All Insights
Unit EconomicsFinance StrategyDeep Tech

Five Numbers Every Deep Tech CEO Should Know Cold

V

Vadis Advisory

March 22, 2026 · 8 min read

Revenue is a vanity metric—at least on its own.

We’ve worked with founders who knew their top-line number to the dollar but couldn’t say whether the business underneath it was healthy. In deep tech that gap is wider than usual, because the costs that decide the outcome—equipment, unit cost, compute—are the ones most likely to be under-modeled.

Here are the five numbers we care about most.

1. Gross Margin (and what’s inside it)

Gross margin is revenue minus the direct cost of delivering it, as a percentage. It tells you how much value you create per dollar of revenue before overhead.

The more important question is: what’s inside it?

For a hardware company, the honest answer includes labor, yield loss, scrap, freight, duty and tooling amortization—not just parts. For an AI product, it includes inference. Every prompt, every agent run and every heavy user shows up in cost of goods sold, and it scales with usage rather than with headcount.

Classic subscription software runs at 70–80%+. AI products and hardware usually run lower, and that’s fine—as long as you know why, and the number is moving the right way as you scale.

What to watch: Is gross margin stable, improving or eroding as volume grows? Margin that erodes with scale is a structural problem, not a timing issue.

2. Cash Runway

Runway is how many months you can operate at your current burn before the money runs out. It’s the most existential number in any company.

The formula is simple: cash on hand ÷ net monthly burn = runway in months.

The catch in deep tech is that burn isn’t smooth. Long-lead equipment, tooling, prototype iterations and test campaigns arrive in lumps, and a model that assumes spend scales with headcount will be wrong in exactly the months that matter most.

What to watch: Model runway month by month with the lumps in it, not as a single average. Deep tech raises take longer than software raises, so plan to start raising with well over a year of cash left, not six months.

3. Fully Loaded Unit Cost at Volume

If you ship a physical product, this is the number your next round will be priced on.

Cost the bill of materials at three volumes—today’s, next year’s and at scale—and state the assumptions behind each. Then add everything the BOM leaves out: labor, yield, scrap, freight, and the tooling you’ll amortize across every unit.

The point isn’t precision at today’s volume. It’s showing a credible path from what a unit costs now to what it has to cost for the business to work.

What to watch: Your yield assumption. Plans often assume yield improves on schedule. Model what happens to margin if it doesn’t.

4. Cost to Serve per Customer

If compute is part of what you sell, this is the AI equivalent of unit cost.

Map inference and infrastructure spend to individual customers, features and usage tiers. Flat per-seat pricing often hides the fact that your heaviest users are your least profitable ones—and they’re usually the customers you’re happiest to have.

What to watch: Gross margin at light, median and heavy usage. If the heavy tier is underwater, fix the pricing—usage caps, credits or usage-based tiers—before that customer segment grows.

5. Operating Cash Flow vs. EBITDA

EBITDA (earnings before interest, taxes, depreciation and amortization) is a common proxy for profitability. In deep tech it can be badly misleading, because so much of the action is in working capital and equipment.

Operating cash flow tells you what’s actually moving in and out of the bank account. The gap between the two is one of the most revealing numbers in the business.

A company with healthy EBITDA but weak operating cash flow might have:

  • Components and labor paid for months before a customer—often a government customer—pays for delivery
  • Inventory and work-in-progress quietly absorbing cash
  • Milestone revenue recognized before the cash arrives

What to watch: Over a rolling 12 months, operating cash flow and EBITDA should converge. And remember that capex sits outside both—a company can look fine on EBITDA while equipment purchases drain the runway.


The point

Knowing these numbers doesn’t make you a CFO. It does make you a better CEO—one who can spot problems before they become crises, test the big commitments before making them, and hold your own with investors and your board.

The best financial leaders we’ve worked with don’t just know these numbers. They know why they’re moving, what’s driving the movement, and what they’re doing about it.

Want help building a model around these numbers for your business? Let’s talk.

V

Vadis Advisory

Strategic Finance & Capital Advisory

Vadis Advisory provides fractional CFO and strategic finance services to funded deep tech companies — AI, life sciences, and defense / hard tech. We bring institutional rigor to every engagement.

Work with us