What we sell, to whom, at what price, and what would have to be true for it to work. Written from the two calls. Every number marked est is a modelled assumption, not a measured fact.
"The one gap that I have is operational intelligence. Everybody talks about AI — utilising it operationally, at ground level, is where nobody is looking. And you don't need to be a huge player to be there. Everybody else is focused on the process. It's the user interface, that simple guy who goes out and needs to replace a fuse in the box at your house." Dirk, 31 July
The enterprise AI market is crowded, expensive, and aimed at head office. The very small operator is already reasonably served — a one-van plumber can take a card payment and a signature on his phone today. The gap is the layer in between: companies large enough to have a real operational problem and a budget, small enough that no serious vendor is building for them.
The specific unmet need is capture at the point of work. A technician finishes a job and the detail of what he did — hours, parts, diagnosis, the photograph of the serial plate — either gets written down badly or not at all. Everything downstream depends on that detail: the customer sign-off, the invoice, the warranty claim, the job costing, the decision about whether that customer is even profitable.
The business we're actually in: we close the gap between work performed and money collected, for companies whose people work away from a desk. Field capture is the wedge. The back office is the business.
This was the sharpest call of the conversation and it came from Dirk. Not the one-van operator — that market has products. One step up. Typically 10 to 40 employees, 5 to 25 trucks, an owner who still knows every customer, a bookkeeper, and no operations function to speak of.
| Segment | Revenue | Served today by | Our fit |
|---|---|---|---|
| Owner-operator, 1–3 vans | Under $500K | Jobber, Housecall Pro, Square — adequate | Skip. No budget, solved enough. |
| The target band | $2–3M | ServiceTitan is too heavy and too dear; spreadsheets fill the gap | This is the wedge. |
| Regional contractor | $10M+ | ServiceTitan, BuildOps, full ERP | Later. Longer sale, real competition. |
| Oil field services | Varies | Bespoke, expensive | Adjacent. Higher value per job — worth a look once we have proof. |
Selling this means satisfying two people whose interests barely overlap. Get either one wrong and it fails.
He does not care about your system, your dashboard, or your data model. If capture costs him more than about thirty seconds he will work around it, and then the whole thing is worthless. Adoption by the technician is the entire technical risk.
He wants the job card signed off by the customer, the invoice out the same day, and a view of which jobs and which customers actually made money. He is the buyer and the budget holder. "I want to be able to invoice that guy and get paid as soon as possible."
The capture app is not the hard part and we should not pretend otherwise — photos, parts, hours, notes, signature. Any competent team can build that. The hard part, and therefore the moat, is what happens behind it: the job flowing into QuickBooks or the ERP, the invoice raising itself, the parts drawing down inventory, the costs landing against the right job.
A $299-a-month point solution can give a contractor a capture form. It cannot make that form talk to his accounting system, his price book and his payroll. That integration is what we are selling, and it is exactly where thirty years of ERP work between the two of us is worth something.
Dirk's constraint: "You're not going to create a solution for every single team." The unit of delivery has to be a configurable agent, not a bespoke build — otherwise the fifth client costs as much as the first and there is no business. Everything we ship should be assembled from parts we already own.
Every technician already carries a camera. Photograph the unit, the serial plate, the failed part, the finished work — and let a vision model extract the detail rather than asking a man with dirty hands to type it. This is the single biggest reduction in capture friction available to us, and it is the thing that was not possible when Dirk looked at this problem at HCL.
Start on free tiers — Cloudflare, a hosted database, our own accounts. Proving the model costs effectively nothing. Move to AWS, Google Cloud or Azure when client volume or a client's own procurement requires it. Two delivery shapes, and we should offer both: managed (we host and run it, recurring fee) or client-owned (their cloud account, we build into it).
"It's all about the package. You want to sell it in a package. You pay me — that's the amount you get for us to service you." Dirk
Not hours. Not a bespoke scope every time. A named offer with a number on it, tuned per client rather than rebuilt per client. Then let the buyer choose how to pay, because the choice itself does sales work — it tells a sceptical owner that we are willing to be measured.
Fixed fee against defined milestones. The client sees something finished before money moves. Familiar, low-friction for buyers who want a contract they understand.
Little or no money up front. We take a share of what we create — faster collections, recovered billable hours, reduced leakage. Our incentive and theirs point the same direction.
"My incentive is to make you succeed. Because the more success you have, the more money I make." Dirk, on why alignment beats fees
Eduardo's addition: "You create enough alignment and the whole machine becomes a flywheel." Dirk already runs this structure with his referral partners — a percentage of what they bring, "and whatever percentage of zero is zero."
The catch nobody has solved yet. Outcomes-based pricing requires a baseline both sides accept — days-sales-outstanding before and after, billable hours captured before and after, invoice disputes before and after. Most $2–3M contractors do not have clean enough books to establish that baseline. Which is itself an argument for a short paid assessment first, to measure the starting point. That may be the real first product.
In the 1990s vendors would place hardware free for three months with a licence that started nagging near the end. The client had nothing to lose on day one, and by the time the nag started they were dependent. Same psychology, updated: land it, make it indispensable, convert. The modern version costs us almost nothing to run.
Illustrative model, not a forecast. Built to test whether the shape works, not to predict revenue. Every figure is est.
| Per client, year one | Model A | Model B |
|---|---|---|
| Implementation fee | $25,000 | $0 |
| Managed service, 12 months | $30,000 | $0 |
| Share of value created | — | $40,000 |
| Gross revenue | $55,000 | $40,000 |
| Infrastructure & model costs | ($3,600) | ($3,600) |
| Delivery effort | ~120 hrs | ~120 hrs |
| Contribution | $51,400 | $36,400 |
| Effective rate | ~$428/hr | ~$303/hr |
The point of that table is not the numbers, which are invented. It is the shape: infrastructure cost is a rounding error and the whole business is our time. That means the only two questions that matter are how many hours a client takes and how much of the second client's build we can reuse from the first.
The reuse question is the business. If client two takes 120 hours like client one, we have a consultancy that tops out at three people's capacity. If client two takes 40 hours because the agents already exist, we have something that compounds. Every scoping decision should be made with that test in mind.
Rather than argue the case, here is what we would need to believe. If any of these is false, the venture is different or it isn't there.
Everything rests on this. If capture takes more than about thirty seconds, or needs typing, or fails without signal in a crawlspace, adoption goes to zero and the data never arrives. Test: put a prototype in front of five working technicians before writing a line of integration code. Watch them, don't ask them.
Dirk believes this band is underserved and Albert reportedly lives in it. But underserved and willing-to-pay are different claims. Test: Albert gets us in front of five owners. We ask what they currently lose to bad job capture — and whether they can even answer.
Our advantage is that ERP and accounting integration is genuinely hard and we are genuinely good at it. That is a real moat against a $299/month product. It is a weaker moat against ServiceTitan deciding to move down-market, and a weakening one as AI makes integration cheaper for everyone. Test: how long is the window, and what do we own at the end of it besides code?
All three of us are consultants by training. The default failure mode is a bespoke build per client, a good living, and no asset. Test: before client two, write down what will be reused. If the honest answer is "not much," fix the architecture before selling again.
Dirk is a serving Accenture MD. Eduardo is running a job search alongside this. Albert is unconfirmed. Test: agree honestly how many hours a week each of us has before we commit to a client delivery date. Also: Dirk should check his Accenture obligations on outside activity early rather than late.
See the warning above. If we cannot establish a clean baseline in a typical target client, Model B is unsellable and we are a fixed-fee shop. Test: try to construct the baseline for one real company and see how hard it is.
The risk both of you already named: "It's easy to want to go too big too soon." The counter, also from the call — "put out three fishing rods, see which one bites, then score that one." The discipline is picking three small, cheap, genuinely different bets rather than one large one, and being willing to kill two.
"We're going to tell the shoemaker: focus on making shoes. We will help you run the business so you can make more shoes." Dirk, 31 July — Eduardo's response: "Beautiful. Simple."
It works because it says nothing about AI. The owner of a $2.5M HVAC company does not want an AI strategy; he wants his trucks billable and his invoices out. The technology is how we do it, not what we sell.
The companion point, on why we can charge properly for this — from Eduardo, recounting someone pointing out that AI can do a great deal now, "but you need to know how to operate it. It's a machine. And this is what we bring. We know how it works." The conclusion both of you reached: don't undersell.