An unreasonable amount of domain detail
You know the workflow better than the people who own it. The specifics arrive without being asked for.
Thesis
The useful layer is usually the boring one.
Software that becomes infrastructure earns its margin slowly and then all at once. We spend our time one level below the headline, where the work is harder to copy.
The bet
Depth compounds. Novelty does not.
Every cycle produces a wave of companies built on a temporary advantage: a model release, a platform loophole, a regulatory gap. Most of them are gone in four years.
The ones that last build something a customer cannot easily unplug. A ledger of record. A data pipeline the operations team depends on. A device already installed in ten thousand sites. That takes longer to sell and much longer to displace.
So we underwrite for durability rather than early velocity. We would rather back a company that takes eighteen months to find its first real customer and then keeps that customer for a decade.
Where we invest
Five areas we know well enough to be useful in.
Infrastructure software
Compute, data movement, storage, and the tooling around them. We like companies whose customers are engineers, because engineers tell you the truth about your product quickly.
Applied AI
Products where the model is a component, not the pitch. The defensibility sits in the data, the evaluation loop, and the workflow you replaced.
Financial rails
Payments, accounting, and treasury for businesses operating across borders. Slow to build, heavily regulated, and very hard to rip out once installed.
Vertical software
Tools for industries that still run on spreadsheets, WhatsApp, and paper. Unglamorous, sticky, and usually underpriced.
Operations in regulated markets
Health, logistics, and compliance. Software that moves paperwork, evidence, and money between institutions.
Terms
What a check from us looks like.
- Stage
- Pre-seed and seed. We are often the first institutional money, and sometimes the first money at all.
- First cheque
- Sized to the round rather than a fixed band. We would rather fit the company than defend a template.
- Ownership
- Meaningful but not greedy. We have never lost a company we wanted over a couple of points.
- Reserves
- We keep room for the next round and use it where it matters, including the flat ones.
- Board seats
- We take one when we lead. We give it up cheerfully at Series B when a better operator should hold it.
- Geography
- India first. We back teams here who sell to customers anywhere, and we show up in person often enough to be useful.
Signals
What tends to make us lean in.
A wedge that already works
One narrow use case, one buyer, and evidence that they came back. We can help with the second product. The first one has to be yours.
A hiring bar that is already set
Founders who have turned down good candidates for the right reasons usually build teams that survive scale.
Honest numbers
A founder who volunteers the churn figure before we find it earns more credibility than one with a better chart.
A reason it is possible now
Not a trend. A concrete change in cost, regulation, or capability that opened a door in the last two years.
A customer who would fight to keep it
The strongest reference call is the one where the buyer sounds slightly worried we might not fund you.
What we pass on
Honest limits, so you do not waste a week.
We are not the right firm for consumer social, gaming, crypto trading, or anything whose growth depends on a single distribution channel we cannot underwrite. Not a judgement on the category. We simply do not have the pattern recognition to be useful, and a passive check from a firm with no opinion is the most expensive money you can take.
We also pass when a round is priced past what we can defend, when the cap table has no room left for the team, and when a founder needs an investor who will agree with them. There are firms that do that well. We are not one of them.
Keep reading
Think you fit the thesis?
Send the one paragraph version. If we are wrong about a category, a good company is how we find out.