How to Choose the First Workflow to Improve with AI

Start with work that would be worth improving for the business. Check whether AI can help, whether the information it needs is available, who can change the work, and how results will be tested. Consider what happens if it makes a mistake and where a person’s judgement is needed. A simpler change may be the better first project.

That is the whole decision, in one place. The rest of this piece is how to make it with a clear head when the team is already trying several tools at once.

Why “start anywhere” rarely helps

When AI interest is high and the first project is vague, teams often start in three places at once. Someone tries a new assistant on client email. Someone else builds a private research habit. A third person asks for a tool rollout. Each effort can be useful on its own. Together they do not answer whether the business is better at a specific kind of work.

Without a chosen workflow, there is no shared outcome to check. Without an owner, there is no one responsible for deciding if the change should stick. The result looks busy and still leaves delivery effort high.

Useful individual practice can sit next to that noise. Someone already drafts faster or prepares better. That progress is real. It still does not answer which shared workflow should become the first team bet.

A first project is not a tour of what AI can do. It is a bounded bet on improving work you already do.

What this article is choosing

This piece is for a first improvement to existing work: a recurring delivery task, an internal handoff, a report the team already produces, a review step that eats hours.

New products and exploratory work sit in Mode Lab’s offer. They need a different evaluation. Do not force them through the scorecard below. If what you want is a new service or a prototype of something you do not yet sell, say so and use a design path built for that. The criteria here are for improving work that already exists.

Selection criteria

Use these six questions on every candidate. Write short reasons, not only scores.

1. Business value and recurrence – Does this work matter enough, often enough, that improving it would be felt by the business or its clients?
2. AI suitability – Is there a clear place where AI can draft, summarize, structure, or check against known inputs without pretending to own the result?
3. Available information – Can the team point to the sources, examples, and constraints the work needs, or is that still locked in one person’s head?
4. Ownership and authority – Is there someone who can authorize a change to how this work gets done, and who will be responsible for the result?
5. Testability – After a few real cycles, can you tell whether the new way is better, worse, or only different?
6. Consequences of mistakes and human judgement – What happens if the output is wrong, and where must a person stay in the loop?

More judgement-intensive work is not automatically a better first choice. High judgement with unclear ownership, thin information, or hard-to-check results is often a later project. A simpler workflow with clear inputs and a named owner is often the stronger first bet.

NIST’s AI risk framework, Map section offers broader guidance on context, benefits, alternatives and consequences. This scorecard is our practical method.

A qualitative scorecard

Compare two or three candidates side by side. Keep the language plain. If information, ownership, or risk management is missing, resolve that before you rank the candidates.

Not yet means one of these is true: the necessary information is unavailable, ownership is unclear, results cannot be checked, or the consequences of mistakes cannot be managed. Fix the gap or pick another workflow. Do not paper over it with a tool.

A practical flow for the decision:

Idea → score with reasons → pick one ready bet (or name the gap) → design the work → test on real cycles → decide whether it sticks.

Read the diagram as text

Compare existing workflows using six criteria. If information, ownership, checks and manageable consequences are in place, choose one bounded bet and test it. Otherwise name and resolve the gap or choose another workflow.

Strong vs weak first bets (hypothetical)

The following shortlist is a labeled hypothetical, not a client result.

Imagine a specialist firm of about fifteen people. Delivery effort is high. Three ideas sit on the whiteboard.

Candidate A – Weekly client status updates. Recurring. Partners already care about quality. Notes live in a shared tracker. A project lead can authorize the change and review before send. AI can draft from agreed inputs. Mistakes are visible and fixable before the client sees them. Strong first bet.

Candidate B – Full proposal strategy for new categories. High business value, but each proposal is different, source material is uneven, and partners disagree on what “good” means. Judgement is heavy and hard to check after one cycle. Later, after the firm can describe the work more clearly.

Candidate C – Auto-filing every inbound email into the CRM. Sounds efficient. Ownership of data quality is unclear. Errors are quiet and expensive. The team cannot easily test whether filing got better. Not yet, until someone owns CRM accuracy and a check exists.

The useful lesson is not that status updates are always best. It is that recurrence, available information, ownership, and testability often beat a glamorous, high-judgement first project.

What needs a different approach

If the idea is a new service that combines your expertise with AI, or an exploratory prototype for something you do not yet offer, do not pretend it is “the first workflow to improve.” It is a different kind of bet. Mode Lab still works on that kind of work. It simply should not be scored as if it were an existing delivery task.

Also defer implementation when information, ownership, or adequate checks are missing. Choosing a workflow is not the same as being ready to build.

Why interesting is not enough

An interesting AI use case can still be a weak first project. High-visibility, judgement-heavy work needs adequate information, clear ownership, and a way to check whether the change helps. Ambition alone does not establish those conditions.

For a first improvement, look for recurring, checkable work with someone who can change how it gets done. Place judgement where a person can actually review. That is why the scorecard above puts testability and ownership next to business value.

After you choose

Once you have one first bet, the next job is to make the work clear enough to share or build: what people need to know, who does what, where AI helps, where human judgement matters, how quality is checked, and who keeps the approach useful. That design work is the subject of the companion piece on what to design before you build.

If people on the team already have useful individual practices for that workflow, you will also want a shared way of running it, not only a private prompt.

When an Assessment helps

If you cannot get two or three candidates onto the scorecard, or every candidate fails “not yet,” an Assessment can help you decide where AI could make a worthwhile difference and what to do first. Use it when you need that decision. It is not a compulsory gate before every project, and it is not a priced package described here.

Mode Lab helps teams improve how their business works – and build new capabilities – with AI. We bring your team’s expertise and AI together to rethink how the work gets done, and help you put it into practice.

If you want to build capabilities yourself through learning and practice, Mighty AI Lab is a separate learning and community pathway for individuals and teams.

Start with one bounded bet

List two or three workflows you already run. Score them with the six criteria. Pick one ready first bet, or name the gap that keeps the others at “not yet.”

When you want help choosing or scoping that first improvement, start a conversation. Bring your top two candidates, or say you cannot choose yet and want help deciding.