Our investor network has underwritten, co-invested in, or passed on roughly 240 hardware robotics deals since 2022. This is the condensed version of the diligence framework our members converge on after enough of those deals have printed. It assumes you have already done the standard team, market and traction work and are ready to look at what the asset on the table actually is.
1. The bill of materials tells you the truth
Ask for a line-item BOM at current volumes and a projected BOM at 10,000 units. The ratio between the two is one of the most honest signals you will get. A founder who cannot produce either document is telling you that they do not yet understand what they are building.
Flag any BOM where a single component exceeds 18% of the total cost. That is concentration risk, and it usually means the founder has not validated a second source. In 2025 we watched three Series-A-ready companies stall for six months each because their primary motor or sensor supplier pushed out lead times by a quarter.
Service and support margins
Hardware revenue is headline revenue. Service revenue is the business. Ask: what does the first-year service contract cost the customer, what does it cost the company to deliver, and what is the renewal rate at month twenty-four? A gross service margin below 55% in year two usually means the robot is not yet field-hardened.
The deals that blew up on us were never the ones with a soft BOM. They were the ones where service cost more than the quarterly subscription fee by a factor of two, and nobody on the deal team had modeled it.
2. Integration risk is almost always underestimated
A robot that works in a lab is a robot that has solved 30% of its deployment problem. Questions we ask on every diligence call:
- How many distinct customer sites has the system been deployed in, and for how long?
- What percentage of deployment time is spent on non-robot integration (racking, power, network, safety fencing)?
- What is the mean time between human interventions across the live fleet?
- Who handles on-site service — the company, the integrator, or the customer?
- What contractual SLA is in place, and what does the penalty schedule look like?
3. Field data quality is the moat
Two robotics companies can have identical hardware and radically different valuations because of the data flywheel. Ask to see the data pipeline end-to-end: how is it collected, how is it labeled, how is it fed back into model training, and how fast does a new firmware version reach the fleet.
The best teams can push a firmware-level model update to 100% of the fleet in under 72 hours and roll it back in under an hour. The median team takes 9 days to push and cannot roll back without a truck roll.
Red flags that survived our 2025 vintage
After reviewing the deals that underperformed from our 2023–2024 vintage, three patterns recur: founders who could not articulate unit economics at 1,000 units; a gap of more than six months between the most recent firmware release and the diligence call; and a service organization that was more than 30% the size of engineering.
4. The team questions that matter
We ask every founder to walk us through the single hardest engineering decision of the last six months, in detail, including the option they rejected. Founders who can do this fluently are almost always ready to scale. Founders who default to marketing language are not.
We also ask who on the team has shipped a physical product before, at what scale, and what broke. A company where the senior hardware leadership has never shipped more than a hundred units into the field will spend its Series B learning things a second-time founder learned before the round closed.
None of this framework replaces judgment. It just prevents the failure modes that have cost our network the most capital — the ones where the deal looked clean because nobody asked the uncomfortable question until six quarters later.