How we evaluate: from the first call to the Investment Committee
How SilverX evaluates deep-tech companies, builds conviction and makes investment decisions.
By Prashant Rathee
Most deep-tech pitches sound convincing in the first thirty minutes. The science is real, the problem is large, and the founder is usually the smartest person in the room on their subject. That is exactly why our evaluation is designed to slow judgment down, not speed it up.
Underneath every conversation, diligence call and committee discussion sits one question: can this company become a fund returner?
In plain terms: can this company reach about ₹1,000 Cr in annual revenue within 5–8 years? And is the market big enough for 4–5 players, incumbents included, to each hold at least 10%?
If we can't see a credible path to that, the deal doesn't go further, however impressive the technology.
How a deal moves
At a high level, every deal goes through three phases.
AI Filter. Every inbound deal is logged and read. It goes to a templatised AI filter. AI has access to the deck/startup material, the web and any prior virtual call that happened. An internal scoring mechanism ensures only the high-quality deals move to the second stage
Build conviction. We talk to the founders, go deep on the technology, speak with customers and domain experts, and model the business. More than one person on the team forms a view independently, so no single person's enthusiasm or scepticism decides the outcome.
Decide. The partners debate the full picture at the Investment Committee (IC). After that come the term sheet and a final round of detailed diligence.
The middle phase is where most of the time goes.
Where AI helps, and where it stops
AI does the groundwork. People make the decisions.
For deals that pass the first filter, AI:
Gives a first-pass score based on the pitch deck, public information and call transcripts.
Shows how much evidence the score rests on. A score built only on web data is flagged as weaker than one backed by a deck and a founder call, so the two are never read the same way.
Explains unfamiliar technology in plain language: the problem, the core insight, the technology from first principles, and what still has to be proven. It deliberately gives no recommendation, so the analyst forms their own view.
Handles pipeline admin, such as spotting duplicate entries and flagging deals that have gone quiet.
AI never votes on a deal.
What we're actually testing
Is the outcome big enough?We size the market from the bottom up, and only the segment the company can realistically win, not the whole addressable market. Then we work backwards: what would this company need to sell, to whom, and at what market share, to reach a fund-returning outcome?
Can this team get there?We look at domain expertise, technical depth relative to what has to be built, and whether the team can hire people better than themselves. We look for the right mix of founders: ideally one who builds and one who sells. Determination matters more in deep tech than almost anywhere else, because the timelines are long and the setbacks are real.
Is the technology real, and is it theirs?Do the technical claims hold up outside the pitch? Who owns the IP, and is the company free to use it commercially? We generally look for a 5x or greater improvement in performance, efficiency or reliability. A 20% improvement rarely justifies the effort of switching.
Will someone pay, and how soon?This is where time to revenue matters. We map how the company sells, how long sales cycles run (especially with governments and large enterprises), and whether the financial projections match that reality. A strong technology that is eight years from its first paying customer is a very different bet from one that is eighteen months away.
Does the funding path lead to an exit?How many more rounds will the company need, and how will those rounds dilute early investors? How far does the current raise get them? Are strategic investors involved? Who could plausibly acquire the company or take it public, and at what scale? We want to see a credible exit path on day one, not hope for one in year seven.
Why doesn't this get killed?What is the moat? What happens when incumbents notice? We look hard at dependence on regulation, on a single customer, or on a single geography. Sovereignty tailwinds matter here too. Countries are working to build defence, semiconductor, critical-mineral and energy capabilities at home, and that creates real demand. But a policy tailwind can open a market without protecting a company within it. We check whether the business survives if one external decision goes the wrong way.
Not all traction is equal
For companies without revenue, we place evidence on a ladder, from weakest to strongest:
Lab result → Prototype → Third-party validation → Unpaid pilot → LOI/MoU → Paid pilot → Design win → Qualification → Order Book/Contract
At each step we ask a follow-up question. For a prototype: what was simplified to make it work? For an LOI: does it commit to volumes and price? For a design win: how easily could the customer walk away?
A few habits we hold ourselves to
We write it down as we go. Our investment case is built up during diligence, not rushed together the night before IC.
We write down what would make us wrong. Before any vote, we are clear about what would have to be true for the investment to fail.
"Not yet" is not the same as "no". When timing is the issue, we note the specific milestone that would change our view, such as a qualification result, a key hire, or a regulatory decision, and come back when it happens.
We call someone the founder didn't suggest. Reference checks include at least one person the founder didn't put forward. Customer calls and independent expert views are usually where a deal is won or lost, so we never skip them.
The goal is not to find reasons to say yes. It is to find the small number of companies that can carry an entire fund, and to be honest with ourselves about everything else.