When Revenue Growth Hides Bad Unit Economics: Why Hometree Stopped Scaling Boiler Installations
Short answer: Stop scaling when each additional sale produces negative contribution after the acquisition, fulfilment, failure and servicing costs caused by that sale. Give the offer no more than two defined improvement cycles, each with a deadline and a target large enough to turn contribution positive. If the economics still fail and there is no credible repeat-purchase value to recover the loss, reduce volume or close the offer even while revenue is rising.
Revenue can rise because you are buying work faster than you are measuring its cost. That is particularly dangerous in installation, ecommerce and marketplace businesses, where cash arrives before defects, returns, callbacks and support expose the full economics.
My view is blunt: growth is not a reason to tolerate negative unit economics. It is a multiplier. Unless you are deliberately funding a measured route to positive contribution, faster growth makes the mistake larger and leaves you less time to correct it.
Use the Growth Without Contribution Stoplight
The Growth Without Contribution Stoplight tests an offer at three levels: the unit, the customer and the cash system. A green result permits controlled growth. Amber requires a time-limited repair. Red means stop adding volume.
| Test | Green | Amber | Red | |---|---|---|---| | Sale contribution | Positive after all variable costs | Slightly negative with one evidenced fix | Loss grows with every sale | | Acquisition recovery | Recovered from first sale or contracted repeat revenue | Recoverable within a funded, observed period | Depends on hoped-for future purchases | | Quality exposure | Defects and callbacks measured and affordable | Data incomplete but capped | Unknown or increasing with volume | | Working capital | Growth releases or consumes affordable cash | Funding covers the next two cycles | Orders create an unfinanced cash gap | | Operational capacity | Delivery quality holds as volume rises | One constrained process has a dated fix | Backlog and failure rise together |
You do not average these results into a comforting score. A red unit-contribution result is enough to stop scaling. A profitable average can also hide a red customer segment, postcode or acquisition channel, so calculate each separately.
What Hometree's installation decision reveals
Hometree originally sold boiler installations directly to consumers. Founder Simon Phelan later said the operation grew beyond £1 million of monthly revenue and completed high hundreds of installations each month. Revenue was real. So was the scale.
The problem was acquisition. In a founder interview about Hometree's early business model, Phelan described customer-acquisition costs as exorbitant and said the economics did not work. About 20 months in, the company stopped pushing the installation model and directed capital towards recurring home-cover subscriptions. His reasoning was structural: local tradespeople had an advantage in winning installation work, while Hometree had been paying heavily for each transactional customer.
The public account does not disclose the cost of an installation, gross margin or acquisition cost, so it cannot prove the loss per job. It does show the decision pattern. The team separated an attractive revenue line from a scalable economic engine and stopped treating volume as validation.
Do not copy Hometree's destination without its evidence. A recurring model is not automatically superior. It works only if retention, claims, servicing and acquisition leave positive customer contribution.
Calculate the contribution of one sale properly
Start with net revenue, excluding VAT or equivalent sales tax that you collect for the state. Subtract every cost that occurs because you made and fulfilled that sale:
- materials or product cost;
- delivery labour and employer costs;
- subcontractor payments;
- payment and finance fees;
- travel, packaging and delivery;
- expected returns, warranty work and callbacks;
- sales commission and customer-acquisition cost.
The result is contribution before fixed overhead and tax. Rent and permanent management salaries still matter to overall profit, but adding them too early can obscure the operating decision. The immediate question is whether one more sale contributes cash towards those overheads or consumes more cash.
Use expected failure cost rather than waiting for every defect. If 5 jobs in 100 require a £240 callback, allow £12 per job. Replace that estimate once your own data improves.
Worked example: CedarHeat Installations
CedarHeat Installations fits domestic boilers around Bristol. Its dashboard celebrates £180,000 of monthly net revenue from 60 jobs at an average £3,000 each. The owners want to double advertising because revenue has grown for four months.
They calculate one job properly:
| Item per installation | Calculation | Amount | |---|---:|---:| | Net revenue | Contract price excluding VAT | £3,000 | | Boiler and materials | Supplier invoices | £1,620 | | Engineer and subcontract labour | Direct delivery cost | £610 | | Survey, travel and disposal | Direct cost | £145 | | Payment and finance fees | Average per job | £75 | | Expected callback cost | 8% × £250 | £20 | | Customer acquisition | Advertising and sales cost per completed job | £620 | | Contribution | £3,000 minus all costs above | -£90 |
At 60 jobs, the business loses £5,400 of contribution before paying office rent, management salaries, software, insurance or tax. Doubling to 120 jobs at the same economics would double that operating loss to £10,800 and probably increase working-capital pressure.
CedarHeat tests two repairs for six weeks. It narrows advertising to postcodes where completed-job acquisition costs have historically been lower and raises the quoted price by £180 on the next 20 suitable enquiries. After the test, average revenue is £3,150 and acquisition cost is £470. Other costs remain £2,470, producing £210 contribution per job: £3,150 minus £2,470 minus £470.
At 60 jobs, that is £12,600 towards fixed overhead, a £18,000 monthly improvement from the previous result. CedarHeat can now consider controlled growth, but only while callbacks, lead prices and delivery capacity remain within the tested range.
These figures are illustrative. Use your actual costs and obtain qualified accounting and tax advice for your treatment of VAT, labour and provisions.
Distinguish a repairable execution problem from a structural one
An execution problem has a mechanism you can change and observe quickly. Examples include poor lead qualification, a supplier surcharge, avoidable travel or a quoting error. Set one target, one owner and one deadline. “Improve marketing” is not a test. “Reduce paid acquisition cost per completed job from £620 to £480 within six weeks while keeping cancellation below 10%” is.
A structural problem persists even when competent execution improves. You may face buyers who purchase once, powerful local competitors with cheaper acquisition, supplier minimums that prevent margin, or service complexity that rises with distance. A large addressable market does not remove those disadvantages.
Practitioners disagree about whether an early company should accept losses to buy growth. Venture-backed businesses sometimes do so to establish network effects or recover acquisition cost through contracted recurring revenue. Bootstrapped operators usually cannot. My view is that losses are defensible only when the recovery mechanism is already visible in cohort data and cash is explicitly budgeted for the test. A story about future scale is not a mechanism.
Set a stopping rule before the next sales push
Choose the red line while you are calm. Define the maximum cumulative contribution loss, the test end date and the minimum improvement required. Include a cash threshold that preserves enough runway to close or change the offer responsibly.
For example: pause paid acquisition if completed-job contribution remains below £150 after 40 more jobs, or if unrestricted cash falls below four months of fixed costs. The exact thresholds depend on your business, obligations and funding. Their value is that they prevent a rising revenue chart from moving the goalposts.
When you stop, protect existing customers first. Honour warranties, explain changes and reserve the cash needed for outstanding work. Closing an acquisition channel is not permission to abandon delivery.
What to do in the next seven days
By tomorrow, choose one offer and calculate contribution for its last 20 completed sales. Use invoice-level costs, staff time and actual acquisition spend. Within three days, split the result by channel, customer type and area. Identify whether one segment creates the loss.
On day four, write one repair test with a numeric target, maximum spend and end date. Freeze expansion outside that test. At the end of the week, set the stopping rule with whoever controls cash. If the existing evidence is already red and no specific repair can turn it green, stop buying more volume now and plan the customer-safe exit.
Related guides
Frequently asked questions
Can an unprofitable sale still be worth making?
Yes, but only when you can identify and measure the later value that recovers the initial loss. A discounted first order may be rational if observed repeat purchases reliably produce enough contribution within an affordable period. The same applies when one product leads contractually to profitable servicing. Do not use total company growth as evidence. Track the original customer cohort, its later revenue, direct costs, refunds and churn. If recovery depends on an uncontracted purchase that most customers never make, treat the first sale as loss-making and price it accordingly.
How many months should I give a bad model to improve?
Use operating cycles, not a universal month count. A weekly subscription can generate evidence in weeks, while installations with seasonal callbacks may require longer. Give the model enough completed transactions to test the proposed repair, but cap both time and cash in advance. Two focused test cycles are normally more informative than six months of undirected persistence. If the same loss mechanism remains after both, stop. A longer runway does not turn weak economics into evidence. It only lets you fund more learning, which must still be specific.
What if gross margin is positive but the business still loses money?
Check whether your gross-margin definition excludes acquisition, fulfilment or servicing costs that rise with sales. Accounting presentations differ, so a positive gross margin does not automatically mean each new customer helps. Calculate contribution using every variable cost caused by the sale, then compare total contribution with fixed overhead. If contribution is positive, more suitable volume may eventually cover overhead, provided capacity holds. If contribution is negative, volume worsens the loss. Ask your accountant to reconcile your management calculation with statutory accounts rather than assuming the labels mean the same thing.
Should founder time be included in unit economics?
Yes, when the founder performs work that would have to be paid for at scale. Record the hours and apply a realistic replacement cost, even if you take no salary today. Otherwise a service can appear profitable only because your labour is free. Keep actual cash reporting separate so you can see both survival and sustainable economics. Founder time spent on genuinely fixed leadership work may sit in overhead instead. The test is whether another unit creates more of that work. If it does, include an allowance in the unit calculation.
Can I fix negative contribution simply by raising prices?
Sometimes. Quote the higher price to a controlled group and measure completed sales, not acceptance anecdotes. A price increase works when the lost volume is outweighed by added contribution on the customers who remain. It fails when the offer becomes uncompetitive, sales effort rises sharply or higher expectations increase delivery cost. Compare total contribution for the test group with the previous group. Do not judge success by win rate alone. A lower win rate with materially better contribution and manageable capacity can be the stronger business.
What if investors want revenue growth despite poor economics?
Show the unit, cohort and cash evidence and ask them to approve a bounded loss budget rather than an open-ended growth target. Investors may rationally accept losses when there is a credible recovery mechanism, but surprise runway reductions damage trust. Present the proposed test, expected learning, maximum cash exposure and stopping rule. Directors also have legal duties that vary by jurisdiction, especially as solvency weakens. Obtain qualified legal and accounting advice for financing, director duties and any decision that could affect the company's ability to meet obligations.
Does stopping one offer mean the whole business has failed?
No. It means one combination of offer, customer, price and acquisition route has failed its economic test. Preserve any assets that remain useful: customer insight, supplier knowledge, trained staff, technology or a profitable segment. Hometree's public account describes redirecting investment from transactional installations towards recurring cover, not abandoning its wider purpose. Do not force a pivot merely to preserve the company name, though. The alternative still needs its own demand and unit-economics evidence. A failed offer is information, not automatic proof that the next model works.
Comments (0)