How to get from an AI demo to an enterprise customer: The 95% Problem

There is broad agreement that AI will change how enterprises work. The cost of building has fallen, and a small team can now put together something that would once have taken months. And yet our conversations with early teams have often swung between euphoria and deep pessimism. Which do not help the founders either. Because our reactions don’t help a founder much in figuring out what to build, how to get it bought, or why a customer will keep paying.

That is where we wanted to spend time. Especially when selling to Indian enterprises, where the economics of cheap labour and the weight of existing relationships have more weight than better technology and easier implementations.

But look at it from their perspective. A voice agent can make an impressive call in a demo. Great, but an enterprise needs it to work through messy calls, fit into an existing workflow, and then improve a number someone is responsible for. That’s difficult to do, and that’s the gap.

Our field note, The 95% Problem, works through this gap in three parts. It draws on conversations with founders and CIOs, and on what we have learned from the go-to-market journeys of Hunar, Bluecopa, and others.

What does it take to get through the adoption barrier?

Adoption requires roughly 95% reliability on some outcome that is important to the enterprise you are selling to. Take frontline hiring. The buyer cares about filling roles on time, not how well an agent talks. AI might screen applicants and follow up on joining documents, while people still make the hiring decision. That bar has to cleared, that understanding has to be there. If that workflow keeps getting more reliable, then you have a play.

What are customers paying for, and how?

Proving value and pricing it are different questions. In hiring, charging per hire takes credit for sourcing and judgment that the product may not actually control. Charging per call minute reduces the metric to airtime. The better unit is something the buyer already counts, can attribute to the product’s work, and can pay for sustainably, such as a lead. 

Direct sales, integrators and global capability centres also offer different routes into an enterprise. Plus Partners can bring trust and reach. Even then, a founder still needs a direct sales motion to retain control over pricing and the customer relationship.

Does India buy software?

We also question the familiar view that India does not buy software. Often, hiring another employee simply offered better economics. AI changes that calculation when it does the work rather than only helping someone record it. But that also brings competition. An enterprise may prefer a services firm rather than trusting a product, and some might start building it for themselves. So execution matters more than choosing where to play.

We also look beyond India, to the Gulf and Southeast Asia. Our view is that the adoption bar travels fine, but the relationships that open the door differ.

This isn’t the final word

We would love your views, especially those that push back at our thesis. At Blume, we want to keep backing teams doing the patient work of building world-class AI and software businesses that customers can rely on. 

Please write to us at yash@​blume.​vc and arpit@​blume.​vc.