How to tell whether a workflow is worth changing with AI

Not every repeated task deserves an AI build. The first useful system usually sits at a workflow that is genuinely high-leverage, has decent context already, has a clear point where a human reviews the result, and is the kind of work the team would actually like to be unburdened from. If three of those four are missing, fix the workflow first.

Why this matters

An AI system that lands on top of a confused workflow makes the confusion faster and harder to see. The team ends up reviewing more outputs, with less context, in a tool they did not choose. That is not progress. So before picking a place to build, it is worth running a sober check on the workflow itself.

The four signals

1. Real leverage

Leverage means a measurable business outcome moves when this work moves: revenue, retention, cost per customer, time-to-decision, error rate. If the only honest answer to “what changes if this is faster?” is “we feel less behind”, the workflow may not be the right first target. Look for work where speed, quality, or consistency in this specific step changes a number the business already cares about.

2. Existing context

AI is at its most useful when the context it needs already lives somewhere readable: the CRM, the inbox, the documents folder, the data warehouse, a clearly written internal guide. If the context lives only in two people’s heads, the engagement starts with capturing that context, which is real work in its own right and worth knowing about up front.

3. A clear review point

Useful AI systems give people clearer drafts, sharper routing, faster checks, and better starting decisions. They do not remove judgement where judgement still matters. So the workflow needs a natural place where a person reviews the output before anything irreversible happens: a customer email gets sent, a payment is moved, an offer goes out. If there is no review point and the work is high-stakes, build the review point first.

4. The team would welcome being unburdened

This sounds soft, but it is the single most reliable signal. AI systems that the team wants to use get used. AI systems that the team feels surveilled by, or threatened by, quietly die. If the people doing the work today would treat a useful assistant as a relief, the system has a chance. If they would treat it as a wedge to replace them, the system needs a very different conversation around it.

What to do if three of the four are missing

Fix the workflow first. The first useful intervention is usually not AI; it is a small change to how the work is structured, where the context lives, or who reviews what. Once those pieces are in place, the AI build becomes much easier, and the team is far more likely to adopt it.

What to do if all four are there

Treat it as a real candidate for a first system. The diagnose stage of a Lucumo engagement is built around a check like this, applied across the few workflows in the business that might be worth changing. The result is usually one clear first target, not a backlog of ten.

If you would like a second pair of eyes on whether a specific workflow is worth changing, the intake chat is a good place to describe it in your own words.