For providers
Effort reveals intent: the evidence behind detailed project requests
Homeowners who take the time to describe their project are, on average, more serious buyers. What the research shows, where it stops, and why a flat subscription keeps a platform’s incentives on the provider’s side.

- 15–18%
- lower hire rate on effortless job posts
- +10%
- pro win rate when homeowners chose
- 33%
- fewer sign-ups after one extra step
- 0 €
- fee per request on Fynd.lu
The short version
- Effort is a signal. When writing a job post became effortless in a randomised experiment, more posts came in, but from employers less likely to hire.
- In home services, Angi saw the same pattern: when homeowners had to choose their pros, lead volume fell and pro win rates rose by about 10%.
- More effort only helps up to a point. Questions matched to the job capture the signal; long forms and long free text lose good customers too.
- No study has measured this in Luxembourg yet. Fynd.lu is collecting its own data for deeper local research.
Current as of 4 November 2026.
A roofer gets three requests on a Monday morning.
One reads “need roof fixed, how much?”, with no address and no photos. One comes from a homeowner who is collecting prices for a project they might do next year. The third gives the street, the slope, the number of broken slates, a photo of the water stain in the attic and a deadline before the next heavy rain.
All three take time to answer. Only one is likely to become a job.
Homeowners who take the time to describe their project properly are, on average, more serious buyers. The research behind that is real, but mostly indirect, so this article sets out what it shows and where it stops.
The cost providers pay: requests that never hire
For a tradesperson, the most expensive request is the one from someone who was never going to hire. It takes the same time to read, call back and price as a real job, and it pays nothing.
In the United States the problem reached the regulator. In 2023 the Federal Trade Commission finalised an order against HomeAdvisor, which sold leads to contractors for a membership fee plus a fee per lead. The FTC alleged that since at least 2014 the company had misrepresented the quality of those leads, including how likely they were to turn into jobs. HomeAdvisor settled the allegations and agreed to pay up to 7,2 million US dollars.
The lesson is about incentives as much as about one company. When a platform earns money on every lead, more leads mean more revenue, whether or not they turn into work. Volume is easy to grow. Intent is what pays the provider.
That is why Fynd.lu works on a flat monthly subscription with no fee per request. A request that never becomes a job earns us nothing. Our interest is the same as the provider’s: fewer requests that waste time, and more that are worth quoting.
Sources: FTC, January 2023 and FTC, April 2023
The idea: effort is a costly signal
A signal is believable when it costs something to send, and costs a casual browser more than a committed buyer. That is the logic of signalling theory in economics, and answering questions about a project fits it well.
A homeowner who plans to renovate soon already knows the room size, the rough budget, the timing and the constraints. Answering takes little effort and feels useful.
A homeowner who is only curious has to stop and think, and some of them stop there. That drop-out is the filter working.
Economists call such deliberate frictions ordeal mechanisms: small costs that people sort themselves through according to how much they want the outcome. In a field experiment in Indonesia, asking people to travel and apply for a welfare programme produced a better-targeted group of beneficiaries than enrolling them automatically. The same study carries a warning this article comes back to: making the application harder still did not improve targeting further. It only added cost.
There are two reasons why detailed answers can predict hiring, and both help providers. Selection: serious buyers are more willing to answer, so the requests that arrive are more serious. Commitment: answering may itself make a buyer a little more committed.

Evidence 1: when writing a request got easier, intent fell
The clearest causal evidence comes from a randomised field experiment by Emma Wiles and John Horton at MIT. It ran on a large online labour market where employers post jobs and freelancers apply, a set-up much like homeowners posting projects and providers quoting.
Some employers were randomly offered an AI-written first draft of their job post. The versions of the study report:
| Outcome | Effect of making the post effortless |
|---|---|
| Likelihood of posting a job | about 20% higher |
| Time spent writing the post | about 40% lower |
| Total number of hires | no increase |
| Hire rate of posted jobs | 15 to 18% lower, depending on the version |
The extra posts came from employers with lower intent to hire, and the posts were more generic. Before AI, effort and detail moved together, so applicants could read a detailed post as a sign of a serious employer. Removing the effort removed the signal.
This is the provider’s situation in reverse. The effort a homeowner puts into describing their own project is a signal of intent. Removing that effort, with an empty one-line form or a request written for the homeowner by an AI, invites lower-intent requests and blurs the signal. The lesson is not to remove effort, but to make sure it goes into answers that are specific to the job.
Sources: MIT Sloan and MIT DSpace
Evidence 2: costly signals sort serious participants elsewhere too
The same principle has been tested on the other side of online marketplaces. Where it is free to express interest, interest stops meaning anything. A modest cost brings the meaning back.
An availability badge. In a field experiment, freelancers could pay to show an “available now” badge. Those who did were the more reliable ones, employers sought them out, and matching improved. Two years later, badge renters still drew substantially more attention.
Boosted applications. In a related experiment, job seekers could pay to boost an application. The boosted applications came from stronger candidates and were more likely to end in a hire.
Welfare targeting. As above, an application step steered aid towards the households that needed it most.
The common thread: a step that costs a little more for the less motivated works as a filter, and those who pass through it are more likely to complete the transaction.
Sources: Filippas et al., EC ’24, MIT IDE and Alatas et al., NBER
Evidence 3: in home services, one active step raised win rates
The most direct home-services evidence comes from Angi, which runs the HomeAdvisor network. In early 2025 it stopped matching homeowners to pros automatically. A year earlier about 40% of its leads had been auto-matched; by the first quarter of 2025 almost none were. A pro received a lead only when the homeowner actively chose them.
Lead volume dropped: network service requests fell 33% and network leads 57%. But management reported that the pro win rate, the share of leads that turn into a job, rose by about 10%, and that homeowners were significantly more likely to hire a pro they had selected themselves.
These are company-reported figures and they mix several changes, so they are not a controlled test. They still show the pattern from the experiments: one deliberate step by the homeowner cut lead volume sharply and made the remaining leads more likely to convert.
Source: Angi Q1 2025 earnings call
Beyond selection: answering can deepen commitment
Detailed questions may also do more than reveal who is serious. Three findings from consumer psychology suggest that working through the answers can slightly increase follow-through. The effects are small and come mostly from settings outside home services, so they are a secondary benefit, not the main argument.
| Finding | What it shows | Size of the effect |
|---|---|---|
| Question-behaviour effect | Asking people about something they intend to do makes them more likely to do it | Small: d = 0,24 across 116 tests; g = 0,14 in a second meta-analysis, smaller still in better-controlled studies |
| IKEA effect | People value what they put effort into, but only when the effort is completed | Shown in lab studies |
| Foot in the door | Agreeing to a small first step raises agreement to a larger one | Small: r = 0,09 to 0,17 across earlier meta-analyses |
The practical point: a homeowner who has worked through scope, budget and timing has, in a sense, already started the project in their head. Completion matters, so the intake should feel finishable and end with a clear summary of what they asked for.
Sources: Wood et al., 2016, Wilding et al., 2016, Norton, Mochon & Ariely, 2012 and Burger, 1999
Why structured, adaptive questions beat long forms
“More effort is better” is only true up to a point. The research points to three rules.
Ask for answers that decide the quote, not for long descriptions. On a large IT-services marketplace, longer project descriptions drew more bids, but also a larger share of weak bidders, and were linked to a lower chance of the project being contracted. That study was observational, and people who write long posts may differ in other ways. Either way, the useful signal is specific information about scope, budget and timing, not word count.
Do not add friction the job does not need. In the Indonesian experiment, a harder application step cut the number of applicants without improving targeting further. In US health insurance, one small administrative step cut enrolment by 33% and, despite screening people out, made targeting worse overall. In home services the same risk applies to a busy homeowner with a large, valuable project who gives up on a 25-question form.
Match the depth to the project. Unblocking a drain needs four or five answers. A kitchen renovation justifies many more, because a provider needs them to quote accurately. Matching the effort to the size of the job keeps the signal useful and avoids losing serious customers to pointless friction.
A note on conversational AI. In a field study with about 600 participants, a chatbot that asked open questions and followed up drew more informative and specific answers than a standard web survey. A controlled experiment also found less careless answering in a chatbot than in a web survey. The results are not uniform, and a 2024 study that followed respondents over two weeks found no advantage for the chatbot. The design matters: the assistant should interview the homeowner, not write the request for them, which is exactly the shortcut that weakened intent in the Wiles and Horton experiment.
Sources: Liang et al., 2022, Shepard & Wagner, Xiao et al., Kim, Lee & Gweon, 2019 and Zarouali et al., 2024
What it means for a provider’s numbers
For a provider, the number that matters is the cost per job won, not the number of requests. Requests that pass through a well-designed intake improve it in four ways.
Fewer dead ends. If a deliberate step filters out low-intent homeowners, as it did at Angi, a larger share of the requests you receive can become jobs.
Faster, more accurate quotes. With the scope, size, access, timing and budget already known, you can often price without a discovery call, or decline quickly when the job does not fit.
Fewer surprises on site. Structured answers and photos narrow the gap between the job you quoted and the job you find, a common source of disputes and lost margin.
Better choices on your side too. Clear requests let you skip jobs outside your trade instead of finding out the hard way.
A worked example, with purely illustrative numbers rather than platform data:
| Requests | Hire rate | Jobs won | |
|---|---|---|---|
| Many vague requests | 20 | 10% | 2 |
| Fewer specific requests | 12 | 20% | 2,4 |
More jobs, from 40% fewer requests to answer.
How Fynd.lu applies this
On Fynd.lu, a homeowner describes the job to an assistant that asks follow-up questions chosen for that job: what needs doing, the property, access, timing, budget and photos. A drain and a kitchen are not asked the same things, and the number of questions is capped so the request stays quick to finish. The last question is how far along the homeowner is in their decision. The assistant asks; the homeowner answers.
The homeowner then chooses which providers receive the request, and reviews it before sending. That is the same kind of active step that raised win rates at Angi.
What reaches you is the job, the property, the timing, the budget and the customer’s own notes, with a line showing how ready they are, for example “comparing offers · start within 3 months · budget defined · photos attached”. You decide whether to quote, and declining has no effect on your profile.
And because you pay one flat monthly price with no fee per request, we have no reason to send you more requests than are worth your time. See how Fynd.lu works for businesses.
Limits of the evidence
We set out the limits because a provider who checks the sources should find nothing overstated.
No published study tests this exact claim in home services. The strongest causal evidence comes from online labour markets. Angi’s figures are company-reported and combine several changes made at once.
Willingness to answer is partly a trait. Some people are simply more thorough. That does not weaken the signal, because it still predicts who hires. It does mean that forcing everyone through more questions will not necessarily create more hires overall.
The commitment effects are small. The question-behaviour effect is smaller in the best-controlled studies, and the foot-in-the-door effect is modest.
Friction has a cost. Too many questions, or irrelevant ones, lose good customers along with poor ones. Adaptive questioning is meant to manage that, and it has to be tested rather than assumed.
AI can help or hurt. An assistant that fills in the answers for the homeowner removes the very effort that makes the signal useful.
What we are doing in Luxembourg
None of the studies above was run in Luxembourg, and Fynd.lu does not yet have enough data of its own to publish meaningful numbers. We are not going to present early figures as proof.
What we are doing is collecting the data that would answer the question locally: how detailed requests compare with short ones, by trade and by size of job, from the first reply to the job won. We will publish that research when there is enough of it to be worth reading.
Until then, the evidence above is the reason we built the intake the way we did. Your experience of the requests you receive is the other half of the test.
This article summarises published research as of 4 November 2026. Figures from companies are as reported by them. Nothing here is a promise about the number or conversion of requests a provider will receive.
Research sources
- Wiles, E. & Horton, J. — Generative AI and labor market matching efficiency; MIT Sloan summary
- MIT DSpace — Thesis chapter reporting the same experiment (20% more posts, 18% less likely to hire)
- Filippas, A., Horton, J., Parasurama, P. & Urraca, D. — Costly capacity signaling increases matching efficiency (EC ’24)
- MIT Initiative on the Digital Economy — Can costly signals fix hiring?
- Alatas, V. et al. (2013) — Ordeal mechanisms in targeting, NBER Working Paper 19127
- Shepard, M. & Wagner, M. — Do ordeals work for selection markets? American Economic Review 115(3), 2025
- Liang, C., Hong, Y., Chen, P.-Y. & Shao, B. (2022) — The screening role of design parameters for service procurement auctions, Information Systems Research 33(4)
- Angi Inc. — Q1 2025 earnings call (transcript)
- US Federal Trade Commission (2023) — FTC order requires HomeAdvisor to pay 7,2 million US dollars
- US Federal Trade Commission (April 2023) — FTC approves final order against HomeAdvisor
- Wood, C. et al. (2016) — The impact of asking intention or self-prediction questions on subsequent behavior: a meta-analysis
- Wilding, S. et al. (2016) — The question-behaviour effect: a theoretical and methodological review and meta-analysis
- Norton, M., Mochon, D. & Ariely, D. (2012) — The IKEA effect: when labor leads to love, Journal of Consumer Psychology 22(3)
- Burger, J. (1999) — The foot-in-the-door compliance procedure, Personality and Social Psychology Review 3(4)
- Xiao, Z. et al. — Tell me about yourself: using an AI-powered chatbot to conduct conversational surveys
- Kim, S., Lee, J. & Gweon, G. (2019) — Comparing data from chatbot and web surveys, CHI 2019
- Zarouali, B., Araujo, T., Ohme, J. & de Vreese, C. (2024) — Comparing chatbots and online surveys for (longitudinal) data collection, Communication Methods and Measures 18(1)
