EnterpriseLLM.ai

Home › Where would this actually pay off?

Use cases

Choosing your first AI project

Updated November 2026 · About 5 minutes to read · Watch the 4-minute video

The short answer

Pick one question people ask often, whose answer is already written down, used inside the company first, with a number you can measure today.

Its job isn't to change the company. It's to show you, cheaply, whether this works with your documents and your people.

Think of it as a new hire's first assignment

You wouldn't give a promising new hire your biggest client on day one. You'd give them something real, with someone checking their work.

Real, frequent, checkable, and safe to get wrong while it learns.

Where the picture stops working: a new hire learns by themselves. The AI improves only when people fix the documents it reads, so someone has to own it.

From twenty ideas to one project

  1. 20+

    1. Collect the questions people really ask

    From help desk tickets, shared inboxes, and whoever gets interrupted most.

  2. 6

    2. Score each with the five-question test

    Keep the ones that score four or five. See the test.

  3. 3

    3. Check the documents

    Can a person find the answers to twenty real questions in your files? If not, neither can the AI.

  4. 1

    4. Name an owner and a checker

    One keeps the documents current. One reviews answers in the first months.

  5. Go

    5. Write down the "before" and the finish line

    How long answers take today, and what result by what date means keep going, fix, or stop.

Start on the bottom rung

The same project can take three shapes. Climb one rung at a time.

  1. Start here

    A helper for one team

    It suggests answers, and experts check every one. Smaller savings, and mistakes get caught.

  2. Next

    Self-service for staff

    Employees ask it directly, with the source on every answer and a "this was wrong" button.

  3. Later

    Answering customers

    The boldest step. What it tells a customer, your company has told them.

A worked example

A made-up company of 800 people scores three ideas. The numbers are illustrations, not data.

  • HR policy questions

    5 of 5

    About 60 a week, 10 minutes each. One handbook, with an owner.

    First project

  • Customer order questions

    3 of 5

    Frequent, but mistakes reach customers, and order status is a live number.

    Later

  • Lessons from past projects

    2 of 5

    Asked rarely, and mostly never written down.

    Not yet

The most exciting idea is rarely the best first project. It's usually the obvious second one.

What goes wrong

  • Choosing by enthusiasm. The boss's favorite wins, whatever it scores.
  • Three pilots at once. They share one set of experts, and none finishes well.
  • Skipping the document check. Months go into discovering the answers were never written down.
  • No finish line. The pilot never ends. It just fades.

Questions to ask your team or vendor

  1. Which ideas did we consider, and why did this one win?
  2. Can a person find the answers in our documents today?
  3. Who owns those documents, and who checks the answers?
  4. What's our "before" number?
  5. What result, by what date, means keep going, fix, or stop?

Other names you may hear

Pilot
A small, time-limited first project with a clear measure of success.
Proof of concept (POC)
A smaller test that something can work at all, often on sample documents.
Baseline
The "before" number.
KPI
Key performance indicator: the one or two numbers you judge the project by.
Test questions, "golden set"
Real questions with known right answers, used to check the AI after every change.
Scope creep
A project slowly growing past what was agreed.

Photo by Mikhail Nilov on Pexels