Almost nobody in this industry publishes prices, which is how a market ends up with a five lakh quote and a fifty lakh quote for the same piece of work. Here are ours, in rupees, along with the three costs that tend not to appear in anyone's proposal until later.
The three shapes this comes in
Nearly every AI purchase an Indian business makes falls into one of three shapes, and they are separated by an order of magnitude each. Knowing which one you need saves more money than negotiating within one.
A website assistant: ₹0 to about ₹6,000 a year
This answers questions on your public content for people visiting your site. It is a product, not a project. You install it, point it at your pages, and it works the same day.
Our version, Berry, is free on the Seed plan and runs ₹499 to ₹4,999 a month on annual billing depending on volume. If someone quotes you two lakh to build a chatbot for your website, this is the shape they should have sold you instead.
A private AI on your own data: ₹2.5 lakh to ₹7 lakh, plus a care plan
This is the one most people mean when they say they want AI. It reads your internal material, your ledger, your quotations, your drawings, your reports, and answers questions about your business with the source attached.
Melon starts at ₹2.5 lakh for one team and one core problem, ₹4.5 lakh for a whole-company assistant, and ₹7 lakh for a multi-entity or on-premise build. Each carries a monthly care plan from ₹25,000, which is not a maintenance fee for the sake of one, and we will come back to why.
The important thing about this band: it is a project with a defined scope, delivered in about eight weeks, and you should see it working on your own data in week one. If you are being quoted this without a demo on your real data first, ask why.
Bespoke development: it depends, honestly
Custom AI means software that did not exist before, built for how your company works. Pricing starts around ₹2.5 lakh but the number is genuinely a function of scope, and anyone giving you a firm figure before understanding the workflow is guessing.
What we would say is this: most companies who think they need this need the ₹2.5 lakh version of the previous shape first. Solve one painful question completely, then decide.
The largest unbudgeted cost in every AI project is the time your own people spend explaining the business to it.
The three costs nobody quotes you
1. Getting the data ready
This is the big one, and it is where projects overrun. Your data is spread across an ERP, a shared drive, someone's laptop and fourteen years of email. Connecting to it is engineering. Making sense of three spellings of every customer name is engineering. Deciding what counts as the current revision of a drawing is a business decision that will take three meetings.
A vendor who has priced this at zero has either not looked at your data or is planning to have the conversation after you have signed. Ask directly: what is included for data preparation, and what happens if it turns out to be worse than we thought.
2. Running it
The model itself costs money per question. For a normal business this is smaller than people fear, typically a few thousand rupees a month rather than lakhs, but it is not zero and it scales with use. Hosting, backups and monitoring sit alongside it.
The care plan covers this plus the part people forget: your business changes. You add a product line, change your ledger structure, acquire a company. A system nobody is maintaining degrades quietly, and the first sign is people going back to asking Suresh.
3. The time your own people spend
This is the largest unbudgeted cost in every AI project. Somebody in your company has to explain what a dealer is, which report is the one people actually trust, and why the March numbers look wrong. Realistically that is a few hours a week from someone senior for the length of the build.
If you cannot spare that person, the project will produce something technically complete and commercially useless. That is not a cost you can pay a vendor to avoid.
Three questions for any proposal you receive
- What is included for data preparation, and what happens if the data is worse than we thought?
- What does this cost to run each month once it is live?
- How many hours a week do you need from our people, and from whom?
A proposal that has priced all three at zero has not priced them. It has postponed them.
How to think about whether it is worth it
Skip the productivity-percentage arithmetic. Nobody believes it and it is not how the decision gets made.
Do this instead. Take one question that costs somebody real time every week. What did we quote this customer last year. Which dealers are over sixty days. Which SKU lost margin last quarter. Work out roughly what that costs you today in hours, in delayed decisions, and in the occasional expensive mistake made from memory.
If a ₹2.5 lakh build removes that entirely, most manufacturers and distributors we have worked with find the arithmetic obvious within a quarter. If it does not look obvious, that is a real answer, and the right move is to not do it yet.
On the word affordable
We are wary of it, and it is worth saying why on a page about price.
Affordable invites you to compare us with the cheapest thing available, and the cheapest thing available is a ₹40,000 chatbot that will be dead in four months. That comparison is bad for you as well as for us, because the money you lose there is not the ₹40,000, it is the two years your company will now spend believing AI does not work for businesses like yours.
What we would rather claim is that the entry point is small and honest. ₹2.5 lakh to solve one real problem, on your data, with a demo before you commit. That is a size a mid-sized Indian company can decide on without a board meeting, which is the actual thing people need. Not cheap. Just not a leap of faith.
If you want to test that, we will run a free evaluation on a sample of your own data under NDA and show you what it can answer before any money changes hands.
