A sensible way into AI for a small team

Try one small thing.

Pick a job where a mistake is cheap, such as a first draft of a reply, a summary of a long document, or a list of ideas. Write down what you hoped for: time saved, better quality, or more ideas. Note where the AI helped and where it did not.

Get your information in order.

AI can only work from what it can see. Decide where your key information lives, remove duplicates and out-of-date copies, and agree who may see what. Tidying is not a one-off, so plan a regular check.

Automate one repetitive job.

Choose something that eats the same hour every week, like a weekly report, answers to questions you always get, or a summary of project updates. Run it with two or three people first, compare it with the old way, then widen it.

Chain jobs together.

Once one job works, connect the next: a new enquiry comes in, the details are recorded, a draft reply is prepared, and a person approves it before it goes. Each added step has its own check and its own sign-off point.

Whatever the step, a person decides.

Whatever the step, a person decides. AI prepares, you approve. Keep a record of what it did, and keep sensitive information to the people who need it.

How we check an AI's work.

Saying 'done' is not proof. Before we let an AI do a job, we write down what a correct result looks like, and we check against that, not against the AI's own opinion.

• A check that can fail. We decide in advance what good looks like, for example the totals match, every required field is filled, and the source is shown. If the AI cannot fail the check, it is not a check.
• Sign-off by how hard it is to undo. A fix that is easy to reverse, like a draft or a tidy-up, can go ahead and be reviewed afterwards. Something that needs cleaning up afterwards gets a second look first. Something that cannot be undone, like an email sent, a record deleted or money moved, always waits for a person to approve it.
• Limits that do not depend on good behaviour. We control what the AI is allowed to see and do through permissions, not through instructions. It keeps a record of what it did, so you can see every step.
• Stop when it is stuck. If the same problem comes back twice, or the time or cost limit is reached, it stops and tells a person, rather than trying again and again.

How we help.

AI Opportunity Snapshot

We map your offer, define the visible gaps worth looking for, then review a focused set of prospects to identify the clearest sales opportunities.

Opportunity Analyst Sprint

We turn your offer into a repeatable AI opportunity model, score prospects against it, and prepare evidence-led recommendations your team can approve before outreach.

Monthly Opportunity Analyst

Ongoing opportunity analysis and contact-history discipline for a managed pipeline.

How to ask AI properly.

Role

Who it should act as.

Format

What shape you want back.

Background

Paste in the facts it needs.

Think it through

Ask it to check its own answer before replying.

Never paste anything you would not email to a stranger.