A useful first AI project starts with a job your team wants help doing. It might be preparing a handover note, finding an answer in company documents or drafting a response from known facts. Pick a specific job and you can explain what success looks like, who benefits and what needs to happen next.

The challenge is choosing between several promising ideas. A short discussion with the people doing the work can give you a much clearer starting point than a list of AI products.

Northloom builds AI products and advises businesses on private AI deployments. Our planning work connects a defined workload with practical model and infrastructure choices. This guide helps you describe that workload before making those decisions.

Put three real jobs on the table

Imagine a small software company considering three uses for AI:

  • Help staff find answers in product guides.
  • Turn support conversations into notes for the next person handling a case.
  • Draft customer replies using the conversation and relevant documentation.

This is a fictional worked example. Each idea could be useful, but each needs different information and a different way of checking the result.

Ask the team to describe a recent instance of each job. Who did it? Where did they look for information? What did they produce, and who used it? Find out how often the job comes up and where the effort goes.

For the handover note, the difficulty might be reading a long conversation to find what has already been tried. For product search, it might be working out which guide is current. Those details help identify the part AI should assist with.

Compare the ideas with five questions

Use the same questions for each candidate. A few sentences under each heading are enough for the first discussion.

Does the job happen often enough to matter? Look for recurring effort or an important delay. Ask what would improve if the team could do this part of the job more easily.

Is the information available? Identify the documents, conversations or records the system would need. Establish who can authorise their use and whether the people using the result should have access to them.

Can someone check the result? Name the person who can recognise a useful answer and explain how they would check it. Reviewing a handover note against one conversation gives the first project a clear review point.

Will it fit the way people work? Describe where the result will appear and what the person does with it. Include the effort of moving information between systems and making corrections.

Who will own it? Choose someone who understands the job and can make decisions about examples, feedback and changes. They need time to work with the people trying it.

These questions give you a practical comparison. For our example company, the discussion produces this shortlist:

  • Product answers: useful across the team, but someone first needs to agree which guides are current and which staff can read them.
  • Handover notes: a recurring job with a conversation to summarise, a known reader and an existing place to save the result.
  • Customer reply drafts: useful potential, with a reviewer who must also check advice and any commitments made to the customer.

The team chooses handover notes because it can describe the job clearly and organise a useful first version.

Choose the first useful part

The first version takes one support conversation and produces a draft internal note containing the customer’s question, steps already tried, current position and next action. A support engineer checks it against the conversation, edits it and saves it in the case record.

That gives everyone a shared picture of the project. The next engineer receives the information they need to continue the case. The reviewer knows what to look for. The person building it knows the expected input and output.

Keep the first scope clear. Here, sending customer replies remains a separate project decision. The team can improve the handover workflow and learn from its use before adding another responsibility.

Write a one-page brief

Capture the decision in a document the team can discuss with an engineer or supplier. For this example, it could read:

  • Current job: support engineers read a conversation and write a handover before passing the case to a colleague.
  • Intended users: the engineers preparing and receiving the handover.
  • Input: the conversation for one case, supplied through an approved process.
  • Result: a draft note with the question, steps tried, current position and next action. It should make missing information clear.
  • Reviewer: the engineer handing over the case, who checks the note before saving it.
  • Information access: use the selected case within the team’s agreed access and data-handling arrangements.
  • First scope: prepare an internal note and support review and correction in the existing workflow.
  • Owner: the support lead, responsible for examples, staff feedback and the decision about broader use.
  • Useful outcome: the next engineer can continue the case, and preparing and checking the note takes less effort than writing it from scratch.

Add any practical constraints: where the information may be processed, how long people can wait and what budget is available. A clear brief gives technical discussions something concrete to resolve.

Take the brief to your first project meeting

Bring the person doing the job, the person receiving the result and whoever can make decisions about the information and systems involved.

Before the meeting ends, agree:

  • The one job and part of it you will start with.
  • Who owns the project and checks the output.
  • Which examples can be used through an approved process.
  • Where the result goes and how staff will give feedback.
  • Who will investigate the technical approach and report back.

With those choices made, you can move on to testing a model against the workload. If a private deployment is the right direction, the brief also provides the starting point for planning its hardware.

Northloom’s private AI planning assessment helps turn an agreed workload into recommendations on models, hardware and capacity. Arriving with a clear job, owner and desired result makes that next step productive.

Further reading

Google PAIR’s User Needs + Defining Success guide explores understanding existing workflows and choosing how AI can assist the people doing them.