The words, not the policy: why a renovation firm's calls stopped converting
A Belgrade bathroom-renovation company had no confirmed orders ahead of it for the first time in months. The calls hadn't stopped — they'd stopped turning into orders. Briefed on a Monday evening, the research came back by Tuesday midday with a finding the founder didn't expect: the problem wasn't his transparent pricing. It was the way he had worded it.
A full pipeline, an empty order book — and a crew depending on it
Six months after launching, the company was running up to five simultaneous projects. Then the conversions quietly stopped. Enquiries still arrived: prospects asked questions, booked inspections, deferred, and never came back. Behind the order book were plumbers, tilers and finishers the founder had spent months finding and onboarding — if the schedule didn't refill, he'd be telling reliable people there wasn't enough work.
He had deliberately rejected the category's "complete renovation from €999" convention. If a real job costs several thousand euros, he reasoned, say so before someone spends their time calling. The question the research had to answer was whether that transparency was costing him the business — or something else was.
The price wasn't filtering anyone. It was generating phone calls.
The panel: 72 True Twins™ across all four Serbian regions (Vojvodina 22, South & East 18, Belgrade 17, Šumadija & West 15), average age 45, each given a standard 3–4 m² Belgrade bathroom scenario and shown a recording of the live mobile site.
Transparency was working. The wording was undoing it.
Respondents liked exactly what the founder intended them to like — real jobs, itemised costs, no deposit, a price agreed upfront. But the headline figure at the top of the page and the detailed examples further down described different things, and nobody reading it for the first time could tell. One respondent named the contradiction precisely.
Both reactions came from the same page. The policy was an asset; the phrasing around it was the liability — and six months of pricing bathrooms daily had made the distinction invisible from the inside.
Why the campaign was teaching itself to deliver bad leads
The deeper issue wasn't any single line of copy — it was a self-reinforcing loop. Because every phone call registered as a conversion, the bidding algorithm was being rewarded for producing exactly the enquiries that never converted, and went looking for more people like them.
Every call counts as a win — so the machine optimises for calls
Result: a reinforcing loop. The system learns to deliver precisely the low-quality calls the business is trying to avoid.
Qualify before the call — and change what counts as success
The lever: not the ad copy — the success signal. Feed the algorithm qualified enquiries and it starts learning from buyers the offer actually fits.
Tested head to head, the risk-free ad beat the price-led one decisively (8.6 vs 4.9 on a 10-point scale) — but it lifted volume across all segments rather than filtering. That's why the fix had to move the filter earlier, onto the page and the form, instead of relying on the ad.
A sequenced plan, with every action tied to evidence
Beyond the findings, the study included a strategy layer — each action labelled as either a direct finding from the data or a hypothesis to be confirmed by a test, and sequenced so the cheapest, highest-certainty fixes came first.
Remove the contradiction
Prove the bigger bets
Change what you measure
He changed the wording, not the principle
The transparent pricing stayed: real completed-project costs, first available slots, the commitment to agree the price upfront. Only the presentation changed — reworded so first-time readers understood it the way it was meant. What followed came fast, and the founder is careful about what it does and doesn't prove.
"The calls hadn't disappeared. They had simply stopped turning into orders. Potential customers would ask questions, arrange inspections, defer decisions and never get back to us."
"The research suggested that the problem wasn't transparency. It was the way I had expressed it. Some of the pricing formulations that seemed perfectly clear to me — after spending six months dealing with more than 25 bathroom renovations — appeared inconsistent or potentially misleading to people encountering them for the first time."
"Yesterday, the first relevant call in weeks converted into an order that same evening. Today, two more relevant enquiries came in. One converted into another order this evening; the other customer is discussing our offer with their family after today's inspection. All of this happened despite a significant fall in relevant Google search volume."
"What I take from this is fairly simple: what you think you are communicating and what your customer actually understands can be two very different things. For a business owner who spends every day inside their own product, that distinction can become surprisingly difficult to see."
"What happened next is obviously too small a sample to establish causality, but the timing was striking." We agree, and we'd rather say so: two orders in two days is an encouraging signal, not proof. The research itself flags the same discipline — its recommendations are explicitly split into findings that follow from the data and hypotheses still to be confirmed by A/B tests in the live account.
What a small business can take from this
The curse of knowledge is a revenue problem, not a style problem
Six months and 25+ renovations had made distinctions obvious to the founder that no first-time reader could see. Nothing about his pricing policy was wrong — the words carrying it were doing the opposite of what he intended. This is the single most common failure mode for an owner-operated business, and it is invisible from the inside by definition.
A price only filters if the reader trusts it's the real total
Searchers arrived with a reference price roughly €1,500 below the real bill, so a headline figure didn't screen anyone out — it created a question to be settled by phone. Publishing a number isn't transparency; publishing an unambiguous total that a stranger can reconcile with the examples is.
Fix the success signal before you rewrite the ad
The highest-leverage change wasn't copy at all — it was what the campaign counted as a win. While every call scores as a conversion, better ad text simply buys more of the wrong calls faster. Qualified-enquiry tracking changes what the system optimises toward; everything else is downstream of that.
Give people a way in that isn't a phone call
Half the sample would rather submit details than call — and a short form asking for budget does the qualifying the headline price was failing to do. What reassures is concrete and mundane: no deposit, a free inspection, a fixed agreed price, one contractor covering every stage.
An outside view of an offer he'd stopped being able to see
For less than the cost of a week of misdirected ad spend, a company of this size got customer research it could not realistically have commissioned through traditional polling — and got it fast enough to act on before the schedule emptied. The fix cost nothing to implement: the pricing policy stayed exactly as it was, and only the wording changed.
It took less than a day to identify a problem I had been looking at for weeks without seeing it.
Research a company this size couldn't otherwise commission
Atani True Twins™ are individual AI models, each built from a real, recruited respondent. For a small business with a live commercial problem, that meant proper customer research at a speed and price that made sense:
Overnight, not next quarter
Brief on Monday evening, findings by Tuesday midday — fast enough to act while the order book was still emptying, which is the only speed that mattered here.
Reactions to the real thing
Twins responded to the actual ad variants and a screen recording of the live mobile site — not a description of the offer, but the offer as a customer meets it.
Outside eyes on an over-familiar offer
The core value wasn't a statistic. It was a first-encounter reading of copy the owner had been staring at for six months and could no longer evaluate objectively.
Findings plus a plan
At the client's request the report carried a strategy chapter: each action tagged as finding or hypothesis, sequenced into do-now, test, and monitor.
We fielded a 15-question instrument to 72 validated True Twins™ across all four Serbian regions. Each respondent was given the same scenario — a complete renovation of a standard 3–4 m² bathroom in Belgrade — then shown the company's real ad variants and a recording of its live mobile site, and asked about expected costs, trust, fears, and what they would do next.
Closed answers (ratings, rankings, choices) were extracted programmatically. Open-ended answers were coded by LLM reading across 33 dimensions — attitude, trust, fears, price perception, site reactions, verification behaviour, budget fit. Three budget segments were derived from the responses rather than from declared income. Every quote was validated against its source response before inclusion.
This was a national panel, not a Belgrade-only one: 17 of 72 twins were from the Belgrade region, and Belgrade purchasing power and prices are likely above the sample average. The answers are stated intentions — click and booking ratings given in a questionnaire, without competing ads, a live auction or a real checkout — so they indicate direction and should be confirmed by A/B tests in the live account. Some items drew near-unanimous answers; those are best read as a strong signal rather than an exact percentage. The three budget segments were derived by coding, not declared income.
We hold no identity data. Our AI twins are built from anonymised interviews conducted by an independent research agency, processed under strict data separation, and governed by a Code of Ethics that covers every client, every use case, and every prompt. GDPR-compliant by design.