AI Readiness: 5 Questions to Ask Before Automating Anything

I get a lot of inbound that starts the same way: "We need an AI agent for [thing]." Sometimes they're right. Often the thing they want to automate isn't the thing that's actually costing them hours — and sometimes the process underneath isn't ready for an agent to touch it.

Before I scope anything, I walk people through five questions. They're not a sales filter. They're the cheapest way to find out whether automation will pay off or quietly rot. If you're thinking about automating something in your team, run these on yourself first.

1. Is the process actually repeatable?

This is the one most teams skip. They feel the pain of a task being manual, so they assume it must be a process. But a process that runs differently every time is a series of one-offs, and one-offs are what agents are worst at.

The test: could you write it down as a checklist that a competent new hire could follow without asking you a question? If yes, it's automatable. If the answer is "it depends on judgement every time," an agent will frustrate you — automate the boring 80% around it and leave the judgement to a person.

If you can't write the process down, an agent can't read your mind.

2. Where does the data live, and is it any good?

Every agent is only as useful as what it can reach. The single biggest predictor of whether an automation project lands well is whether the underlying data is in a known place, in a consistent shape.

• If your team's knowledge is scattered across Notion, Google Drive, Slack threads, and three people's heads, you don't have an automation problem yet — you have a consolidation problem. Solve that first, or scope the agent to the one source you trust.
• If the data is structurally messy (free-text fields where there should be categories, duplicates, stale records), the agent will hallucinate confidently. Cleaning the data is cheaper than building guardrails around dirty data.

3. What's the cost of being wrong?

This determines how much you can trust the agent. An agent that drafts an internal summary you'll read before it goes anywhere can be wrong often and still be useful. An agent that replies directly to a customer on WhatsApp needs to be right almost every time, with a clean path to a human when it isn't.

Be honest about which category you're in. A lot of disappointment comes from treating a low-stakes task like a high-stakes one (over-engineering it) or a high-stakes task like a low-stakes one (shipping it without an off-ramp).

4. How will you know it's working?

If you can't name a number that should move, you won't be able to tell the difference between a successful automation and an expensive toy. "Saves time" isn't a metric. "Cuts the hour the ops lead spends every Monday on the status round-up to ten minutes" is.

Pick the metric before you build. It does two things: it forces you to be specific about the win, and it gives you a clean decision later about whether to expand, maintain, or retire the thing.

5. What happens to the person doing it now?

This one isn't technical, and it's the reason good automations get quietly sabotaged. If the person whose task you're automating doesn't understand it, isn't trained up on the new flow, or feels threatened by it, the automation will stop being used within a month.

The automations that stick are the ones where the person who used to do the manual work becomes the person who owns and refines the automated version. They know the edge cases. They're invested in it being right. Ship with them, not around them.

---

What this looks like in practice

These five questions are the difference between "we built an agent" and "we got hours back every week." The first is a deliverable. The second is an outcome. Clients don't pay for the first — they pay for the second.

If you've got a process in mind and you want a honest read on whether it's worth automating, that's most of what a strategy session is. No audit theatre — a real decision about what pays off.

Book a call and we'll work through your five questions together.