AI readiness is less a question of budget and more a question of whether the process you want to automate is already documented, has accessible data, and has someone on your side who will own it. The checklist below lets you check that before you pay for a first prototype.

Is your process ready — check these off

  • The process is written down step by step — even informally, on a page or talked through out loud. It doesn't need to be formal documentation, but "we know how we do it, it's in our heads" without any write-up isn't enough for an outside team to design against.
  • You have access to representative data — examples from the last few months, covering typical cases as well as harder ones, not just the easiest slice.
  • You know the volume — roughly how many times a day or week the process actually happens. That's what determines whether automating it is even worth the cost.
  • You have an accountable owner — someone on your team who will oversee the deployed solution once the implementation team's engagement ends.
  • The systems the process runs through have an API or another way to integrate — not just a click-through interface that's hard to pull data from programmatically.
  • You're clear on error handling — who intervenes when the agent gets something wrong, and what the escalation path to a human looks like.

Signals that it's too early to start

Honest red flags matter as much as the checklist itself — here's when it's better to wait:

  • The process changes weekly. If the way of working is still being settled, automating it now just locks in something you'll soon need to redesign anyway.
  • The volume is too low to justify the cost. A task performed a few times a month rarely earns its keep as an agent-driven automation — a simpler fix (a template, a checklist) may be enough.
  • Nobody on your side will take ownership after launch. A solution with no owner tends to stop working correctly within a few months, because nobody reacts as the surrounding context changes.
  • Data lives in emails and paper notes with no way to export it — that's a problem to solve before, not during, an AI implementation.
  • Expecting 100% automation with zero tolerance for human oversight. Every agentic system makes mistakes; if the organization won't accept any margin for review, that's a sign the process isn't ready for this approach yet.

Who in the company should fill out this checklist

This checklist works best when answered by whoever actually runs or directly oversees the process day to day — not leadership and not IT in isolation, but someone in between: understanding the business purpose of the process and having real visibility into how the surrounding systems are actually used. It's common for leadership to rate readiness optimistically because they see the process from a distance, while the person running it daily sees dozens of exceptions nobody above them knows about. It's worth getting those two perspectives in the same room before the first call with an implementation team — otherwise the checklist effectively gets filled out twice, with two different answers.

A good test is asking that person to describe the last three unusual cases they had to resolve by hand. If they can list them easily, the process is probably understood well enough to start a conversation about automation. If the answer is "that almost never happens, but when it does I flag my manager and we figure it out" — that's a sign the process isn't stable enough yet to be a good fit for an agent.

What to do if you don't check every box

Missing part of the checklist doesn't automatically mean "no" — it means starting with a smaller step first. Sometimes that's writing the process down before the first call with an implementation team. Sometimes it's naming an owner before the budget conversation happens. A readiness conversation is part of the first stage of any well-run project — not something a client has to figure out alone before reaching out.

Does "we have it in a spreadsheet" count as ready

A common assumption is that if data exists in any digital form, the company is ready. That's only half the answer. A spreadsheet where every person records things slightly differently — different column names, different date formats, statuses typed in by hand — can be harder to work with than a paper notebook, because it looks like organized data when in practice it isn't. Before anyone starts designing an agent around a source like that, it's worth checking whether the data can be exported in a consistent format, and whether two different people looking at the same row would read it the same way. If not, the first step isn't an AI implementation — it's cleaning up how the data gets recorded in the first place, which is often enough of an improvement on its own to make the process run better even without automation.

What comes next

If most of the checklist is already true for you, the natural next step is a prototype — a scoped test of the idea on real data. We cover it in What an AI Proof of Concept Looks Like Before You Commit to a Full Build. If you're wondering whether you need an outside team for that, or a freelancer will do, read AI Implementation Studio vs. Freelancer vs. Agency. More on our approach on the Orkestra Labs homepage.