Executive summary
"We need clean data first" is the most common AI-blocking sentence inside small and mid-sized businesses. It is partly true and largely an excuse. True: AI applied on top of scattered context produces scattered output. Excuse: the data is never going to be perfect, and waiting for perfection is a permanent way to avoid the harder operating work.
The real reason most teams stall
The blocker is usually not data quality. It is the absence of a clear workflow, an owner, and a decision criterion that the AI is supposed to support. Without those, even perfect data produces directionless output.
The five readiness layers
- Workflow clarity. The workflow AI will support is mapped end to end and named consistently across the team.
- Data visibility. The data the workflow depends on lives somewhere reachable — not perfect, just findable.
- Ownership. A named person is responsible for the operating outcome the AI is meant to improve.
- Tool integration. The AI output lands where the team actually works, not in a separate chat thread or doc.
- Decision accountability. There is a review cadence at which the team decides "keep, change, kill."
Minimum viable AI readiness
For a single use case, MVR means: one mapped workflow, one named owner, one source of truth for the input data, one place where the AI output lives in the operating system, and one weekly review. Five constraints. Two weeks of work. Enough to begin safely.
Quick wins for messy teams
- Meeting summarization that lands as structured notes in the CRM, not as transcripts in a drawer.
- Inbound intake summarization that pre-fills a scoring rubric.
- Weekly account briefs assembled from existing signals — even if those signals are imperfect.
- Decision packets for recurring choices (pricing exceptions, scope changes) so partners stop reading raw threads.
Risks to avoid
- Customer-facing automated outreach before internal readiness exists.
- Workflows where the underlying decision criteria are still being argued.
- Bulk migration projects ("clean all the data first") with no operating use case attached.
- Tooling decisions made before the workflow decision.
The 14-day readiness sprint
- Day 1–2. Pick one operating outcome that matters. State the baseline.
- Day 3–5. Map the workflow that produces that outcome. Eight attributes — trigger, owner, input, decision, handoff, data, output, feedback.
- Day 6–8. Identify what is broken about the workflow that is not an AI problem. Fix it.
- Day 9–11. Define the one AI insertion point that compounds. Pick the model and surface.
- Day 12–14. Ship into the system of record. Establish a weekly review. Read the first signal.
Two weeks. One workflow. One owner. One read-out. That is what readiness looks like in practice — and it is dramatically more useful than another quarter of "we need to fix the data first."