Your AI product has no moat. A 90-day plan to build one.
If a well-funded team could rebuild your product on the same model in a quarter, you do not have a moat. You have a head start. Here is how to turn one into the other.
If your product is a well-designed layer on top of a model anyone can rent, your advantage is time, not ownership. In 90 days you can map what is genuinely proprietary, file on the technical core, secure your data rights and license one outside technology your competitors cannot get.
Every founder we meet this year has the same slide. It says the product is ‘powered by proprietary AI’. Press on it and the proprietary part is usually a prompt, a user interface and a sales team. All three are valuable. None of them is a moat.
The uncomfortable test
Imagine a competitor with ten crore rupees, the same foundation model you use and three good engineers. Give them one quarter. What part of your product could they not rebuild? If the honest answer is ‘not much’, you are not in trouble yet. You are simply early. The next 90 days decide whether you stay that way.
Days 1 to 30: find what is actually yours
List every component of the product and sort it into three piles. Rented: models, cloud, open-source libraries. Built: your code, workflows, integrations. Owned: things the law or a contract lets you keep for yourself. Most teams find the third pile is nearly empty, and that the most interesting technical work is sitting, unprotected, in the second.
Look hardest at the unglamorous engineering. The way you clean sensor data before it reaches the model. The technique that cut inference cost by 40 percent. The method for keeping outputs consistent across languages. These are often patentable in India because they produce a technical effect, which is exactly what examiners look for under Section 3(k).
Days 31 to 60: protect the core
File provisional applications on the two or three technical innovations that would hurt most to lose. A provisional is quick and inexpensive, and it gives you 12 months to decide what is worth a complete filing. If you are a DPIIT-recognised startup, the SIPP scheme covers your facilitator's professional fees.
At the same time, lock down your data. Check your customer contracts: do they actually give you the right to use their data to improve your models? Many do not, and that clause is the first thing an acquirer's lawyers will ask about. Proprietary data you cannot legally use is not proprietary.
Everything you decide not to patent, protect as a trade secret. Restrict access, sign NDAs with every vendor, and put confidentiality and IP assignment clauses in every employment contract. Courts in India protect secrets that were treated as secrets.
Days 61 to 90: borrow someone else's decade
The fastest moat is one somebody else already built. Indian universities and CSIR laboratories hold thousands of patented technologies, many in sensing, materials, energy and diagnostics, that could give an AI product a physical-world edge no competitor can rent. An exclusive licence in your field turns their ten years of research into your position.
By day 90 you should be able to redraw that slide. Not ‘powered by proprietary AI’, but a short list: what you have filed, what you keep secret, what data you have the right to use, and what you license exclusively. That list is your moat. Everything else is a very good product.