Primeiro Saque

Product case study · São Paulo · 2026

A fake doorworth opening

Testing whether tennis players in São Paulo would pay to try a BRL 2,000 racket for a week before buying it — without buying a single racket first.

Fake-door MVP Demand validation Live since Apr 2026 Still running
38
Reservation requests
R$ 0
Paid media spend
100%
Organic acquisition
0
Rackets purchased
01 — The problem

A two-thousand-real
decision, made blind

A serious amateur player in Brazil buying a new frame is making a BRL 1,500–2,500 decision with almost nothing to go on.

Shops let you hold the racket. Holding a racket tells you nothing about how it plays. The things that actually decide whether a frame suits you — string pattern, swing weight, balance, how it behaves in the third hour when your timing goes — only surface across several sessions on a real court.

Demo programmes solve this in the US and Europe. In São Paulo they are rare and informal, usually limited to whatever one shop happens to stock. So players decide on brand loyalty, a coach's opinion, or which pro endorses the frame.

Get it wrong and you lose the money twice — once on the frame, and again on every match you play with it.

I play tennis. I have watched this happen to people, and I have done it myself. That is where the idea came from — not from a market report.

02 — The hypothesis

Four bets,
stated separately

Vague hypotheses produce vague results. Splitting the idea into four claims meant each one could be probed — or shown to be untestable by this experiment, which turned out to matter just as much.

H1 · Problem

The pain is financial, not athletic

Players feel racket choice as money at risk, not as a performance or injury problem. So the page leads with the cost of being wrong.

Probed · single framing only
H2 · Solution

A week on court is the fix they recognise

A one-week, delivered, court-ready trial is legible enough that people will raise their hand for it without explanation.

Probed · 38 requests
H3 · Pricing

BRL 250 a week doesn't kill it

Around 12% of the purchase price is acceptable to buy down the risk — and tiering pushes people toward comparing several frames.

Partial · nobody was asked to pay
H4 · Acquisition

Search intent finds them cheaply

Someone searching for a specific racket review is already inside the decision window and can be reached affordably.

Untested · campaign never launched
03 — The MVP

One page that
behaves like a business

A single landing page presenting Primeiro Saque as an operating service: four real rackets with real specs, real prices, and a reservation flow that ends in a spreadsheet row.

Catalogue
Four real frames
Babolat, Wilson, Head, Yonex — actual models, actual specs, the frames people are already deciding between.
Offer
One week, delivered
Fresh strings at your tension, new overgrip, delivery and pickup included across São Paulo.
Pricing
250 / 450 / 600
One, two or three rackets. Shown before the form opens, with a live total as you select.
Request
Five questions
Name, WhatsApp, which frames, how soon, which neighbourhood. Nothing else earned its place.

The last two fields are the ones that make this an experiment rather than a lead form. Urgency is the only available proxy for real purchase intent — it separates someone mid-decision from someone browsing. Neighbourhood decides whether the business is even physically possible: delivery and pickup are included in the price, so if requests scatter across São Paulo instead of clustering, the logistics eat the ticket.

Everything the page could have promised and didn't is deliberate. No contact window. No named channel. No "only 3 spots left". The button says request, not book, because nothing is being booked.

04 — Why we didn't build more

The cheapest credible
version of the question

A working version of this business needs inventory, stringing, city-wide logistics, payments, deposits against damage, and someone to run it. Call it BRL 15,000–30,000 and a few months — all committed against an assumption nobody had tested.

The riskiest thing was never whether it could be built. It was whether anyone wanted it.

So none of this got built:

InventoryBRL 1,500–2,500 per frame, several per model to cover concurrent rentals
LogisticsTwo trips per rental across São Paulo — possibly margin-negative, still unknown
PaymentsNo checkout, no Pix, no card. Nobody has been charged anything
AccountsNo login, no profiles, no password to lose
A backendA spreadsheet behind a 60-line script. At 38 rows, that is the right tool
A frameworkOne HTML file. A copy change stays a one-line diff

The constraint that drove every technical decision: nothing may make a copy change slower. During validation you rewrite headlines, not architecture. Total running cost, five months in: a domain.

05 — The experiment

What happens when
you knock

Step 01
Landing page
Static HTML on a CDN. Visitor arrives from organic search.
Step 02
Reservation form
Rackets selected, price calculated live, request submitted.
Step 03
Apps Script
Validates, escapes, appends. Sixty lines standing in for a backend.
Step 04
A spreadsheet row
The system of record. Pivot tables included, free.
Not built
Everything after
Fulfilment, payment, delivery. The door does not open yet.

GA4 runs alongside, recording traffic and a conversion event carrying the value of each request. It never touches the contact data — traffic analytics and lead data stay in separate systems with no shared identifier.

Where the design is uncomfortable

A fake door tests whether people will walk toward it — it does not entitle you to leave them standing there. Real people submitted real contact details expecting eventual contact. The success message was written to be literally true, and no timeframe or channel was ever promised. That is mitigation, not absolution, and closing this loop is the first job of the next experiment.

The result · 14 April – September 2026

38
Organic reservation requests
Zero paid media · zero outbound · zero rackets owned

Nobody was driven to this page. No campaign ran, no ads were bought, no lists were emailed. People searching for rackets found it, read it, and asked for something that does not exist yet — at prices they could see before they asked.

07 — Reading the number honestly

What 38 means —
and what it doesn't

This is the part where product work usually goes wrong. A number that flatters you is the easiest thing in the world to over-read.

What it means

Supported by the data
  • The problem is recognised without explanation. Nobody had to be taught why testing a racket first is worth something.
  • The offer is legible. People understood a one-week trial well enough to choose specific models.
  • Organic search reaches this audience at zero cost — the page was found, repeatedly, unassisted.
  • Price transparency did not stop them. Everyone who submitted had already seen BRL 250.
  • Demand is non-zero. That was the actual open question, and it now has an answer.

What it doesn't mean

Not supported — do not claim these
  • Not 38 customers. Nobody has been charged, quoted or served. There is no way to accept money.
  • Probably not 38 people. The sheet contains repeat submissions from the same contacts.
  • Not willingness to pay. Raising your hand is free. Every unknown below traces back to this.
  • Not a conversion rate. There is no modal-open event, so drop-off before the form is invisible.
  • Not a clean sample. The live page carries fabricated testimonials — an unmeasurable influence on this data, and the experiment's worst decision.
  • Not even a firm 38. The form posts blind, so failed submissions look identical to successes. It is a floor.
The experiment ran for five months without a success threshold. So 38 cannot be scored as a pass or a fail — only described.

That is the most useful thing this project taught me, and it is a methodological failure, not a win. Any bar written now would be reverse-engineered from a number I already know. The next experiment defines its criteria before the first call is made.

08 — Still unknown

The gap between
interest and a business

Would anyone pay?
The entire hypothesis. Nobody has been asked for money, so nobody has said no.
What does delivery cost?
Two trips across São Paulo against a BRL 250 ticket. Possibly margin-negative.
How often does a frame rent?
Utilisation sets the capital requirement for the entire model.
What breaks, and how often?
Damage, loss and non-return are unpriced. One bad week erases many good ones.
What does a customer cost?
CAC under paid media is unmeasured — the acquisition bet was never run.
Does anyone come back?
Someone who buys a racket may not need another trial for years.
09 — The next experiment

Ask them for
the money

The fake door has given everything it can. Every remaining question needs someone to reach for their wallet.

A paid, closed-loop pilot

Hypothesis: a meaningful minority of the people who requested a trial will pay BRL 250 up front for a real one.

Small on purpose. Two to four frames from the most-requested models, one neighbourhood cluster, four to six weeks. Inventory bought only after money arrives — low thousands of BRL, recoverable by resale.

  1. Resolve the fabricated testimonials before collecting another data point.
  2. Add the missing funnel event, so drop-off stops being invisible.
  3. Write the success criteria down — conversion bar, maximum delivery cost, stop date — before the first call.
  4. Contact the existing respondents, starting with those who said "this week". Be straightforward about what happened.
  5. Take payment first. Buy frames second. Deliver third.
  6. Log every decline and, more importantly, every reason for declining.

If they say no at BRL 250, the honest reading is that this measured curiosity rather than demand — and that is a genuinely valuable finding, bought for the price of a few rackets instead of a business.