Every business has three jobs that could run themselves. The skill is knowing which three.

You already have a shortlist in your head. The thing someone retypes every morning. The question your team answers forty times a month from the same document. The pile that waits for one person to come back from leave.

The hard part was never the technology. It was knowing which of those is worth automating, and being able to prove it worked.

What it costs you

Not the wages of the person doing it. The waiting.

The invoice sits for two days. The ticket comes back on Thursday. A customer waits while somebody rebuilds an answer that already exists in a document.

Waiting never appears on an invoice, and it is usually the biggest cost of the lot.

Why now

Google and the India SME Forum asked 3,249 small and mid-sized businesses about this. Their finding: putting AI into everyday operations could lift profits by 30 to 35 per cent, and add more than $490 billion across the sector.

Those are numbers for businesses your size, in this market. What changed is not that AI became possible. It is that using it on one ordinary back-office job got cheap enough to be worth doing.

What we do differently

We tell you which job first, in writing. Before anything is built, you get a short call about the job you bring us. What can be automated, what it would be worth, and what has to be true first. That answers the question with evidence rather than enthusiasm.

We measure it on your work, not on a sales demo. We take real cases where you already know the right answer, and that becomes the target. You get a score on your own files before it touches anything live. So "three people, three answers" becomes one answer you can check.

A person approves anything expensive. Where being wrong costs money, someone checks the work first, and you can see what is waiting.

Ask it from your phone

Everything we build can be asked questions directly, using the standard the whole industry now agrees on.

You, in a car park: "How many supplier invoices are waiting on someone, and which are over thirty days?"

The system: Fourteen waiting. Four over thirty days. Three with Priya, one nobody has picked up since the 2nd.

You: "Give the unassigned one to Priya, and remind me about the other three tomorrow."

The system: Done. Recorded against your name at 16:42.

You did not open anything. It was still your permission and your record. Asking is just a new door into the same house.

This is not a promise about what we will learn to do. We already publish software that works this way.

browesp runs browser tasks in your own browser, with your own logins. CodraGraph lets an assistant answer questions about a codebase. So does this website.

Orfyn is ours too, and it is live. It runs on your own infrastructure, under your own keys, and does the integration work that decides whether a pilot ever reaches production. Our engineers use it on client work. That is why the systems we build for you arrive ready to be asked questions, rather than having it bolted on later.

How it works

Read. We look at your real cases and read the results by hand, before automating any judgement about them.

Measure. We build the test set and give you an honest starting score.

Build. The system, with a person checking anything that costs money to get wrong.

Watch. Ongoing checks, and the same test set run against every change.

What we guarantee

We agree what is included in writing before the clock starts. What the system will do, what it will not, and how we will both know it worked. That document is what makes a fixed date mean anything, and you get it before you commit to a build.

We give you an accuracy number after the first week, once the test set exists and we have measured something real on your files. Not before. A number quoted before we have seen your data is a guess, and you would be right not to trust it.

Some work has to be right every single time. Tax, statutory filings, payroll sums. There we build the checks and the human approval around the work. We do not hand the decision to a machine. That is a deliberate choice, and we will explain it on the call.