Three Things the Data Room Doesn't Show About Industrial AI Companies
- ecoxaiconsulting
- May 4
- 3 min read
I've spent 18 years deploying AI into manufacturing, oil and gas, and energy environments. Here is what I keep finding in deals that looked clean on paper.
1. The AI is real. The deployment is not.
The model exists. The demo works. The data science team is credible. But when you ask which customer sites are running the AI autonomously in production — not in a pilot, not with manual override, not with an engineer babysitting the output — the answer is usually one site, or zero.
In a recent engagement, a cleantech IoT platform was 30 days from closing a $22M Series B. The AI demand forecasting engine was central to the valuation. It had never run in a live building. The controlled lab environment it was demonstrated in did not have the sensor density or integration depth of a real industrial site. The AI was real — but 18 months from being deployable at scale.
That is a $5–6M valuation premium sitting on a demo.
The question to ask is not "does the AI work?" It is "which specific customer sites are running it autonomously today, and can I speak to the field team at those sites?" The answer to that second question tells you everything the data room doesn't.
2. The ARR includes revenue that won't recur.
Hardware deployment fees. Commissioning services. Custom integration work. One-time implementation charges. In physical-world software businesses, these show up bundled into ARR because the company genuinely believes they will recur — and sometimes they do. But stripping out the non-recurring components and rebuilding the ARR bridge from actual contract terms almost always produces a different number than the data room shows.
In that same engagement, $2.4M of $7.2M ARR was non-recurring install revenue. True SaaS ARR was $4.8M. The implied revenue multiple was overstated by 33%. The growth rate looked different too.
This is not fraud. It is optimism — and in industrial AI businesses, optimism is now an asset class. The job of diligence is to price it accurately.
The question to ask is not "what is the ARR?" It is "can you reconcile this ARR number to individual contract terms, line by line, and separate recurring from non-recurring?" Most companies cannot do this cleanly on the first ask. That gap is the signal.
3. One person owns something critical and nobody has documented it.
In industrial software businesses there is almost always a key-person dependency that is invisible in the org chart. Not the CTO — everyone checks the CTO. The person I mean is the engineer who wrote the firmware, or the data scientist who built the feature extraction pipeline, or the field technician who knows how to commission the hardware across the twelve building or facility types the platform actually supports.
In that same engagement, the entire IoT device stack — firmware, OTA update pipeline, hardware abstraction layer — lived in one engineer's commit history with no documentation. A departure would have halted hardware deployments for three to six months. The risk was unmitigated and undisclosed.
We made firmware documentation a closing condition. It was met at signing.
The question to ask is not "who are your key people?" It is "if your top three engineers left tomorrow, which parts of the product or deployment process would stop working, and where is that documented?" The answer reveals the real organizational depth behind the org chart.
These are not exotic risks. They appear in the majority of physical-world AI deals I assess. They are invisible in a well-curated data room and visible within two weeks of asking the right questions to the right people.
The gap between what the data room shows and what the business can actually deliver is where investment cases quietly fall apart — not at signing, but eighteen months later when the AI milestone is missed, the ARR bridge doesn't hold, and the key engineer is gone.
Traditional diligence checks whether the technology looks credible. That is a necessary question. It is not a sufficient one.
If you are evaluating a deal in industrial SaaS, energy tech, climate tech, or built environment where AI is central to the valuation — I am worth 30 minutes before you close.
Ramya Ravichandar, PhD is the Founder and Principal of EcoX AI Consulting. EcoX provides AI execution diligence for growth equity and PE investors backing physical-world software businesses.
Comments