Notes from work that actually shipped.

Restaurants26 August 20264 min read

We ran two POS systems on the same counter for nine days.

A restaurant owner in Surat was paying for established POS software and was not especially unhappy with it. We thought ours was better. There is no honest way to prove that in a demo — a demo is a controlled environment where the salesperson knows every keystroke and nothing is at stake.

So we proposed something else. Run both systems, at the same counter, on the same live bills, for nine days. Every order punched into theirs and into ours. Same staff, same Saturday rush, same confused customer changing their order after the KOT had gone to the kitchen.

What that actually tests

Feature lists are the wrong comparison. Both systems could produce a GST bill. What we wanted to see was the stuff that never appears in a brochure:

• How many taps it takes to punch a modified order when three people are waiting
• Whether the kitchen display actually keeps up, or drifts a few seconds behind and starts causing arguments
• What happens when the internet dips for forty seconds during peak hour
• Whether a cashier who has never seen your software can use it by day three without asking
• Whether the day's totals reconcile against the cash drawer without someone doing sums on paper

Nine days is roughly the point where the novelty wears off and staff stop being careful. That's the useful part. A system that survives careless use survives.

What it cost us

It's an expensive way to sell. Nine days of double-entry is real friction for the client's staff, and every problem in our software was visible to the owner in real time rather than discovered quietly and patched. Twice we fixed things the same evening they were noticed.

But it converts a sales conversation into an operational one. The owner didn't switch because we argued well. He switched because he had watched both systems handle his own counter and formed his own opinion. There was nothing left for us to persuade him about.

Where it stands now

Those figures come from the production database, read on 25 August 2026. We publish them because the nine-day argument only works if the numbers afterwards are checkable too.

The transferable part: if you're being sold software and the seller won't run it in parallel with what you have, ask why. The answer is usually that they're not confident, or that switching is designed to be irreversible. Neither is a good reason to sign.

Search26 August 20265 min read

AEO, explained for a business that isn't in tech.

Here is the change, without jargon. Ten years ago a customer looking for a façade contractor typed something into Google and got ten blue links. They clicked a few, formed an impression, and called two of them. Your job was to be one of those ten links.

Increasingly, that customer opens an AI assistant instead and types a full question: "who makes architectural metal façades in north India and what are the typical lead times?" They get back a paragraph. Sometimes that paragraph names three companies.

Why this is different from SEO

Search engine optimisation is a competition for position — being higher on a list. Answer engine optimisation is a competition for being quotable. An assistant writing a paragraph needs a specific, verifiable, well-structured fact it can state without hedging. If your website says "we deliver excellence and quality solutions", there is nothing there to quote. If it says the thickness, the lead time, the cities you serve and the year you started, there is.

This is why so much marketing copy is now actively expensive. Adjectives were always weak, but they used to be harmless. Now they're the reason a machine skips you and describes your competitor instead.

What actually helps

• Answer the question in the first sentence. Don't build up to it. Assistants lift the sentence that answers directly.
• Structured data. Schema markup — Organization, LocalBusiness, Service, FAQ — is how you hand a machine facts in a form it can't misread. Every page on this site carries it.
• Specifics over claims. Numbers, dates, cities, measurements, prices where you can. "Founded 2026 in Amritsar" is quotable; "a leading provider" is not.
• Real FAQ sections in the customer's own words, including the awkward questions. The awkward ones are what people actually ask assistants.
• Consistency across the web. Your name, address and phone number identical everywhere — website, Google Business Profile, directories. Contradictions make an assistant hedge, and hedging means it names someone else.

The part people get wrong

AEO is not a separate department from SEO, and it isn't a trick. Everything on that list also makes a human trust your site faster. The difference is that a human will forgive vagueness and dig around; a machine won't. AEO is mostly the discipline of saying true, specific things and structuring them so they can't be misread.

The practical test: open an AI assistant, ask the question your best customer would ask about your category, and read what comes back. If you're not in it, that's your answer. If a competitor is, read what the assistant said about them and ask where it got that.