Choosing your first AI project: a scoring guide for mid-market teams

Ask a leadership team where AI could help and you'll get twenty ideas in ten minutes. The hard part is choosing one. Pick well and you get a measurable win that funds the next project. Pick badly and you get a demo nobody uses.

Score each idea on five questions

Give each candidate a 1 to 5 on each question, then add them up.

1. Volume. How often does this task happen? Daily and repetitive beats occasional and complex.
2. Pain. How much time, money or customer frustration does it cost today? If nobody complains about it, nobody will celebrate fixing it.
3. Data. Does the information needed to do the task already exist in documents, a CRM or a help desk? AI is only as good as what it can read.
4. Tolerance for mistakes. If the AI gets it wrong one time in twenty, what happens? Drafting a reply a person reviews is forgiving. Quoting a price to a customer is not.
5. Measurability. Can you count the before and after? Minutes per ticket, quotes per day, response time.

Good first projects usually look like this

• Drafting first responses to common customer or patient questions for a person to approve.
• Searching and summarizing internal documents, policies or past proposals.
• Turning intake forms, emails or call notes into structured records in your CRM.
• Producing first drafts of proposals, scopes or reports from a template and past examples.

Projects to postpone

• Anything customer-facing that can't be reviewed by a person at first.
• Projects that depend on data you don't have yet, or that lives in someone's head.
• Broad "AI assistant for everything" initiatives without one clear job.

Before you buy anything

Write down the baseline: how long the task takes today and how often it happens. Check whether a tool you already pay for added the feature you need. Then run a small pilot with five to ten real users for four to six weeks and compare against the baseline. If it doesn't beat it, stop and pick the next idea on your list.

If you'd like help scoring your list, a 60-minute AI strategy call is usually enough to pick the first project.