Your Next AI Win Starts With One Annoying Problem
Last week, in “AI;DR is now a thing…”, we looked at automated content, automated comments and the growing habit of pretending to care at scale.
The problem wasn’t that AI could write a sentence. It was that more activity was being mistaken for a better relationship.
This week, let’s take that same question inside the business: when AI makes us busier, faster or more visible, what actually improves?
You probably don’t need another list of AI tools.
You need the same enquiry to stop falling between two people. Or the Friday report to stop eating Thursday. Or the quote that was nearly ready on Tuesday to actually reach the customer before they buy elsewhere.
Those are useful starting points. “We should be doing more with AI” is a much harder instruction to act on.
This week, let’s turn one recurring annoyance into a small test. No grand transformation. No collection of subscriptions. One problem, one owner and something you can check afterwards.
1. Start where the work gets stuck
Pick a workflow that repeats often enough to observe. New enquiries. Preparing quotes. Collecting information for a report. Following up an agreed next step.
Then write what actually goes wrong.
“Our admin is inefficient” could mean almost anything. “New enquiries sit in a shared inbox because each person thinks someone else is answering” gives you something to investigate.
It also stops you reaching for the wrong solution. If nobody owns the handoff, a faster writing tool may simply produce unanswered emails more efficiently. Impressive demo. Same problem.
Before asking which tool to use, ask: where does this piece of work wait, repeat or lose its owner?
2. Get a number you can explain
You don’t need a dashboard to begin. You do need to distinguish a measurement from a feeling.
Look at a small, clearly described sample. Record what happened and how you checked it. If you are estimating, label it as an estimate.
For an enquiry workflow, you might record the time between receiving the enquiry and sending the first useful reply. An automatic acknowledgement doesn’t necessarily count as useful. Decide what counts before measuring.
Keep the comparison fair. Business hours and elapsed hours are different. A complicated enquiry is different from a straightforward one. A missing timestamp is a gap, not permission to invent one.
The point is to establish a starting position you can defend, rather than a number that makes the next slide look good.
3. Change one thing before changing everything
Here is a fictional practice example, not a customer result.
A small team finds enquiries waiting for a first useful reply. Their first test is to name an owner for each enquiry and use a short reply checklist. A person checks the information before anything is sent.
They might use AI to draft a response from approved information. They might discover the ownership change does most of the work without AI. Either finding is useful.
Choose a small sample and a review date. State who can approve the test and what is outside its scope. Don’t give a drafting assistant permission to make commitments just because it writes confidently.
And write down when you would stop. Unsupported information in a reply? Missing approval? A handoff nobody can account for? Define the exception before the exception arrives.
“Let’s try it and see” becomes much more useful when you have agreed what you are trying and what you are looking for.
4. Count the work that moved somewhere else
If drafting gets quicker but checking takes longer, the checking still counts.
So do corrections, rework and dealing with exceptions. A task leaving your desk is not automatically a saving for the business.
At review time, compare the same measure with the baseline. Record what else changed. A quieter week, different customer mix or extra staff member can affect the result.
A better number is a reason to investigate, not proof that AI caused it. A worse number can still teach you where the problem really sits.
Your decision can be simple: stop, refine the test or repeat it with another small sample. Expanding should follow evidence, not enthusiasm.
5. Give yourself a working rhythm
You can use a month for this if the workflow needs time to repeat: observe and establish a baseline, design the test, run it, then review.
You can also move faster. There is no prize for stretching a useful decision across four weeks.
The useful habit is returning to what you said you would check. Put the review date in your calendar. Keep the test brief somewhere you will find it. Name the person who makes the next decision.
That is the part a shiny tool launch rarely demonstrates: coming back after the excitement and asking whether the work actually improved.
Your next action
Before opening another AI tab, finish these three sentences:
The recurring problem is…
The person who owns the next decision is…
The number we will check is…
If you can’t finish them yet, start with observation. That is useful work too.
AI is OK is now available beyond LinkedIn
Read the weekly newsletter and ten past issues free on our website, or subscribe by email. No LinkedIn account needed. The newsletter remains free.
Read and subscribe: https://aiisok.app/case-studies/#newsletter
Past issues: https://aiisok.app/case-studies/#past-issues
Start small. Keep people accountable. Check what actually changed.
Until next time, stay curious, stay scrappy, and remember: AI is OK.
Steve AGI / C51