Direct answer: choose the firm that can prove, inside one real workflow, how it will connect business priority, task redesign, safe-use guidance, manager capability, value measurement and transfer to the internal team. Workshops, campaigns and training hours are inputs, not the outcome. Adoption exists when people use AI in the right situations, apply the right level of human judgment and produce a verifiable process improvement without adding unmanaged risk or unnecessary dependency.
A common mistake is buying change management as a communications layer applied after the technology. With AI, change begins earlier: which tasks should be automated, augmented or kept human; who is accountable for errors; which data may be used; when a user must review or escalate; and how performance will be evaluated. NIST calls for clear roles in human-AI configurations and a culture of critical thinking and safety. A partner must convert those principles into decisions, routines and observable behavior, not simply publish them.
Training, implementation or transformation?
A training provider teaches tools and practices. An implementation partner configures platforms, integrates data and delivers working features. An adoption firm redesigns how work happens and enables leaders to operate the transition. These scopes can coexist, but their deliverables differ. If the platform does not work, do not label the problem adoption. If it works but employees do not know when to trust, review or embed it in the process, training alone will not fix it. The work is sociotechnical: technology, task, incentive, skill and control change together.
- Buy training when the primary gap is foundational knowledge or proficiency in a stable tool.
- Buy implementation when product, integration, data, security or technical quality is missing.
- Buy adoption and change when capability exists but use, process, roles and value remain inconsistent.
- Use an integrated model when the technology and the work must be redesigned in the same cycle.
The AI Adoption Proof-7
MAKINAI uses seven proofs to compare finalists. Score each from zero to four: zero means absent; one means a promise; two means a documented method; three means evidence from a comparable setting; four means a demonstration in a controlled workflow from your organization. The maximum is 28. Below 20, narrow the scope and authority. Some failures are knockout gates regardless of score: a surveillance-led plan, no participation from affected teams, metrics limited to logins and training, or an attempt to transfer employment and management decisions to the vendor.
- 1. Priority: connects change to an outcome, baseline, population and explicit scale decision.
- 2. Work design: maps tasks, exceptions, handoffs, human judgment and role-level impact.
- 3. Trust: establishes guidance, boundaries, feedback channels and transparent error handling.
- 4. Capability: builds role-specific skills through contextual practice rather than generic courses.
- 5. Management: equips sponsors, managers, owners and champions to remove barriers and sustain habits.
- 6. Measurement: combines qualified use, quality, time, risk, experience and operating results.
- 7. Transfer: leaves playbooks, assets, learning data, governance and internal capability.
How to evaluate proposals
Request a compact evidence room. For priority, ask for a baseline and hypothesis example. For work design, require a before-and-after task map. For trust, review sample guidance and a feedback-response loop. For capability, inspect role-specific learning paths. For management, examine the cadence for sponsors, line managers and champions. For measurement, require a dashboard that separates access, qualified use and outcomes. For transfer, list every asset that will remain with the client. The UK AI Playbook treats adoption planning, organizational impact, change, principles and governance as connected components.
Compare the proposed team as carefully as the methodology. A credible partner combines service or process design, organizational behavior, data and measurement, communications, learning, governance and enough technical depth to challenge the solution. Be cautious of a team composed only of trainers or communicators, and equally cautious of a purely technical team that treats resistance as a knowledge deficit. The UK's 2026 adoption plan reports that culture, behavior and workforce conditions influence adoption as much as technology, while poor early experiences damage trust.
Metrics that help and metrics that mislead
Activated licenses, logins, workshop attendance and prompt volume are activity signals, not proof of value. Measure a ladder. Start with access and readiness. Then measure qualified use in defined tasks. Next observe process change: cycle time, rework, escalation, quality and experience. Finally connect those signals to business outcomes and risk. Segment by role, team and use case because averages hide pockets of value and friction. OECD research associates training and worker consultation with better outcomes, supporting participation as a design input rather than a late communications tactic.
- Keep: accepted task completion, quality, cycle time, rework, safe use, calibrated trust and operating outcome.
- Use carefully: active users, prompts per person, isolated satisfaction and training hours.
- Avoid: individual usage rankings, context-free volume targets and productivity claims inferred from raw telemetry.
- Track unintended effects too: review burden, exclusion, manual workarounds, displaced effort and emerging risks.
Test the partner in a six-week adoption sprint
Choose one consequential workflow, a bounded population and an existing baseline. In week one, align outcome, risk and stakeholders. In week two, observe the work and map tasks and exceptions. In week three, redesign the flow and define guidance. In week four, run contextual enablement and support managers. In week five, operate the pilot and collect quantitative and qualitative evidence. In week six, decide what to stop, repair or scale and complete the handoff. Do not use an opt-in pilot made only of enthusiasts as a proxy for the organization.
Five knockout criteria
- The proposal starts with a campaign or training calendar before observing the work.
- The vendor promises productivity without a baseline, task definition or evaluation method.
- The plan excludes consultation, feedback and participation from affected employees.
- Measurement depends on individual surveillance or context-free usage volume.
- There is no internal owner, governance, transferable asset list or explicit exit condition.
Use this scorecard with MAKINAI's guides to in-house versus consulting, strategy versus implementation, and choosing an AI governance firm. Those decisions establish the delivery and control model; the AI Adoption Proof-7 tests whether a partner can turn available capability into better work. If you already have a platform or pilot and need adoption evidence before expanding the investment, MAKINAI can help select the workflow, establish the baseline and run a measurable validation sprint.