Software that lets us grow without growing the team.
The unified, AI-native platform across the entire customer journey — configure → order → prep → install → service — built to decouple growth from headcount.
Quatt is winning sales. But every time we sell 50% more, we’ve had to hire ~50% more people to deliver it.
QuattOS breaks that weld.
One platform that runs the whole journey, with software and AI agents doing everything except the physical install. Growth restored — not by hiring into it, but by building a machine that doesn’t need us to.
Everyone has a decent heat pump at a similar price. You don’t win on the hp alone anymore. You win on experience — easy, smooth, fast, affordable, magical — and on operational cost.
~9 systems glued together: HubSpot (~100 deal stages, 390 active workflows), QST, NetSuite, Skedulo, 362 Make scenarios, spreadsheets, email. No single system knows the complete state of a customer.
Models are finally good enough for real customer work; AI-assisted development means a small team ships what took an army. Whoever builds this operating model in 12–18 months gets a cost structure nobody can catch.
Today, to answer “where is this customer and what’s next?” you query four or five tools. That’s not a foundation you scale on — it’s spaghetti that breaks under load.
The majority of our cost is in the lead-to-install and support process — not the hardware. So that’s the highest-leverage thing we can touch.
The elasticity that powers it: a €500 price cut → roughly +9 deals/week (+12%), ~€2.5M/year.
Anything deletable, delete. Own 100% of the stack.
Where the customer lives the whole way through: configure, upload photos, track the order, complete prep, apply for financing & subsidy, pay.
The portal IS the product.
The configurable brain underneath. It knows what tasks exist, in what order, which are blocked, which run automatically. Country-, product-, channel-agnostic.
Scales to new variants without new code.
The internal cockpit. Real-time visibility, run by humans at first, progressively handed to AI agents. The dashboard is the escape hatch, not the primary interface.
For a board it lands as four plain-English components: a self-serve configurator · an automated post-order workflow · ops dashboards as a single source of truth · AI agents that take over the repetitive human work, task by task.
Every problem runs through the same five steps, in this order:
Automating a broken process just breaks it faster.
The order is the whole point. The build cadence matches: daily commits, weekly ships, hourly iteration. Delete-and-iterate is the operating mode.
Every piece of operational truth has a single home — QuattOS. HubSpot, NetSuite, Skedulo become consumers of that truth, not co-owners of it.
A journey is a set of tasks with dependencies — not a fixed position in a linear stage list. That’s what makes it reorderable.
Humans, AI agents and outside systems all hit the same endpoints — no privileged path. The trick that lets us swap a human for an agent on any task without re-plumbing.
The same destination, reached deliberately. Today HubSpot still holds deal state so sales keeps its view; over time QuattOS becomes the single source of truth and HubSpot’s deal stage is retired.
Why it’s safe: in both pictures only one system writes state. We just move which system is canonical — HubSpot today, QuattOS tomorrow — without ever having two writers.
At any point a customer is in exactly one state and can only move when conditions are met. A signature moves quote sent → quote signed; satisfied work-prep checks move it onward. Nothing else touches those fields.
Country, product, partner and payment are configuration, not code. Sequencing is configurable too — a German regulatory step, a bulk-partner shortcut. An installation partner is just a configuration dimension on the standard pipeline.
Does the actual work: create the NetSuite order, schedule the Skedulo job, charge in Stripe, update HubSpot, notify the customer and the right teammate on Slack — in order, checked, retried on failure, and surfaced immediately if it can’t complete. No silent failures.
Each task carries the logic that makes the system run itself: auto_if · skip_if · due_at · dependencies. Not different systems — different orderings of the same primitives.
And the one task type that is always human: INSTALLATION — atoms are moving.
Genuinely undecided — and worth getting right early. Property persists independently of the customer, so returning customers and shared-address history just work.
Agents are first-class operators of the business, not a chatbot in the corner. Because every action is the same idempotent API call whether a human or an agent makes it, we replace work one task at a time — no big-bang re-architecture. The workflow engine stays in control; AI handles the judgement.
The deeper build philosophy: you don’t need to write so much software — you write pipelines with skills. Skills are editable markdown, so behaviour changes with no code change. And the real differentiator is a self-improvement loop — every day asks “what did we learn — update the definitions, the priors, the decision log, the strategy?” That’s where the compounding is.
Today, at every handover — sales → work prep → installer → ISM — context is lost. The portal keeps everything in one place. The information accumulates and stays visible to everyone, end to end.
Sees their journey; enters info, uploads, signs — or does it with a sales agent who opens the same screen.
Acts on everything the customer entered; both sides notified on every change.
Picks up exactly where sales left off; reviews, approves, builds the placement plan.
One summary: house, piping route, materials, notes. Nothing said aloud and lost.
Full history if something goes wrong after install — approved, communicated, noted.
The portal replaces QST — and quietly does something bigger: every entry, approval and document becomes the structured record that makes the AI world model possible. An assistant answers grounded in the actual record, not a generic FAQ.
The biggest thing between a customer and a yes is hesitation — our biggest conversion blocker. So we design it out.
Postcode + house number auto-fills the home. Twenty click-through questions become one slider. We recommend one right configuration, not twelve. Benchmark: BOXT — best-in-class online heat-pump buying.
A small down payment marks the order “converted.” “Pick your install date” creates ownership. “Finish the configurator fast → guaranteed install in ~2 weeks” selects for high intent. Ask when you’d like to be called, not whether.
Pre-checks run while the customer types. Order with open questions, and an expert resolves them after — never blocking the purchase. Plans auto-generated from the customer’s own photos.
If we rebuilt our installer tooling from scratch, what would it be? Yoshi is QuattOS’s installation & service module — and the tip of the spear of the tooling we’ll eventually sell to installers across Europe. Its customers are both internal and external installers.
Set hardware up fast and correctly — connectivity, settings, full functional verification.
Confirm it actually performs — heat pump, central heating, and electrically.
Field-tech tools — component tests after a replacement, actionable diagnostics.
Surface problems before the customer calls — proactively, from the fleet.
Partners see only their own installs; Quatt sees all. GDPR-structured by design.
Run as a product, not a feature (discovery led by Bob van Iersel & Julie Roland Sørensen). Remote monitoring builds on Lino Hendriks’ AI-first ticket-resolution system — a deterministic flag opens the case before any customer calls. These are the install/service rails the B2B flywheel sells onward to SME installers.
We don’t write a PRD and then start coding. Yoshi is our pilot for an AI-first build loop — capture reality, generate working prototypes from it, let customers react, and let those recordings become the spec the agents build against.
A large set of stakeholders give input — much of it async, via voice messages.
All discovery meetings — including on-site desk visits — are recorded.
Recordings generate working prototypes — possibly live, during the session.
Installers try the prototypes and record their reviews.
Reviews + architecture maps + acceptance criteria become the scaffolding agents build in — orchestrated by product + a software architect.
Once the scaffolding stands, drop into a daily shipping cadence and iterate.
Weekly reviews keep it honest — the standing artefact is a working prototype, not a status doc.
Tuesday evening: configure, pay a deposit, record a video walkthrough. An AI agent analyses the piping, recommends a route, generates the plan, confirms in minutes. She picks Thursday. The installer arrives; the heat pump runs. First click to installed: three days. €4,000 without subsidy.
A PO adds a bulk installation partner to the configuration layer — decoupled ordering, partner invoicing — and builds their portal on documented endpoints. Because the backend behaves consistently, it works in production with no follow-up project. Ordering within days of signing.
A spec, not a prompt. An agent reads the codebase, builds the feature, adds it as a config entry, opens a PR with its own tests. Three AI review passes + full CI. The dev checks for what the pipeline can’t. Ships behind a flag the same day; A/B running by morning.
An early “46.4% post-signature drop-off” turned out to be misleading. Cohort analysis showed converted → installed is actually 95.6% — true post-conversion attrition is only ~2.8%. So the case rests on operational efficiency and growth enablement, not recovering imaginary lost revenue.
We enter heavily-regulated foreign markets with the cost structure of a software company instead of a field-services company — in weeks, not quarters, capital-light.
The wedge is AI-first dossier automation that cuts out the aggregator middleman.
admin cost per install. A structural moat, not a discount.
All-electric, BOXT-style digital journey, freelance install network. Strip ~£800–1,000/install of survey + compliance paperwork.
net price to the customer.
Incumbents have slick front-ends but fragmented back-ends — they can’t mandate a single operating system onto the physical job. We can. That’s the moat.
The same cost edge that wins us customers puts every SME installer under pressure — and the way they relieve that pressure is to run on our rails: buy our hardware, and adopt QuattOS to get the same efficiency.
The strategic intent: make it so hard to win on sales that an SME installer’s best game is to install all day — with Quatt as the sales channel and QuattOS as the brain. Addressable: ~3,000 heat-pump installers in NL, mostly 2–10 FTE still doing CV/boilers. We get there with QuattOS + volume.
A wholesale “Hybrid Core” SKU, priced €/kW with channel margin built in — so the installer marks up and still beats the market. As COGS glides €1,100 → €500, the offer only compounds. The full ecosystem to resell: Hybrid · Chill · HomeBattery · All-Electric.
An installer is just a configuration dimension, not a custom build. A partner portal — bulk ordering, job tracking, house creation — on documented endpoints, onboarded in days, installer↔NetSuite wired. They rent the same lean ops that give us our edge: configurator, auto work-prep, scheduling.
The Kraken move, one rung down: the platform that runs Quatt becomes the platform that runs the installers who run on Quatt. Every installer on our rails deepens the data, the buying power and the moat.
We’re not building QuattOS to sell it — the focus is and stays our own operations. But configuration-driven rules per country/product/partner, a clean API surface, modular workflows and a shared portal are exactly the decisions that make a platform licensable. Designing for that optionality costs almost nothing now and is very expensive to retrofit later.
Solar, EV-charging and HVAC installers — and utilities expanding into home energy — all need to orchestrate a complex, multi-system journey with rules that vary by market. They’d rather not build it themselves.
We shouldn’t build a licensable product today. We also shouldn’t make architectural choices that rule it out.
The machine is running. Scale is what we build toward.
The configurator, the AI coach, the email assistant, the dashboards, the voice agent — all live. The platform spine is being built now. The window is 12–18 months. Discipline to keep deleting before automating is what wins it.
Sources, in order of precedence — the Quatt brain (con-quatt-os, Atlas, Pipeline 2.0, the simpler-installation & direct-install models, the installer-channel & B2B-marketplace work, BT&D notes) · Jorge’s QuattOS vision (8 June 2026) · the prior QuattOS architecture document (May 2026). Where they diverged on where state lives and on the durable-execution / build choices, this deck frames them as one migration story and as open calls, respectively.