The problem
Most businesses have a handful of tasks that eat hours every week and require no judgement at all. Reading an enquiry and deciding who should handle it. Copying figures from an email into a spreadsheet. Pulling numbers from invoices. Writing the same status update forty times.
Nobody has automated them because each one is individually too small to justify a project, and collectively nobody has counted them.
Where we start
We start by counting. A process audit that puts a number on each repetitive task, how many hours it takes, what it costs, and how often it goes wrong. Then we automate in order of return, not in order of how interesting it is.
Common first projects: routing enquiries to the right person with the right context attached, extracting data from documents into your system, drafting replies for a human to approve, and generating the weekly report nobody has time to build.
How we work
- Audit. Two to five days observing and measuring. You get a ranked list with hours, cost and difficulty for each candidate.
- Design. We decide where a human stays in the loop. Anything with money or legal exposure keeps an approval step.
- Build and measure. We run it against your real historical cases and report the accuracy before it touches anything live.
- Roll out. Training, a written runbook, and monitoring that alerts a human when behaviour drifts.
Who this is for
Businesses with a small team handling a large volume of repetitive correspondence, documents or data entry. Typically twenty to three hundred staff, where nobody can be spared to do the copying but nobody can afford to stop.
We start by counting
Most businesses have never measured the repetitive work. It is spread across several people, none of whom spend all day on it, so it never appears as a problem worth a project.
The audit puts a number against each task: how many times a week, how long it takes, what that costs, and how often it goes wrong. That list almost always surprises people, and it decides the order of work by return rather than by whoever complained loudest.
What we automate first
The common early wins are consistent across industries. Routing an incoming enquiry to the right person with the relevant history attached. Reading figures out of invoices, purchase orders or delivery notes into your system. Drafting a reply for a human to approve rather than write. Producing the weekly report that somebody currently assembles by hand.
None of these require judgement a person would struggle to explain, which is exactly what makes them safe to automate.
Where humans stay in the loop
Any step that moves money, creates a legal obligation or makes a promise to a customer keeps a human approval. The automation prepares the work and a person confirms it.
That single design decision is what separates automation that survives contact with reality from the kind that gets switched off after an embarrassing mistake.
Accuracy is measured, not asserted
Before anything runs unattended we test it against your real historical cases and report the accuracy honestly. If it cannot reach a level that is safe for the task, we say so and we do not build it.
We also monitor after launch. Model providers change their models, your inputs change, and a process that was accurate in March can drift by September. Alerts go to a person.
Data and privacy
We use business tier APIs where inputs are not retained for training. For sensitive workloads, or where a client requires it, we can run processing entirely within your own infrastructure using open models.
We document exactly what data leaves your systems, where it goes and how long it is kept, so your compliance answer is written down rather than assumed.
What it costs to run
Model usage is usually a small monthly figure, frequently a few thousand rupees, against many hours saved. We estimate running cost during the audit and monitor it afterwards, so there is no surprise on the third invoice.
Where we say no
If the task is rare, if the inputs are wildly inconsistent, or if the judgement involved cannot be articulated by the person currently doing it, automation will disappoint. We would rather tell you during the audit than after the invoice.
