The 90-Day AI Readiness Sprint: What We Check Before Any Agent Goes Live

This series has spent seven posts making the case for readiness before rollout. This one is the answer to the question that naturally follows: readiness work, specifically, looks like what?

Here’s the sprint, laid out in full rather than kept behind a sales call.

Days 1 to 20: Data foundations

Before anything agentic gets discussed, the customer data underneath it gets audited. Identity resolution accuracy, duplicate rate, consent-field reliability, freshness by source system. This produces a number, not an impression: a baseline you can measure again in six months to know if things actually improved.

Days 21 to 35: Use-case triage

Not every process that could involve an agent should. This phase ranks candidate use cases by two factors: how measurable the outcome is, and how contained the blast radius is if the agent gets it wrong early. The winning first use case is almost never the flashiest one. It’s the one where success and failure are both easy to see clearly.

Days 36 to 55: Governance design

This is where the “human in the center” model gets built for the specific use case, not as a philosophy but as a workflow: who approves what, what the agent is explicitly never allowed to do on its own, what gets logged, and what triggers a human review. This phase produces a document the board can actually read.

Days 56 to 75: Human-in-the-loop design

Governance on paper and governance in the interface are different things. This phase builds the actual approval points into the workflow: the screen a human sees before an action goes live, the escalation path when the agent is uncertain, the audit trail after the fact.

Days 76 to 90: The measured pilot

One use case, live, with the baseline from day 20 as the comparison point and a defined success metric agreed before launch, not invented afterward to justify the spend. Ninety days ends with a real answer: did it work, by the number you said would prove it, or not.

Why the order matters

Every phase above depends on the one before it. Skip data foundations and use-case triage gets built on a false premise. Skip governance design and the pilot has no way to prove it was safe as well as effective. The sprint isn’t slower because it’s cautious. It’s faster than the alternative, which is discovering all five of these gaps live, in front of a customer, after launch.

Where most companies actually are

Somewhere in the middle. Some data work done, no formal use-case triage, governance discussed but not built into the interface. The assessment below places your organization on this timeline honestly, in about the same time it takes to read this post.

Mar Dat runs this exact readiness sprint with enterprise clients. Start with the free AI Readiness Assessment to see where your organization stands on it today.