The small first step

AI Agent Pilot: one workflow, 2–3 weeks, $4,500 fixed.

Pick the one workflow that costs your team the most hours. We build an agent that runs it on your real data, with your people using it and an approval step on anything that matters, and measure the result against your own baseline. At the end you have a working agent, the numbers, and a written plan for what comes next, whether or not that is with us.

$4,500fixed fee, or ₹3,50,000 + GST in India
2–3 weeksfrom scope call to measured result
1 workflowup to two systems, an approval step
Yourscode, prompts, accounts and the plan
What the fee buys

Included, and deliberately not included

Included

  • One workflow, scoped in writing in week one, with a named owner on your side
  • Integration with up to two systems you already run (WhatsApp, email, CRM, calendar, sheets, your practice or accounting software)
  • An approval step on any action that writes to a system of record or reaches a customer
  • A simple review console: what the agent did, why, and a button to correct it
  • A baseline measured before go-live and the same measure taken at the end
  • Weekly working build you can use, not a status report
  • Two weeks of monitoring after go-live, with rule changes included
  • Handover: prompts, tool definitions, runbook and the data-flow diagram, all yours

Not included

  • More than two systems, or a system with no API and no export (we will tell you on the call, and propose the sheet-in-the-middle workaround if it fits)
  • A customer-facing UI beyond the channel the workflow already lives in
  • Production SLAs, on-call support or multi-region hosting (available on a monthly retainer after the pilot; see /ai-agents/retainer)
  • Workflows where the agent would make the decision rather than prepare it: clinical, legal, lending or anything with a regulator who expects a named human
  • Model, hosting and messaging costs: usually tens of dollars a month at pilot volume, billed to your own accounts so you see them
Week by week

How the three weeks run

Day 0

Scope call

Thirty minutes, free. We pick the workflow, name the two systems, and agree the measure. If it is not a fit, we say so here.

You get: A one-page scope and a go or no-go.

Week 1

Rules, data and the baseline

We write down the rules the agent must follow, connect the systems read-only, and measure how the workflow runs today. The checkpoint at the end of the week is where we both confirm the build is worth doing.

You get: The rulebook, the data-flow diagram and the baseline number.

Week 2

Build and dry run

The agent runs on real data with its actions held for approval. Your team uses the console, corrects it, and we tighten the rules daily.

You get: A working agent you have already used.

Week 3

Live, measured, handed over

Approval is relaxed where the log earns it. The measure is taken again. We write the handover and the plan for what a full build would cover and cost.

You get: The numbers, the handover, and a written next step.

Is it a fit?

Good pilots, and the ones we turn down

A good pilot

  • The workflow happens at least a few dozen times a week, so a baseline is measurable in two weeks
  • The rules can be written down, even if nobody has yet (booking rules, document checklists, categorisation conventions)
  • The data is reachable: an API, an export, or a sheet someone already maintains
  • Someone on your side owns it and can spend two or three hours a week on it
  • Getting it wrong occasionally is recoverable, because a person approves what matters

Not a pilot

  • The workflow needs a decision only a licensed professional may make
  • The outcome depends on data nobody can get to without a six-week IT ticket
  • The real problem is that the process does not exist yet, not that it is slow
  • You need a guaranteed result on a fixed date for a launch or a regulator

Unsure which workflow? The readiness quiz ranks your candidates in three minutes, and the industry pages show a worked example each: clinics, fitness franchises, coaching institutes, accounting firms, real estate brokerages, dental practices, law firms, hotels and homestays, logistics and freight forwarders.

Afterwards

What happens at the end

You keep the agent, the prompts, the tool code, the console and the runbook, running in your own accounts. Nothing about the handover depends on us staying.

If you want to go further, the pilot fee is credited in full against the next engagement, which is quoted against the plan the pilot produced: typically the "one workflow in production" band ($8,000–$42,000) or the agentic band ($15,000–$60,000) on our pricing page.

If it did not work, the plan says why, which is worth having before anyone spends more.

All the pricing bands, with INR · What the first week looks like, day by day · Book the scope call, with the five pre-call questions · What a finished reference build looks like · What the pilot measures, and how · What keeping it running costs per month · The one-page proposal as a PDF

FAQ

Before you book

Why a fixed fee rather than hourly?

Because a pilot is a bounded question: does this workflow run well on an agent, and what is it worth? A fixed fee against a written scope means you know the number before you start and we carry the risk of estimating badly. Changes are logged and costed, never silently absorbed.

Which workflows make the best pilots?

High-volume, rule-bound, and currently done by a person who would rather be doing something else: WhatsApp booking at a clinic, lead qualification for an institute, document chasing at a CA firm, review-pack assembly for a franchise. Each of those has its own page under AI agents on this site with the before and after.

What if we decide at the week-one checkpoint that it is not a fit?

Then we stop there. You keep the rulebook, the data-flow diagram and the baseline, which are useful whatever you do next, and we agree a fair share of the fee for the week on the call before anything starts. Pilots that should not continue are the ones we most want to catch early.

Who does the work?

Ketan Khamgaonkar, the founder, with the SigmaticAI team. You deal with the person building it. We work from Nagpur with overlap hours for US Eastern and UK time, on Slack or WhatsApp through the day and a weekly call.

Which models and tools do you use?

Claude by default, with the model chosen for the task rather than by habit; Supabase or your own database; WhatsApp Business API, email and your CRM through their APIs; a small React console. You own every account. The how-we-work page lists the stack in full.

Can we run the pilot on a sandbox rather than real data?

You can, and we will if your data protection rules require it, but a pilot on synthetic data proves much less: the edge cases that decide whether an agent is useful only appear in real messages and real records. The approval step is what makes real data safe to use.

What does the pilot cost in INR?

₹3,50,000 fixed, plus GST, for Indian clients. $4,500 for clients billed in USD. Model, hosting and messaging costs are billed to your own accounts and are usually tens of dollars a month at pilot volume.

Next step

Tell us the workflow

The one that costs the most hours, the systems it lives in, and roughly how often it happens. We reply with whether it is a pilot, and what week one would cover.

Or send the details

Name and email are all we need. The workflow description is what makes the reply useful.