AI automation pricing in India

Three stages, one engagement

What each stage buys you

Most engagements run through all three stages in order, but you can stop after any one of them. The audit is designed to be useful even if the answer is “don't build this”.

• Setup Audit — ₹20,000 – ₹40,000. One to two weeks, fixed, and credited against the build if you proceed. We look at your real workflows and tell you which one an agent can carry and which ones it can't. You leave with a scored shortlist, one measurable metric per workflow, a data-and-integration check, and a written go / no-go.
• Agent Build — ₹1,50,000 – ₹4,00,000 per agent. Four to eight weeks. One agent, built and deployed into your live systems, measured against the metric agreed in the audit. It ships with a cost cap, a kill-switch, and live metric instrumentation. This is per agent, not per workflow you might imagine adding later.
• Managed Running — ₹15,000 – ₹40,000 / month. A rolling arrangement with 30 days' notice. Agents drift, costs creep, and edge cases appear, so we monitor, maintain, and keep it cheap to run. The usage underneath is billed at cost, not marked up.

How the 70/30 build fee works

Seventy per cent of the build fee is fixed. The remaining thirty per cent is only earned if the metric we agreed in the audit moves by the agreed threshold within ninety days of go-live. We accept only metrics your systems already count. If measuring the outcome would mean building the measurement first, it is the wrong metric and we say so before you sign.

There is one honest condition on that contingent third. If agreed data access, environment access, or subject-matter time slips past the dates in the statement of work, the contingent portion converts to fixed. Most outcome-pricing disputes come from client-side delay, and pretending otherwise would make the clause dishonest. We cap the contingent portion on purpose — a firm that bets the whole project makes reckless decisions near the deadline.

What moves a build from ₹1.5L toward ₹4L

The range is wide because builds are not the same size. Five things decide where yours lands, and none of them is your budget. We won't quote a price for any single product before the audit, because the number is a property of the work, not of the agent's name.

• Systems to integrate. One clean system with an API is cheap. Four systems, one of them a spreadsheet nobody owns, is where the time goes.
• Workflow exceptions. A workflow that runs the same way every time is a small build. One with a dozen “except when” branches is a large one, because each branch is a decision the agent has to get right.
• Data quality. Clean, consistent data lands at the low end. Data that has to be cleaned, reconciled, or de-duplicated before an agent can trust it pushes the number up.
• Compliance and audit needs. Audit trails, approvals, and record-keeping are real work. Where a wrong answer has a legal or financial consequence, that work is not optional and it is not free.
• Whether the outcome is already measured. If your systems already count the thing that should improve, the build starts lower and the contingent clause is clean. If not, the measurement becomes part of the job.

Usage is separate, and billed at cost

The build fee and the monthly fee do not include what the agent spends while it runs. Model calls, telephony minutes, and messaging are metered by the providers, and those rates move, so we will not print a number here we would have to keep correcting. What we commit to is the arrangement: usage is passed through at cost, reviewed with you against a cost cap, and sits behind a kill-switch you can reach. A system with no ceiling gets switched off by finance, not by engineering, so we build the ceiling in from the first day.

Do not buy this yet

These prices are real, but they are wrong for plenty of good businesses. If any of the following is true, keep the work with a person for now — automation earns its place against a specific bottleneck, not against a category you are told to be in.

• Low or irregular volume. If the work happens rarely or in bursts, a person absorbs it more cheaply than a system you have to build, run, and maintain.
• No metric. If nothing in your systems counts the thing that should improve, the contingent clause is theatre and you will not know whether it worked.
• No named owner. A committee cannot change how work is done. Without one accountable person, you get a tool nobody adopts.
• The relationship is the point. Some work is cheap to automate and still worth keeping human, because a person on the other end is the value. Automating it saves money and loses the thing that made it work.