The opening went well. Friends came, the first week was busy, and the team is tired but happy. By week three it is quieter. Now every option looks equally urgent: more ads, a second product, a delivery app, a bigger launch offer, a new branch.
The first 90 days are like the first season in a new garden. A few seeds are planted on purpose, and the job is to watch which ones grow before deciding where to dig more. The trouble is that the launch week makes everything look like it is growing. This article is a way to tell the difference, week by week, and to arrive at day 90 with one clear thing to scale.
In short: The first 90 days are for learning, not scaling. Tag where every customer comes from, run two or three deliberate tests with end dates, and watch one number above all others: how many customers come back. At day 90, put more effort only behind the source, offer and audience that bring returning customers. The launch spike is not a signal; repeat behaviour is.
Why the launch spike misleads
The first weeks are full of people who would come anyway: friends, family, the curious, bargain-hunters drawn by an opening offer. They make the numbers look healthy and tell very little about the people who will keep the business alive. A good launch week and a weak business can look identical until about week six.
Eric Ries's The Lean Startup (2011) calls the alternative validated learning: treating each early activity as an experiment with a question, a measure and a decision, in a loop of build, measure, learn. He also warns against what he calls vanity metrics, numbers that go up and feel good but do not guide a decision. Followers, views and total sign-ups in launch week are the usual examples.
A weekly 90-day scorecard
These are the numbers worth writing down every Monday. Five rows are enough. Rough numbers in a notebook work better than a perfect dashboard that nobody opens.
| Number | Why it matters | Good sign | Warning sign |
|---|---|---|---|
| New customers, by source | Shows which channels bring people | One or two sources stand out | “Unknown” is the biggest source |
| Return rate, by start week | Shows if the product keeps people | Steady or rising for newer weeks | Falls with every new week |
| Questions before buying | Shows what is unclear | Fewer repeats of the same question | Same question every day |
| What first-time buyers choose | Shows the real entry product | A clear favourite emerges | Only the discounted item |
| Cost per returning customer | Joins spend to loyalty | Lower in one channel | Cheap leads that never return |
The second row is the one that matters most. Group customers by the week they first came, then check how many of each group are back four weeks later. If newer groups return as well as or better than older ones, the business is learning. If each new group returns less, something is being promised that the product does not deliver.
A small example shows how to read it. Suppose 40 customers first came in week one, and 12 of them were back four weeks later: 30%. Of the 40 who first came in week four, 8 were back: 20%. These numbers are illustrative. The drop matters more than either figure: the week-four customers are coming for something that the week-one customers did not need, perhaps a bigger discount or a different promise.
One caution: small groups move by luck. With only a few dozen customers per group, a difference of two or three people means very little. Treat it as a direction to check in the conversations, not as a verdict.
The 90 days, step by step
- A one-page learning plan, before opening. Write the two or three questions the first 90 days must answer, for example: which channel brings people who return, which entry offer leads to a second purchase, which audience values us most. Output: one page with the questions and the number that will answer each.
- A source tag on every customer, from day one. One field in the booking form, the till or the WhatsApp label: where did this person hear about us? Output: a customer list that can be sorted by source.
- Two or three test cards, in weeks one to four. Each test has a question, what changes, a start and end date, and the number that decides it. Change one thing per test. Output: a card per test, written before it starts.
- A Monday review, every week. Fill in the scorecard, read the test cards, and write one decision: continue, adjust or stop. Output: twelve short entries by day 90, each with a decision.
- Twenty conversations, around weeks five to eight. Ten customers who came back and ten who did not. Ask about what they did, not what they think: what made them try, what they compared, what brought them back or what stopped them. Output: a page of repeated reasons, in customers' own words.
- A cut at day 60. Stop the activities that bring no returning customers, even if they bring traffic. Output: a shorter list of what continues, and freed time or budget.
- A scale decision at day 90. Name one source, one offer and one audience to grow, the number that should rise, and what would make you stop. Output: a one-page plan for the next quarter.
Three tools that make the steps work
Source tags that people actually fill in
A tag only works if it is filled in every time, and that depends on making it easy. Three habits help.
- Fixed choices, not free text. Offer five or six options at booking or at the till: Instagram, a friend, the clinic next door, passing by, search, other. Free text gives “insta”, “Instagram” and “IG”, which then have to be cleaned by hand.
- One owner. Someone on the team is responsible for the field, and the Monday review includes a check of how many customers have no source. If many do, fix the habit before reading anything else.
- A name for word of mouth. When the answer is “a friend”, ask who. It credits the person who sent them and shows you who your best referrers are.
A code or a QR link for each channel is a useful second layer for ads and partners, because it records the source without anyone having to remember. It does not replace the question, since many customers arrive after seeing two or three things.
A test card
A test card is half a page, written before the test starts. It prevents two common problems: changing the rules after seeing the result, and running a test with no end. It needs only five lines.
- Question: what are we trying to find out?
- Change: the one thing that will be different.
- Dates: when it starts and when it ends.
- Deciding number: the single figure that answers the question.
- Rule: what we will do if the number goes up, stays flat or goes down.
Here is an illustrative card for the Pilates studio in the example below. Question: does a three-class starter lead to more class packs than a single discounted class? Change: replace the single discounted class with the starter, for four weeks. Deciding number: the share of new customers who buy a pack within two weeks of their first class. Rule: if the share rises, keep the starter; if it stays flat, keep the cheaper option and test the message instead; if it falls, go back to the conversations. On a real card the numbers come from your own history, not from ours.
The twenty conversations
The conversations in weeks five to eight are the cheapest research in the whole plan, and the one most often skipped. Ten customers who came back and ten who did not is usually enough to hear the same reasons repeat. A few rules make them useful.
- Ask about the past, not the future. “What did you do before you found us?” tells you more than “Would you come again?” People are poor at predicting their own behaviour. Rob Fitzpatrick makes the same point in The Mom Test (2013): ask what people did, not what they would do.
- Ask four things. What were you looking for? What else did you compare? What almost stopped you? What brought you back, or what stopped you coming back?
- Let someone else ask. A customer is rarely fully honest with the person who served them. A teammate from another branch, or an outside helper, works better.
- Write down their words. Copy the exact sentences. The phrases that repeat become the wording of your next message.
For the ten who did not return, a short message is often enough: “We would like to improve. May we ask you three quick questions?” Do not offer a discount in the same message, because it changes the answers.
A worked example
This example is hypothetical, and its numbers are illustrative. A new reformer Pilates studio opens in Dubai, in a neighbourhood of offices and apartment towers. It launches with a discounted first class, Instagram ads, a partnership with a nearby physiotherapy clinic, and a lunchtime offer for one office tower.
Weeks one to four. The discounted first class fills the schedule: around 120 trial bookings, most from ads. The team wants to double the ad budget. But the source tags tell a different story. Of the trial-class people, only a small share buy a class pack. The physio clinic sends only about fifteen people, yet most of them buy a ten-class pack. The office offer brings a steady group at lunchtime.
Weeks five to eight. The conversations explain why. Trial-class people say: “It was cheap, I wanted to see it.” Physio referrals say: “My back needs this, and the clinic trusts you.” Office regulars say: “It fits between meetings, and I go with colleagues.” Two different promises are at work: a fun workout, and a safe way to fix a problem. The second one keeps people.
Day 60 and day 90. The cheap single trial is replaced by a three-class starter that leads into a pack. Ads continue at a smaller level, now speaking about posture and back health. At day 90 the scale decision is clear: grow referral partnerships with clinics, and offer the lunchtime format to two more office towers. The number to watch next quarter is packs renewed. Nothing else needs to grow yet, and the summer slowdown that Dubai sees every year is planned for rather than discovered.
Three outcomes at day 90
Day 90 is a decision point, and it usually points to one of three places.
| Outcome | What you see | What to do |
|---|---|---|
| Scale | One source, offer and audience bring customers who return | Put more effort behind exactly that, and keep measuring the same number |
| Repair | Customers come, few return, and the conversations point to a specific gap | Repair that gap first, then run the same tests again for four to six weeks |
| Rethink | Few customers return from any source or offer | Question the product, the service or the price before spending more on marketing |
The “repair” row is the one teams like least, and the cheapest to act on. A gap such as slow replies, an unclear first visit or a confusing price usually shows up in the conversations, and fixing it often lifts every source at once.
Five mistakes in the first 90 days
- Scaling the launch spike. It happens because the first busy weeks feel like proof. They are mostly friends and the curious. Wait for return rates by start week before spending more.
- Not tagging sources. Without one field per customer, every later decision is a guess. It costs nothing on day one and is nearly impossible to rebuild later.
- Tests with no end date. A test that runs “until we see” never ends and never decides anything. Every test card has a date and a number written before it starts.
- A deep opening discount. It brings people who came for the price and sets a reference that makes the normal price look expensive. A starter that leads into the real product works better than a cheap one-off.
- Adding products or branches to fix a quiet week. New things reset the learning. Fix what the conversations point to before adding anything.
When 90 days is the wrong frame
Some businesses learn more slowly. B2B companies with long sales cycles may need two or three quarters before return behaviour shows. Seasonal businesses should judge against the season, not the calendar: a launch in Ramadan or just before a Gulf summer needs its own baseline. And if very few customers return across every source and offer, the problem is probably the product, the service or the price, not marketing. More reach would only bring more people to the same disappointment. We say so when we see it.
How OG supports the first 90 days
Ideally we start before opening, with launch and market entry strategy: the learning plan, the questions, the entry offer and the channels to test. Analytics and conversion set up source tagging and the weekly scorecard so the numbers exist from day one. During the 90 days, customer research runs the conversations with returning and lost customers. At day 90, the scale decision shapes what comes next: social media ads or wider media for the source that works, and WhatsApp or email for bringing customers back. Scope follows the need; there are no fixed packages.
The strongest objection
“If we do not push hard now, we lose the launch momentum.” Some visibility is needed, or there is nothing to measure. But Les Binet and Peter Field's The Long and the Short of It (IPA, 2013) separates two jobs: short-term activation, which brings sales now and fades fast, and longer-term brand building, which grows slowly and lasts. A loud first month is almost all activation. Steady presence plus fast learning keeps the doors busy and builds something that is still there in month six. Momentum that depends on a discount is not momentum; it is a rented crowd.
A second objection: “We cannot wait 90 days to find out.” You do not wait; you decide every Monday. What takes weeks is return behaviour, because a customer cannot come back before enough time has passed. Plan the cash for that gap, so that one slow week in the middle does not force a panic decision.
Built on: our launch work with new businesses since 2013; Eric Ries, The Lean Startup (2011); Les Binet and Peter Field, The Long and the Short of It (IPA, 2013); Rob Fitzpatrick, The Mom Test (2013). The Pilates studio is a hypothetical example.
Before closing this page
- What are the two or three questions the first 90 days must answer?
- Is every customer tagged with where they came from?
- Of the customers who came in the first month, how many came back?
- Does each running test have an end date and a deciding number?
- Which single source brings the customers who return?
