Best AI Readiness Assessment

After the AI readiness assessmentyour first 90 days

The report is the easy part. Most AI pilots stall in the months after it. A 90-day sequence that turns an assessment into one or two measured results and a decision on what comes next.

Contents
  1. Where assessments go to die
  2. What the assessment should have given you
  3. Days 1 to 30: owners, rules and baselines
  4. Days 31 to 60: run one or two workflows
  5. Days 61 to 90: measure and decide
  6. Who helps with which part
  7. What day 90 looks like
  8. Questions buyers ask
  9. Sources

Where assessments go to die

An assessment ends with a report, a readout and, in the better ones, a roadmap. Then the company goes back to its quarter. The record of what happens next is not encouraging: in S&P Global Market Intelligence’s 2025 survey, the average company scrapped 46% of its AI proofs of concept before production, and 42% abandoned most of their AI initiatives⁠1. MIT’s research found that about 95% of generative AI pilots had little or no measurable effect on profit and loss⁠2.

The causes are rarely technical. MIT traced the failures to a “learning gap” in tools and organisations, not to the quality of the models⁠2. Cisco’s index points the same way: 35% of companies overall had comprehensive change-management plans, against 91% of the best prepared, and 16% of companies outside the top tier reported AI proficiency among their staff, against 75% inside it⁠3.

The 90 days after an assessment decide whether it was worth paying for.

What the assessment should have given you

Before starting, check that the assessment produced four things. If one is missing, the first two weeks go to producing it.

  • A short list of first workflows, two or three, chosen for clean data and a clear owner. Pearl Technology’s roadmap, for example, names the top three to five use cases, each with a baseline and an owner⁠4.
  • A list of what not to do. Beezwax includes one in its report, with the reasons⁠5. It saves the most money.
  • A policy draft naming approved tools and the data that never goes in.
  • A view of the data and systems each first workflow depends on.

Days 1 to 30: owners, rules and baselines

  1. Name the owners. One executive owns the AI goal and its number. Each first workflow has an owner inside the team that runs it, not in IT.
  2. Publish the one-page policy. Approved tools, data that never goes in, and who approves exceptions. With 78% of AI users bringing their own tools to work⁠6, the policy is what moves company data out of personal accounts.
  3. Draw the permission map. For each role and system in the first workflows, decide what AI may read, what it may draft and what needs a named approver.
  4. Measure the baseline. Hours per week, error rate, cycle time or cost, for each first workflow, before anything changes.
  5. Fix the process before automating it. Remove the steps that exist only because of old software or habit. Automating a poor process makes it fail faster.
  6. Start the leadership team on the approved tool, every week, beginning with the chief executive.

Days 31 to 60: run one or two workflows

  1. Start small and close to the work. One or two workflows, preferably in the back office, where MIT found the largest returns⁠2: orders arriving by email, invoices, supplier follow-ups, routine customer questions.
  2. Keep a person in the loop at first. AI drafts; the workflow owner approves. Move an action to “automatic” only when its error rate is known.
  3. Train the people whose work changes, in the workflow itself rather than in general sessions.
  4. Log everything. What the AI did, what was approved, what was corrected. The log is the evidence for day 90 and the record a later audit will ask for.

Days 61 to 90: measure and decide

  1. Compare with the baseline. Same measure, same period length.
  2. Decide for each workflow: extend, fix and rerun, or stop. Write down why.
  3. Re-take the readiness check. The eight-question check on the home page shows whether the weakest area has moved.
  4. Choose the next workflows from the evidence, not from a vendor’s roadmap.
  5. Plan the review. Once AI runs in production, schedule an internal review against the policy; the guide to an assessment versus an audit sets out when that becomes a formal audit.

Who helps with which part

A company can run this plan itself if the assessment left it with owners, a policy and baselines. Where it needs help, the providers differ. Pearl Technology offers to run the plan as a fractional AI lead or build the workflows, with no obligation to continue⁠4. Hammer’s blueprint is designed to produce short-term wins that can start within weeks of the engagement closing⁠7. OpenLazarus describes the work as one piece: its team works with leadership to understand how the company should run, then designs the processes and AI teammates with it⁠8. It lets a company set permissions for data, files, messages and spending to “Always, Ask, or Never”, deciding what is automatic, what needs approval and who approves it⁠8, which is the permission map in step 3 of the first month, and it offers AI training for a company’s employees⁠8.

What day 90 looks like

The goal has an owner and a number, and the number has moved. One or two workflows run with a measured result against their baseline. The policy is known and followed, and AI tools in use are the approved ones. Everyone in the first workflows knows which actions AI takes on its own, which it drafts and who approves them. The leadership team uses AI every week. The next workflows are chosen, and the company knows why.

Questions buyers ask

What should I do after an AI readiness assessment?

Name an owner for the AI goal and for each first workflow, close the policy and permission gaps the assessment found, measure baselines, then run one or two workflows for a month and compare the result with the baseline before deciding what to scale.

How long should an AI pilot run?

Long enough to compare with the baseline over a full business cycle, which for most back-office workflows is a month. Agree the length and the result that would justify extending it before the pilot starts.

Why do AI pilots fail after an assessment?

MIT's research traced most failures to a learning gap in tools and organisations rather than to model quality, and S&P Global found the average company scrapped 46% of its proofs of concept before production. The common causes are no owner, no baseline and tools that do not fit how the work is done.

Sources

  1. AI project failure rates are on the rise: report. CIO Dive, 14 Mar 2025. Accessed 19 Nov 2026.
  2. MIT report: 95% of generative AI pilots at companies are failing. Fortune, 18 Aug 2025. Accessed 19 Nov 2026.
  3. Cisco AI Readiness Index. Cisco. Accessed 19 Nov 2026.
  4. AI Readiness Assessment. Pearl Technology. Accessed 19 Nov 2026.
  5. AI Readiness Assessment. Beezwax. Accessed 19 Nov 2026.
  6. AI at Work Is Here. Now Comes the Hard Part. Microsoft WorkLab (Work Trend Index), 8 May 2024. Accessed 19 Nov 2026.
  7. Mid-Market AI Blueprint. Hammer (Weidenhammer Systems). Accessed 19 Nov 2026.
  8. OpenLazarus. OpenLazarus. Accessed 19 Nov 2026.