Home / Resources / AI Readiness Audit

The AI readiness audit: what it actually checks (and what it costs to skip)

Most failed AI projects don't fail on the model. They fail on the six weeks before the model ever ran — the data nobody checked was usable, the workflow nobody mapped, the approval nobody knew was required. An AI readiness audit exists to surface all of that before you commit budget, not after a pilot quietly stalls.

What "readiness" actually means

It's not a maturity score for its own sake. A useful readiness audit answers one operational question per workflow candidate: if we built this today, would it survive contact with your real data, your real approval chain, and your real team? That means checking five things, in this order.

The five things a real audit checks

1. Data accessibility and quality. Not "do you have data" — almost everyone does — but is it in a system with an API or export, is it consistently structured, and is it current. A CRM with three years of untouched duplicate records is not automation-ready until someone decides what "clean" means.
2. Workflow shape. Does the process have a stable, repeatable structure, or does every instance require a judgment call only a specific senior person can make? High-volume, well-defined tasks (quote generation, report compilation, order status queries) automate cleanly. Highly exception-driven tasks need a human-in-the-loop design, not full automation, at least initially.
3. Tooling and integration reality. What can actually talk to what. An ERP with no API and no scheduled export access is a real constraint that changes the design, not a detail to discover mid-build.
4. Governance and approval. Who needs to sign off on an AI system touching customer data, financial figures or external communication, and how long does that actually take in your organization. This is where timelines quietly double if it's skipped upfront.
5. Team capability and ownership. Who owns the system after it ships. An automation with no internal owner degrades within two quarters — inputs drift, edge cases pile up, nobody notices until a customer complains.

What skipping it costs

We've watched the same failure pattern play out at companies that skip straight to buying a tool or briefing a vendor on outcomes without this groundwork:

Each of these is a 2–3 day conversation to catch upfront and a multi-month setback to catch after the fact. That asymmetry is the entire case for doing the audit first.

What a good audit produces

Not a slide deck of generic AI opportunities. A usable audit ends with a short, ranked list of 3–5 candidate automations, each with an honest readiness rating against the five checks above, a rough cost-to-build, and an expected payback window — the same cost-model thinking we walk through in The ₹40-lakh spreadsheet. If nothing on the list clears a sensible payback bar, a good audit says so, rather than manufacturing a project to justify itself.

Who should run it

It can be done internally if someone senior enough has both the technical literacy to judge data/integration reality and the organizational standing to ask governance questions across departments. Most companies don't have that person free for two weeks, which is why this is usually the first engagement with an outside partner — see our executive's guide to AI agents for the governance checklist that follows once you've picked what to build.