Le modèle d'IA de dépannage HVAC que personne d'autre ne peut entraîner

French HVAC represents €7–8.8B in revenue across ~45,000 companies, ~78% with fewer than 10 employees — a fabric of tradespeople and micro-businesses, our core target. The addressable technician pool (refrigeration, HVAC, and heating engineers working on heat pumps and boilers) is around 70,000–80,000 people in France.

Permanent tension: ~15,000 unfilled positions per year, near-zero unemployment, departure of experienced profiles. Demand overflows; what is missing is the capacity to handle it with less experienced technicians on an increasingly electronic installed base. The fleet is growing fast — ~179,000 air-to-water heat pumps sold in France in 2025, with REPowerEU targeting 10M heat pumps in Europe by 2027.

This is the core market thesis: we only deal with technical knowledge, never regulation. And technical knowledge has no borders. A Daikin VRV, a Mitsubishi heat pump, a Viessmann boiler are strictly identical from one country to the next — same fault codes, same procedures, same failures. Troubleshooting knowledge transfers 100% internationally, where a regulation-centric product would stay locked behind a border.

In practice, the asset we build in France is immediately reusable everywhere. The application is already multilingual — 11 languages in production. Expansion is not a localization project; it is opening a market on a product and model that are already ready:

• English-speaking marketsUK, Australia, New Zealand — same language, same equipment, accessible without adaptation.
• Continental EuropeDE, IT, ES… — identical equipment (AC, VRV, heat pumps, boilers), languages already covered.

A TAM of "1.5M technicians × €228" proves nothing. What matters at pre-seed is a defensible bottoms-up view, market by market. Beachhead: already-digitalized tradespeople and service/maintenance companies running with temporary staff (maximum pain — generic technician, unknown machine, every day).

On a France base of ~50,000 activatable technicians, the issue is not market size but freemium conversion rate. The table below shows full-penetration potential — an upper bound, not a forecast; actual landing pace is detailed in the financial annex:

This potential replays identically on each market, at near-zero marginal product cost (11 languages already in place).

The mobile app in a single country is a few million euros of ARR: a profitable beachhead, not a billion-dollar outcome. The prize size comes from stacking three levers, all unlocked by the same asset:

• 1
  Multi-market replication11 languages already live — expansion means opening, not building.
• 2
  API/MCP layer (P2)Functional, in testing, pricing open: sold to maintenance companies, integrators, and manufacturers — value per account far exceeds €19/month.
• 3
  Proprietary modelThe underlying asset a generalist cannot replicate without the corpus.

The tradesperson market, multiplied by languages already covered, funds building the asset; the asset opens a global market.

Market sources: abcclim.net (French HVAC employment), recrute.gesec.fr (vacancies), European Heat Pump Association (REPowerEU technician shortfall), cedeo.fr / dispart.fr (EasySAV competitor).

No one waits for AskMarcel to troubleshoot. Facing an unknown fault, the technician already does one of these five things:

• 1
  Ask ChatGPT / Gemini / Perplexity — now reflex #1, and our real competitor.
• 2
  Call the manufacturer hotline — when it exists, is open, and is the right brand.
• 3
  Search a PDF manual — if they can find it.
• 4
  Post on a Facebook group / forum / watch YouTube.
• 5
  Call a senior colleague — the historical competitor, the one retiring.

We do not create the need. We replace a workaround technicians already perform poorly.

Each alternative is good at something. The honest comparison is not "them bad, us perfect":

The real investor question: "What if OpenAI or Daikin does it?" Honest answer:

• Against a generalist AIIt will not build a vertical HVAC corpus or a field data loop for a niche trade — and even if it tried, our asset (10,000+ manufacturer docs + 50,000 fault→resolution pairs + field usage) is precisely what model size cannot catch up to.
• Against a manufacturerSingle-brand by nature. Value stops at their own machines; the technician is multi-brand. No manufacturer will aggregate competitors' data. Our multi-brand neutrality is structurally out of reach.

Two channels already drive real traction at near-zero cost. The rule is simple: double down before opening anything new, and any channel that cannot prove its acquisition cost is cut — not maintained out of habit.

• Compound what worksSEO and emailing already drive real traction (+150% traffic, 65k contacts) — we double down before opening a new channel.
• Payer changes, user staysThe distributor channel does not replace the tradesperson user — it changes who pays and how we reach them.
• Credibility through the networkThe founder (ex-Daikin France sales) opens doors no ad budget can buy.
• Discipline on channels that do not convertA channel that cannot prove its CAC is abandoned, not maintained by habit.

Programmatic SEO. 302 technical articles already published, +150% organic traffic. Channel #1: marginal cost, compounding acquisition, and it replicates by language. With 11 languages in production, the content engine that acquires in France is the same one that acquires in the UK, Germany, or Australia. Organic acquisition is, like the product, already international.

FGAS/RGE emailing. 65,000 qualified tradesperson contacts — a direct list at near-zero cost, exactly our target. Freemium activation channel.

This is the advantage neither SEO nor an ad budget can buy. A technician — and a service company — passes through the distributor counter several times a month: the most regular field touchpoint that exists. And I have direct access from 15 years in the sector (CIAT, Daikin) and strong knowledge of HVAC distributor networks.

The counter serves two distinct goals: get tradespeople into freemium (volume + data), and reach service/maintenance companies — payer #1 — where they already buy supplies. Execution detail is in Section 8.

Every channel serves the same mechanism: SEO and emailing fill freemium and feed the model; counters open the door to payer #1 and densify data (one service account = multiple technicians). We are not stacking channels — we are feeding a loop.

This channel is not sold top-down. It is built through usage: we prove a counter generates real usage before formalizing a commission contract. We approach HVAC distributor networks, and usage numbers decide what happens next — not a sales deck.

One counter, two distinct targets activated differently:

• Tradespeople enter freemium via the counter (QR code, referral code). Once usage is proven, the distributor resells the subscription and earns commission. This fills volume and data.
• Service/maintenance companies buying at the same counter are identified and worked through direct sale (payer #1, multi-seat contract).

Everything is measured, not declared: counter QR code, referral code, Supabase tracking. We know exactly which counter produces which usage.

We start with independent and regional distributors. They are more agile, decisions happen fast, and critically they currently have no diagnostic tool to offer their customers. We bring a service that differentiates them, on a neutral, multi-brand product.

Nationals come second, from a position of strength once the model is proven. It is also a competitive necessity: some, like Saint-Gobain (CEDEO/Dispart), already push their own app (EasySAV) — so either locked or head-on. No point attacking them first; better to prove the model where the field is open.

Their existence proves the need. Their dependence on a parts catalog is precisely the door we take: an independent does not want to send customers to a national competitor's tool.

Thibaut Marcelot — founder, sole operator. 15 years in HVAC engineering: field, sales, and distribution (CIAT, Daikin), plus energy renovation and solar. I build the product and lead commercial. My distribution background gives direct access to the counter channel.

Development of certain features uses occasional freelance — no hiring ahead of need, no heavy structure. Spend when there is proven need, not before.

We do not hire based on a forecast. The first hire — likely a commercial profile to replicate the distributor channel — only happens after the first signed contract. Product/support reinforcements follow API layer scale-up, not the reverse. Same discipline as channels: we fund what is proven.

A solo founder supported by occasional freelance is minimal fixed cost and maximum decision speed during the most uncertain phase. The risk it creates (key-person dependency) is real — I address it head-on in Section 11 rather than hiding it.

This pragmatic approach is the project foundation. It enables a deliberately progressive trajectory, limiting risk and strengthening credibility with users.

The HVAC technical assistance standard, in Europe and beyond. A product already multilingual, in a market where equipment is identical country to country, backed by a proprietary model whose quality deepens with every intervention. The tradesperson beachhead funds the asset; the asset opens service companies, integrators, and manufacturers.

• A field assistance standard for HVAC in Europe
• A reference tool for skills development
• Reliable support in a structurally strained sector

Need is documented and growing, the asset is already seeded (10k+ docs, 50k faults, 11 languages, API/MCP in test), and the founder knows both sides of the table — field and distribution. The window is open: players like Saint-Gobain are already moving. Funding transforms a seeded asset into a model whose quality creates revenue — not masking a structural weakness.