Hi ! I’m Renan, and I lead OSS Ventures, the most awesome venture builder specialized in software for operations in Europe (to my knowledge, we’re the only ones). We typically build 4 software companies per year, and per last 18 months we’ve been all in on AI, creating 22 software companies deployed in 3000 factories globally. I try and take time to reflect each week on a particular, hands-on topic coming from us building and our portfolio informing us. Today, we reflect on selling AI software for industrial companies.
There is a strange paradox unfolding in industrial software today. On one hand, AI finally gives us the technical leverage we’ve been dreaming about for decades: the ability to automate expert work, to turn messy processes into structured systems, to replace hours of human reasoning with a few seconds of computation. On the other hand, selling this new category of software has never been harder.
Not because industrial clients don’t understand AI. They understand it better than most people think. But because delivering real, tangible, bank-account-level value with AI — and proving it — requires a different way of selling, deploying, and even thinking about software.
After deploying AI systems across dozens of factories, building companies that automate everything from visual inspection to quoting to waste optimization, and watching some models fail while others quietly transformed operations, a few patterns have emerged. These are not “truths” — reality always has the last word — but they are observations from the trenches. They explain why selling software in the AI era looks nothing like selling SaaS in the 2010s.
It starts with the simplest and most uncomfortable fact: AI now replaces labor. Not figuratively. Literally. And that changes everything.
When your product replaces labor, your client profile changes
In the SaaS era, software automated workflows and made people faster. In the AI era, software often does the work.
At OSS Ventures, we’ve seen this across multiple categories: purchasing, CAPEX management, quality control, quoting, supply chain forecasting, customer service. In each case, the promise is no longer “your team will be more efficient” — it’s “you will need half the team.”
That possibility demands a certain type of buyer: one who is comfortable with deep transformation, who can take the political heat, who can align their organization behind a step change. In our experience, those buyers represent roughly 15% of the industrial world. They are innovators — executives who see technology as leverage, not as a threat.
These innovators share a trait: they want to believe, but they need to see. Not a dashboard. Not a PoC. Not a simulation. Live value. Because they’ve been burned too many times by vendors selling smoke and mirrors.
And so, paradoxically, AI pushes sales back toward something much more tangible: “show me the gains, in my factory, with my data, on my machines.”
Pre-sales is no longer talk — it’s engineering
Selling AI solutions today looks closer to consulting than to SaaS. One of our portfolio companies routinely sends two engineers on-site for a full week before a deal is even signed. They connect to raw data sources, crunch months of signals, and produce a quantified model of what the AI could deliver: fewer changeovers, faster cycle time, fewer false defects, fewer people needed for procurement or planning.
This isn’t a nice-to-have. It’s the only way to build conviction.
Traditional SaaS taught us that scalable companies avoid costly pre-sales. AI teaches us the opposite: you earn the right to scale by proving the production reality upfront. Until the model sees the real world, everything is theory.
Interestingly, clients love this approach. Factories respect engineers who get their hands dirty. They respect people who show up at 6 a.m. for a shift change and open a PLC cabinet to understand the data flow. They respect those who say “this will work” only after testing it.
AI, as a category, has forced the industry back to expertise, humility, and on-site reality.
When it works, pricing flips completely
The upside of this visceral proof-based sales motion is that, once the system is live, trust becomes extremely high. Factories — at least the innovative ones — are transparent, rational, and surprisingly generous when aligned incentives kick in.
A pattern emerged across our companies: once value is demonstrated live, pricing becomes outcome-based. We often charge roughly 20% of the gains we create. When a purchasing engine reduces costs by a few million euros, when a quoting AI turns unprofitable parts into profitable ones, when a visual inspection system frees up dozens of operators — the platform earns its place.
Outcome-based pricing in AI feels natural because the value is not theoretical. It’s visible on the P&L the following month. And clients in industry are used to supplier relationships where long-term alignment matters more than short-term procurement tactics. In that sense, industrial buyers are far healthier than many software buyers in the tech world.
The other uncomfortable truth: AI exposes which clients are not ready
AI is unforgiving. It amplifies the underlying reality of a client: their data hygiene, their governance, their operational discipline, their leadership maturity.
One company in our portfolio became successful by doing something counterintuitive: they actively cut clients early in the process. If they sensed that the factory would not be able to unlock the gains — because of organizational rigidity, misaligned incentives, outdated equipment, or political turbulence — they ended the pilot.
This is not arrogance. It is essential. AI’s value is multiplicative: if one link in the chain is weak, the entire outcome collapses. You cannot brute-force success.
This is also why AI sales cycles feel both faster and more brutal. Compatibility becomes obvious very early.
The economics of inference are misunderstood — by vendors and buyers
One of the most common mistakes in “AI SaaS” has been trying to charge per token or per inference, as if AI were just a variable cost commodity. We tried it. It failed.
Industrials don’t want tokenized unpredictability. They don’t want to feel like they’re paying for cloud provider margins. They don’t want their cost model to depend on the whims of model pricing. Many have been burned by cloud era over-usage, and they won’t relive that.
The fairest model we’ve found is the opposite: let clients host the model themselves, or pass inference costs at exact rate with no markup. Everything else sits on top: the application, the business logic, the deployment expertise, the safety constraints, the validation loops, the organizational change.
It’s more ethical, more aligned, and more compatible with industrial cultures of trust.
And it leads to deeper, longer, healthier relationships.
What about margin ? Isn’t it lower than SaaS ?
We made more than one fancy excel for more than one company and the short answer is : it’s actually better if you do it well.
High-tickets can begin with shorter cycle with more humble proof and paid-for engineers proving the system in a first factory. Scale happens sooner. And the total value is higher.
And the cost structure is better.
If you sell the way you sold SaaS in the 2010s though, the awakening will be brutal : internalizing the inference costs in the face of uncertainty, not being humble and proximate enough from your clients to have a strong ROI, compete for scaling with other initiatives — good luck with that.
So, how do you sell software in the AI era?
If SaaS sales in 2015 were about features, funnels, and top-down pitches, selling AI in 2025 is about something simpler and older: earn trust through reality.
Spend time on the factory floor until the problem is not abstract anymore.
Build models on real data, not generic assumptions.
Show value before signing.
Cut early when the setup isn’t viable.
Charge on outcomes when the system delivers.
Align economics with real-world value, not cloud abstractions.
This is not “best practice.” It is simply what works when the software actually changes how people work.
And it needs to be said clearly: none of this is definitive. AI is evolving too fast, and we learn something new in every factory we touch. Some of these principles will break. Others will become industry standards. Reality will keep surprising us — and that’s the privilege of building in this space.
What is certain, though, is that selling AI software is no longer about convincing people to adopt a new tool. It’s about partnering with people who want to change how their factory operates — and being humble enough to let the truth of the field guide the journey.
The market is ready. The innovators are listening. And the next decade of industrial transformation will belong to the builders who respect the reality of factories, not the fantasy of pitch decks.