Direct answer: hire an AI strategy firm when the main uncertainty is what to prioritize, how to govern or how the business must change. Hire an implementation company when the use case, owner, data, integrations, success criteria and boundaries are already clear. Choose an integrated partner when strategy and delivery must evolve together—a common AI condition because prototypes expose constraints and opportunities that presentations cannot.
The expensive mistake is buying the wrong partner type. Strategy without a testing mechanism can produce an elegant but infeasible roadmap. Implementation without a clear decision can automate the wrong process or create an ownerless pilot. Official procurement guidance recommends defining the problem, testing viability, evaluating suppliers and planning the full lifecycle. NIST, GAO and ISO also treat governance, data, performance and monitoring as continuing responsibilities.
The MAKINAI Strategy–Build Fit-6
Evaluate six dimensions: Mandate, Portfolio, Evidence, System, Governance and Transfer. For each, record two scores from zero to four: decision uncertainty and internal readiness to deliver. High uncertainty indicates strategy work. Low uncertainty with low delivery capacity favors an implementer. High uncertainty plus low capacity calls for an integrated partner. High readiness on both dimensions may justify narrow specialist support.
1. Mandate: is there a concrete executive decision?
A strategy consultancy should turn generic ambition into choices: objectives, problems, boundaries, risk, horizon, investment and an executive owner. It fits when functions disagree about value, AI changes the operating model or multiple business units compete for priority. Require recorded decisions—not only trends, workshops or an opportunity catalog.
An implementer should receive an executable mandate: user, process, outcome, systems, data, risk, budget and autonomy. If the company must discover all of that before estimating, the engagement contains strategy and should be contracted and governed accordingly. Do not hide strategic discovery inside a fixed-price build; it encourages assumptions or change orders.
2. Portfolio: must the company select one use case or balance many?
When there are dozens of ideas, cross-unit dependencies and scarce resources, seek portfolio strategy capability. The partner should compare value, feasibility, risk, time to learning, asset reuse and competitive advantage. It also needs to say what not to do now. A ranking without assumptions or investment decisions is merely an ordered list.
- Strategy evidence: value thesis; use-case inventory; criteria; dependencies; sequence; funding; operating model; risks; benefit owner. Implementation evidence: journey backlog; architecture; data; integrations; evaluation; security; operations; release plan.
3. Evidence: which learning is required before commitment?
A strong strategy firm designs tests that reduce commercial, operational and technical uncertainty. It links hypothesis, experiment, sample, metric and decision rule. A strong implementer turns the test into observable software with evaluation, versioning and safe behavior. NIST SP 800-218A adds secure development practices specific to AI models and systems.
Ask both partner types for the same artifact: a table showing what is known, what is assumed, how the team will learn, the cost and the decision that follows. Strategy must get close enough to technology to avoid recommending the impossible. Implementation must get close enough to business to avoid optimizing a metric without consequence.
4. System: is the difficulty selection or integration?
If the central problem is choosing target architecture, make-or-buy, model, vendors or modernization sequence, strategic capability weighs more. If architecture is defined and the challenge is integrating ERP, CRM, identity, data, channels and operations, evaluate engineering, product, security, DevOps and support. Architecture slides do not demonstrate the ability to operate a system.
Require evidence from a realistic environment: code, tests, logs, failures, costs and dependencies. Ask how components can be replaced, versions controlled and data removed. An implementer need not master every platform, but it should declare limits and avoid unplanned lock-in.
5. Governance: who accepts risk and tracks performance?
Strategy should design decision rights, inventory, risk classification, policies, exceptions and assurance. Implementation should turn those into access controls, evaluations, approvals, telemetry, incidents and change management. GAO organizes accountability around governance, data, performance and monitoring; ISO/IEC 42001 treats the management system as continual improvement rather than a launch checklist.
Do not let each partner transfer responsibility to the other. The RACI should include sponsor, product owner, risk, data, security, operations, strategy firm and technical provider. With two vendors, assign one owner for trade-off decisions and shared artifacts; otherwise, gaps between scope and build become contractual disputes.
6. Transfer: will the client become more capable?
The strategy firm should transfer method, criteria, decisions, financial models and operating model. The implementer should transfer code, documentation, configuration, tests, evaluations, runbooks and access. In both cases, require applied training and evidence that the internal team can execute one critical activity without the vendor. Permanent dependence should be an explicit economic choice.
Three engagement models
- Strategy first: 6–10 weeks for thesis, portfolio, operating model and test design, followed by implementation procurement. Direct implementation: clear use case, known architecture and accountable product owner; use short discovery and technical gates. Integrated partner: one team works from decision to production in cycles; use when technical evidence changes strategy and learning speed matters.
How to test partners before the main contract
Give each bidder the same case, available data and constraints. Ask the strategy firm for a prioritized decision, hypothesis, business case, governance and experiment. Ask the implementer for architecture, risks, integration plan, evaluation and operations. Require both from an integrated partner. Score clarity, evidence, trade-offs, feasibility, execution, transfer and honesty about gaps.
Red flags and next step
- Roadmap without a test; use-case list without funding; strategy ending in slides; implementer accepting vague requirements without discovery; prototype presented as production; proprietary architecture without an exit; governance disconnected from controls; price without assumptions; no benefit owner; vendors blaming each other; no transfer.
Compare capacity models at https://makinai.co/insights/en/in-house-ai-team-or-ai-consulting-firm, structure requirements with https://makinai.co/insights/en/how-to-write-rfp-ai-services and estimate investment at https://makinai.co/insights/en/how-much-ai-consulting-services-cost. MAKINAI connects decision, build and operation at https://makinai.co/services/en/ai-strategy-transformation-consulting.