The short answer

Start with one repeatable piece of work that has clear inputs and a result someone can check. Preparing a report, drafting a customer response or organising information is usually easier to test than asking AI to run a whole department.

Start with a frustrating task, not a list of tools

For a week, notice the work that makes you think “I did this yesterday”. It could be gathering the same information for each enquiry, preparing an update from a spreadsheet or turning meeting notes into a handover.

Write down what starts the task, what information you use, what you produce and who checks it. That gives you something concrete to improve. “Make this weekly report easier to prepare” is a much more useful starting point than “we need to use more AI”.

Four places worth looking

Look for work where a person spends time interpreting, organising or drafting information. The first version should help them do that job, with the final decision still clear.

  • Customer enquiries: categorise a request and prepare a draft response from approved information.
  • Documents: turn a structured brief into a consistent first draft.
  • Reporting: collect recurring inputs, apply defined calculations and prepare commentary for review.
  • Internal information: help a colleague find an answer in a small set of current, approved documents.

Separate judgement from predictable steps

A workflow may need very little AI. Moving a completed form into a system is often a rules-based task. Interpreting a loosely written enquiry or drafting a useful summary is a different sort of step.

Keeping that distinction makes the result easier to explain. Use fixed rules for required fields, calculations and permissions. Use AI for a clearly bounded interpretation or draft, then decide where somebody needs to review it.

Test the awkward examples as well as the easy ones

Try the proposed workflow with examples from the real task, using appropriate or fictional data. Include an incomplete request, an unusual document and an input that should be rejected. A convincing demonstration on one tidy example tells you very little about everyday use.

Compare the result with the current process: how much work was still needed, what had to be corrected and whether another person could follow it. The aim is a repeatable improvement, not simply an impressive answer.

What to bring to a first conversation

You do not need to know which model or integration to use. Bring an example of the task, a list of the systems involved and an explanation of where it gets frustrating. Avoid sharing confidential customer information at this stage.

At Get AI Working, we use that discussion to assess whether there is a sensible implementation to explore. If an ordinary process change or an existing feature looks more appropriate, that belongs in the conversation too.

Choose one task, define a checkable result and test it with the people who do the work.

Further reading

Use the linked provider documentation to check current product capabilities and access for your particular setup.

Explore the connected topic

Put the idea to work

AI Setup & Integration

You already have the knowledge. It is in your documents, your software and the heads of the people doing the work. We help put the right parts within reach of a useful AI workflow.

How we can help