How to choose your first AI workflow

“Where can we use AI?” is a broad question. “What work keeps getting in our way?” gives you a place to start: a task your team recognizes, understands, and wants to improve.

Start with the work everyone hates doing

The best starting points are rarely dramatic. They sound like this:

• copying details from an email into a spreadsheet, then into the CRM
• checking a shared inbox for the same kinds of request, all day
• chasing a status update from three different people
• researching the same background before every call
• routing work from one team to another by hand

If people keep asking, “Why are we still doing this manually?”, you have found a candidate. It is not one big task. It is a hundred small interruptions, and the person doing them has become the integration between your systems.

What makes a good first workflow

A good candidate usually has most of these qualities:

1. A clear trigger. Something arrives: a request, a document, a form, a ticket, a question.
2. Repeated steps. The same sequence happens many times, with variations people can describe.
3. Information that already exists. The context lives in systems you use today, such as a CRM, a help desk, shared drives, or internal databases.
4. A tangible output. The work ends in something you can check: an updated record, a drafted reply, a summary, a routed task.
5. A natural review point. There is an obvious moment where a person should look before anything important happens.
6. An owner. Someone is accountable for the workflow today and will care whether it improves.

Notice what is missing from that list: “the most strategic process in the company”. Your first workflow should be important enough to matter and contained enough to prove. Ambition can come later, once the first result has earned it.

Map the real work before choosing a tool

Before anything is built, write down how the work happens today. Not the process diagram in the handbook, but the real one: who does what, in which tools, where the hand-offs are, and where it gets stuck.

This step often changes the answer. Sometimes the friction is a missing field in a form, an unclear owner, or an approval nobody needs. Those are process problems, and fixing them is cheaper than building anything.

When the map shows repetitive preparation between systems, AI starts to make sense. The useful question is specific: which steps can be prepared automatically, and which need a person’s judgment?

1. 01 / InputA customer asks for an update.
2. 02 / PreparationAn agent checks records and drafts a reply.
3. 03 / ReviewA person approves before sending.

Decide what stays human

A workflow is only ready to automate when its boundaries are clear. We agree four things before anything goes live:

• What can happen automatically. For example, reading approved sources and preparing a draft.
• What needs approval. For example, sending anything to a customer.
• What should be escalated. Uncertain cases, missing information, and exceptions go to a named person.
• What the system must never do. Anything beyond its agreed access is blocked.

Writing these down is not bureaucracy. It is how a team learns to trust the result, and it is what keeps a helpful system from quietly becoming a risky one.

Prove it small, then expand what earns it

Resist the urge to build the biggest version first. Each stage should answer one question before the next begins:

• Prototype: can the model do the job reliably with your real data?
• Pilot: will people use it, and does the workflow actually improve?
• Production: now that it works, what deserves to expand?

At every gate, “rework” and “stop” are valid answers. A pilot that shows the data is not ready has still saved you from an expensive rollout.

Decide how you will judge success before the pilot starts. Choose a measure that fits the work, such as time to first response, the number of manual hand-offs, or how often a draft is accepted without major edits. Otherwise, “it seems useful” becomes the only evidence you have.

Know when the answer is not AI

Not every problem needs AI. Sometimes the right answer is simpler automation, a better process, or “not yet”. A rule-based integration may do the job with less cost and less risk. A missing owner may be the real blocker.

A good first project leaves you with more than a working tool. It leaves your team with a shared way to judge the next idea: understand the work, prove the useful version, and build on what earns it.

If you already have a workflow in mind, show us how it works today. We will help you judge whether it is worth automating, where AI helps, and where it does not. That is the question a Process / ROI Assessment is built to answer.