VANKPA
The useful takeaway
Readiness is the ability to operate and evaluate a specific AI workflow, not simply access to a model.
An AI readiness assessment should answer a practical question: which task can the business improve responsibly, and what must be true before a pilot begins? Buying access to a model does not settle the quality of the source information, the consequences of mistakes, or the person responsible for reviewing the result.
Begin with a workflow the team can describe and measure. Document the input, the current method, the output, the user, and the next decision. If those details are unclear, use the assessment to establish them before comparing model features or asking for a broad automation proposal.
Measure business value
McKinsey's August 2026 survey of 1,719 participants reported that 80% saw improved individual productivity from AI, while 37% attributed some enterprise-level EBIT impact to it. These are self-reported, different indicators—not sequential stages of a conversion funnel. McKinsey research They reinforce the need to evaluate business outcomes separately from how useful a tool feels to an individual.
For your pilot, define the intended benefit in operational terms. It might be less time assembling a draft, more consistent classification, or faster retrieval of approved information. Decide how review effort, corrections, escalations, and tool costs will be included before describing the result as a saving.
McKinsey global AI survey · 2026
Individual usefulness and business impact differ
Check whether the knowledge is usable
Identify the information the workflow requires and who owns it. Is it current, internally consistent, accessible to the intended user, and permitted for this purpose? A model cannot resolve an unresolved business policy reliably by guessing which document should take precedence.
Create a small set of representative questions or tasks with reviewed expected outcomes. Include missing information, contradictory documents, and requests outside the intended scope. These cases help reveal whether the system can recognize uncertainty rather than merely generate convincing answers when the input is easy.
Define the boundaries
Decide which steps the AI may perform, which require approval, and which remain outside the pilot. Drafting a reply is different from sending it. Recommending a classification is different from changing an authoritative record. Make the distinction visible in the interface and enforce it in the system.
Stanford HAI's 2025 AI Index described a gap between advancing capabilities and standardized responsible-AI evaluation. Stanford HAI research Our application is to treat a model demonstration as a starting point for evaluation, not a substitute for it. Test the workflow with the permissions, documents, and failure conditions it will actually encounter.
Build the smallest reviewable pilot
Select a bounded group of users and a manageable task set. Record the baseline, evaluation criteria, ownership, cost limits, and stopping conditions. Keep a manual alternative available while the team learns which cases are suitable and which require another approach.
At review, examine successful outputs and failures together. Ask whether the workflow improved the task after verification and correction time was included. If the evidence is mixed, narrow the use case or improve the underlying information. Expansion should follow demonstrated usefulness within the chosen boundary.
- Choose a task with a clear input and outcome.
- Assign ownership of the source knowledge.
- Build representative evaluation cases before launch.
- Define permitted actions and approval points.
- Measure quality, review effort, cost, and exceptions together.
How much planning is needed?
Not necessarily. A focused assessment can establish the first useful workflow and the controls it needs. Broader planning becomes more valuable as multiple teams, sensitive information, integrations, and shared operating responsibilities enter the picture.
Evidence behind the guidance
Sources & context
Published research informs this article. VanKpa's frameworks and recommendations are practical applications; illustrative data is labeled where used.
- McKinsey — The state of AI in 2026: On the road to ROI ↗2026-08-25
Self-reported global survey: 1,719 participants in 97 nations, fielded May 4–June 8, 2026. Different indicators are not stages in one funnel.
- Stanford HAI — 2025 AI Index Report ↗2025
Historical reference on AI technical progress and responsible-AI evaluation gaps, not a current adoption estimate.
What could this change?
Bring the question, the current workflow, and the result you want to improve. We can help define a useful next step.
