You do not need a list of AI products to find a starting point. You need a clear view of one piece of work: what starts it, what someone does today, and what a useful outcome looks like. Begin with a task your team recognises, rather than a technology you want to try.
Look for work you can describe.
Notice the moments where someone repeats the same preparation. They collect an enquiry, find the context, move information between tools or prepare a reply. A useful candidate has an identifiable beginning and end. ‘Improve operations’ is too broad. ‘Prepare incoming enquiries for review’ gives you something you can examine.
- What event starts this work?
- Where does the information come from?
- Which judgement requires a person?
- Who knows when the task is complete?
Give the workflow a boundary.
Choose one result, one owner and a manageable set of tools. If the first version needs to replace every system, narrow it. An automation can move known information using rules. AI can help interpret or prepare material where a rule alone is insufficient. Neither removes the need to decide what the system is allowed to do.
From an enquiry to something ready for review
Collect
An enquiry arrives through an agreed channel.
Prepare
Relevant context is organised and a draft response is prepared.
Review
A person checks the context and decides what to send.
What a small starting point looks like.
Try the nursery enquiry
Illustrative exercise. A customer writes: “Can I collect 12 lavender plants on Friday? Small pots, please.” The stock sheet says “Lavender, 2L: 18”, last checked on Monday. Today is Thursday. Open each step and compare it with your own answer.
1. What do we know?
The customer wants 12 lavender plants and Friday collection. “Small” is not a confirmed pot size. The sheet lists 18 plants in 2L pots, but Monday’s count does not establish Thursday’s availability. Extract the request exactly; keep stock and customer preference separate.
2. What should the workflow prepare?
Prepare a review note: requested quantity 12; requested collection Friday; pot size needs confirmation; stock record 18 × 2L, checked Monday; fresh count required. Route this to the nursery owner. Do not reserve stock or confirm collection.
3. What could the reply say?
“Thanks for your enquiry about 12 lavender plants for Friday collection. Would 2L pots suit you? I’ll check current availability before confirming.” This is proposed wording for the owner to review, not an automatically sent message.
4. Try the awkward case
Change the enquiry to “Can I get a dozen of the purple ones?” The workflow should flag the plant identity as unclear. It should ask for a name or photo rather than quietly classifying the request as lavender. A draft that sounds helpful but chooses a plant has failed this test.
Take this into your work
Your starting brief: When [trigger], use [approved sources] to prepare [output] for [reviewer]. Stop when [missing or conflicting information]. Never [unapproved action]. Test one clear request, one ambiguous request and one stale source before connecting any send action.
Find the useful first step
Give the AI one piece of work. It will suggest where rules, AI preparation or human judgement could help, and turn that into a small starting brief.
Run it, change one detail, then compare the decision. Use sample text without customer or confidential information.
Test the starting point before you expand.
Write down how you will recognise useful work. That might be less manual preparation, clearer handoffs or fewer missing details. Establish the current situation before comparing it with a new process. Include the awkward cases: incomplete information, a duplicate enquiry or a request that should not be handled automatically.
Start with a useful result your team can recognise. Then decide what should help produce it.
A practical check
Tick these as you work. Checks reset when you leave this page.
Write a one-page starting brief.
Choose one task this week and describe its trigger, inputs, prepared output, reviewer and stop conditions. Walk through a normal case and an awkward one with the person doing the work. If the two of you cannot agree what a useful result looks like, resolve that before choosing software. The brief becomes a shared reference for deciding what to build and what to leave out.
A small first workflow is not the whole strategy. It is a way to learn from something specific. Keep what proves useful, revise what does not, and expand only when the people using it can explain why it earns its place.
Useful to someone you know?
Pass the article along.
