requires.
- Supply-chain transparency — employees, external specialists, subcontractors, access locations and flowed-down duties.
- Continuity — substitution, escalation, redundancy, documentation and knowledge recovery.
- Transfer — repositories, runbooks, training, decision records, evaluation assets and increasing buyer autonomy.
Five mandatory gates
Reject or clarify when the sales team is not the proposed delivery team; no one owns technical integrity or the business outcome; critical roles lack allocation and a start date; subcontractors or data-access locations are undisclosed; or continuity and knowledge transfer are absent. A high average cannot compensate for any of these conditions.
Ask for evidence without demanding confidential client names
For each key role, request two examples of consequential decisions and one redacted artifact: an evaluation plan, architecture, risk register, product backlog, runbook or incident analysis. Providers can anonymize clients and data. The objective is to see how a person reasons, communicates limits and creates reusable mechanisms—not to collect logos.
Distinguish laboratory, deployment and production experience. A strong researcher may not have led integration, operating change or incident response. A platform engineer may not know how to design decision metrics. Team composition should match the engagement stage and risk profile rather than a generic org chart.
Validate allocation, authority and dependencies
Build a person-by-person table with role, weekly allocation, start, minimum tenure, backup, time zone, language, location, expected access and authority. Ask how many concurrent engagements each person supports. Twenty percent allocation may suit specialist review; it is fragile when the same person must clear daily blockers.
Document who can approve architecture, accept risk, change scope, release to production and escalate to the executive sponsor. Without decision rights, the engagement gains meetings and loses speed. Name corresponding buyer-side owners so the consultancy does not inherit decisions the enterprise must make.
Run a paid finalist working session
Hold a 90- to 120-minute session with the proposed people, not substitutes. Provide context, constraints and one failure scenario: quality below baseline, a data change, cost above the ceiling or an unavailable integration. Ask the team to frame questions, assign ownership, propose tests, state trade-offs and produce a one-page decision record. Pay for the exercise; do not request free production work.
Use the same scenario, time box and evaluators for every finalist. Score collaboration, question quality, business-to-technical reasoning, uncertainty handling, security awareness, willingness to say no and artifact quality. Capture evidence immediately so charisma and brand reputation do not replace observation.
Trade-offs buyers should make explicit
- Seniority versus availability: a renowned advisor for two hours is not present delivery leadership.
- Specialists versus integrated team: depth helps, while excessive handoffs create queues and context loss.
- Local versus distributed: proximity supports collaboration; distribution expands talent and coverage.
- Named people versus flexibility: commitment reduces risk; controlled substitution preserves provider capacity.
- Speed versus transfer: isolated delivery may move quickly; buyer autonomy requires documentation and joint practice.
Turn the proposed team into SOW commitments
Define key people or minimum role profiles, allocation, substitution notice, buyer approval for critical changes, subcontractor disclosure, access locations, equivalent downstream duties, onboarding and transition. Tie milestones to artifacts and outcomes. The European Commission's AI model clauses are a useful reference for responsibilities and documentation, but counsel should adapt them to governing law and risk.
Keep using the scorecard after award. Compare the promised team with observed time, decisions, artifacts and outcomes. Reassess risk when a person changes, allocation falls, a subcontractor appears or another data class becomes accessible. The goal is not to freeze staffing; it is to make changes visible and controlled.
Connect team proof to proposal, pilot and contract
Use https://makinai.co/insights/en/how-to-evaluate-ai-consulting-proposals-scorecard for the full proposal, https://makinai.co/insights/en/how-to-run-paid-ai-pilot-before-hiring-partner for delivery evidence and https://makinai.co/insights/en/what-to-include-ai-services-contract-sow for commitments. Explore https://makinai.co/services/en/ai-strategy-transformation-consulting.
When to involve MAKINAI
MAKINAI can help define the delivery model, build the scorecard and facilitate comparable technical-commercial sessions with finalists. The result should be a traceable decision about who will deliver, with what capacity, under which controls and how capability will remain with the buyer.