AI Readiness Is a Workflow Property, Not a Company Personality Test

Replace “Are we AI ready?” with a scoped question

The broad question produces broad answers about leadership, data, talent, infrastructure, governance, and culture. Those subjects matter, but an enterprise average can conceal the decision a sponsor needs to make. One team may have approved content, a visible workflow, ownership, and a baseline; another may rely on paper records and person-dependent decisions. One corporate grade cannot describe both.

A useful assessment asks whether defined users can improve a named decision inside a named workflow, using identified data and systems, within an agreed risk boundary, and prove the outcome against a baseline. NIST's risk framework is contextual, while ISO/IEC 42001 provides organization-level management-system guidance. Organization-wide governance creates the environment; workflow-level evidence supports the specific decision.

• Which workflow and current pain are in scope?
• Which decision, task, or handoff should improve?
• Who performs the work and who owns the outcome?
• What data, systems, and permissions are required?
• What consequence follows if the system is wrong, stale, unavailable, or misused?
• What baseline and evidence will support the next decision?

Assess seven connected dimensions

The dimensions are connected. Accessible data creates little value if nobody trusts its meaning. Willing users cannot rescue an undefined objective. Strong policy does not compensate for the absence of a process owner.

Problem readiness asks whether the pain is material, specific, and owned. Process readiness asks whether the current work, exceptions, handoffs, and decision rights are visible enough to redesign. Data readiness asks whether the required information is available, understandable, timely, representative, permissioned, and maintained. Technology readiness covers integration, identity, access, performance, security, and operating support.

People and adoption readiness examines whether the users see the need, have a meaningful role, can judge the output, and receive manager support, training, and safe escalation. Governance readiness addresses purpose, accountability, privacy, security, impact, vendor, human oversight, incident, and recordkeeping controls proportionate to consequence. Measurement readiness asks whether the team has a credible baseline, success criteria, evaluation set, cost model, and decision gate.

Score evidence, not confidence

One-to-five self-ratings can reflect optimism or familiarity rather than operating conditions. Record the claim, evidence, owner, date, confidence, and action needed to close the gap. Unknown is a valid result.

“We have maintenance data” is only a claim until a sample shows relevant asset coverage, joinable service and failure records, aligned timestamps, understood gaps, access, and permission. Use four evidence states: documented and tested; documented but untested here; partially evidenced or dependent on unresolved work; unknown or absent. The next action may be a data sample, not a model build.

• Record the exact scope to which the evidence applies.
• Link or name the source rather than relying on workshop memory.
• Give every gap an owner and review date.
• Separate a fact from an assumption and an assumption from a preference.
• Recheck evidence after process, vendor, model, data, or policy changes.

Use readiness to make one of three decisions

The assessment should produce a decision for the scoped workflow. Prepare means foundations are too weak for a responsible pilot. Pilot selectively means a bounded test is feasible with controls and unresolved assumptions turned into evaluation tasks. Scale with governance means the intervention has evidence across quality, use, operating outcome, risk, and economics and can be supported beyond the pilot.

These are decision states, not prestige levels. Prepare may be right for a high-value opportunity; scale is not permanent approval and requires monitoring. The assessment may also conclude that standard work, data capture, search, reporting, or rules automation is better than AI.

How to run a readiness assessment that changes a decision

Begin with preparation, not a large survey. Name the sponsor, workflow, users, desired decision, access boundaries, confidentiality requirements, and what decision the assessment must enable. Ask for a small evidence pack in advance: process documents, representative records, volumes, performance measures, system list, known incidents, and existing policies where relevant.

Observe the work with the people who do it. Compare the formal process with real handoffs, workarounds, queues, judgment calls, language needs, and exception paths. Review a sample of the actual data rather than accepting a data catalogue as proof of usability. Identify where a human uses tacit knowledge and what would happen if an AI output were plausible but wrong.

Then describe several intervention options, including non-AI options. Assess them against the seven dimensions and record the evidence state. Prioritize one starting point by value, feasibility, time to evidence, user impact, and risk. The assessment should leave behind a current-state workflow, gap register, bounded solution hypothesis, owner map, measurement plan, and next decision—not a catalogue of tools.

• Interview the sponsor, process owner, data/system owner, risk owner, and representative users.
• Walk through normal cases and exceptions using real artifacts.
• Sample data for meaning, quality, access, and permission.
• Define a no-AI or lower-complexity alternative.
• Freeze baseline gaps and success criteria before building.
• Agree a dated decision gate and the evidence required there.

Adoption belongs inside readiness

Adoption cannot be postponed until after a technical pilot. The workflow may change what information people see, which decisions they make, how performance is judged, and where accountability sits. Users need a meaningful role in defining the task, testing representative cases, setting escalation, and interpreting the output.

Microsoft's 2026 Work Trend Index reports associations between organizational conditions and self-reported AI impact in its sample; it does not establish universal causation. The practical implication is limited but useful: individual enthusiasm cannot compensate for work that has not been redesigned and supported.

Readiness evidence should include user pull, manager sponsorship, role clarity, training on verification and exceptions, a non-punitive feedback route, and time to learn. Track edits, overrides, escalations, abandoned use, and operating outcomes alongside logins or task volume.

Pause when the foundations are not defensible

Pause when nobody owns the outcome, the baseline cannot be measured, the process cannot be observed, required data use is not approved, material vendor questions remain unanswered, representative evaluation is impossible, or users cannot challenge the system.

Also pause when the case depends on an unsupported percentage, unrestricted write access is requested before lower-risk value is proven, or launch itself defines success. The next deliverable may be an owner, process map, data sample, risk review, evaluation set, or simpler workflow. End with a narrow, dated statement: ready for what, for whom, with which data and systems, under which controls and unresolved assumptions.