Your Customers Are Already Telling You What to Automate

SOMEONE ASKED YOUR BUSINESS

Open your WhatsApp Business, your DMs, your inbox, your call log. Go back through the last fifty enquiries and write each question down in plain language. Not summarised. As asked.

STEP 2: COUNT HOW MANY ARE THE SAME QUESTION

You will almost always find something uncomfortable: somewhere between eight and fifteen distinct questions account for the overwhelming majority of everything anyone ever asks you.

Price. Are you open. Do you deliver. Is it in stock. How long does it take. Where are you located.

We call this the 12-question rule, because in practice the number is usually around twelve.

If a dozen questions make up 70% of your inbound volume, and a trained human is typing those answers out by hand every day, you have just found your first automation.

You did not need a strategy document to find it. You needed a tally.

STEP 3: SORT EVERY TASK INTO ONE OF THREE BUCKETS

Now widen it beyond enquiries and look at the work itself.

HIGH VOLUME, LOW JUDGEMENT — answering repeat questions, sorting enquiries, chasing invoices, generating first drafts, formatting reports, tagging photos.

This is where AI earns its keep immediately. Automate here first, always.

HIGH VOLUME, HIGH JUDGEMENT — negotiating a contract, diagnosing an unusual fault, handling an upset client.

Here AI assists but never decides. Give the human better information, faster; let the human make the call.

LOW VOLUME, HIGH JUDGEMENT — hiring, pricing strategy, deciding what to build next.

Leave it alone. Automating something you do four times a year buys you nothing and costs you real money.

Nearly every failed AI project we have been asked to rescue started in the wrong bucket. Somebody automated the interesting problem instead of the frequent one.

STEP 4: MEASURE ONE THING BEFORE YOU BUILD ANYTHING

Pick a single number and write it down today: average response time, hours spent per week on the task, number of enquiries lost without a reply.

If you cannot state the number before, you will never be able to prove the system worked, and you will end up arguing about feelings instead of results.

THE PART NOBODY SAYS OUT LOUD

The hardest part of an AI project is almost never the AI.

It is the plumbing — connecting the systems you already have, cleaning up data nobody has looked at in three years, and agreeing internally on who owns what when the machine hands a conversation back to a human.

That is the actual work.

The model is the easy part.

WHERE TO START THIS WEEK

Do step one.

Fifty questions, twenty minutes, a pen.

If a dozen of them repeat, you already know what to build first — and you know it from your own business, not from a conference stage.

If you would like a second pair of eyes on what came out of that list, message us. We will tell you honestly whether it is worth automating or whether you simply need to fix your routing.

Sometimes the honest answer is the second one.

DeMario Glow builds AI and digital systems for businesses in Ghana and worldwide — https://demario.pro⁠�