NETTOCase study and numbers · Terms and dataRUENKZUADEFRESITPTPLNLTR

NETTO case study: numbers, periods, how we calculated

Updated 27.09.2026. Data through 29.09.2026.

All numbers come from one education center in Astana in 2026. NETTO grew inside this center, and the owner allowed us to show the numbers. NETTO has no other clients yet: we started by setting it up in one center and are now turning a working system into a product.

How well a number is proven

1. Same daily budget: before and after the switch Before and after

The main comparison. When the budget is cut, return on ad spend can rise on its own: the cheapest audiences remain. So we compare periods with the same daily spend. From 17.08 the center had already cut the budget to the September level but was working the old way. On 31.08 the center switched to the new setup: budget based on payments, signals to Meta, sales team oversight, AI agent.

Does return on ad spend rise on its own when the budget is smaller? We checked every day of June–August, before the switch. Sales were counted within the first 14 days after the lead, so all days had the same maturity.

Before the switch, return on ad spend barely depended on daily spend. So the growth after 31.08 is not explained by the smaller budget.

Limitations. The sample is small: 7 sales before and 21 after. At the end of August the ad accounts were being moved and the ads were relearning, which could also have pulled August down. The NETTO dashboard, which also links sales by phone number, shows the same: return on ad spend 1.20 → 2.35, cost per client $526 → $339.

2. A month before and a month after Before and after

The second comparison: two continuous months of the same length, with half the budget in the second. Ads for one of the center’s lines (in-person classes), all of its ad accounts.

The “before” window leaves out one more ad account that ran for 3 weeks: $2,139, 1 sale. With it, the cost per client before setup would have been $462 and return on ad spend 1.37, which would make the comparison more favorable to NETTO. We kept the stricter version.

3. Every week in a row, including the bad ones Before and after

A single week proves almost nothing: sales per week range from 3 to 21, and the cost per client swings several-fold. On top of that, a week’s numbers change while its deals mature: as of 14.09, the week of 03.08 showed $840 per client and return on ad spend of 0.94; as of 27.09 it was already $560 and 1.49. So we show the best week separately and don’t present it as a typical result.

Best week after setup: 31.08.2026–06.09.2026. Cost per client $202, return on ad spend 4.54, spend $1,615, 8 sales. This is a small sample. Weeks with a similar cost per client also happened in June–July: 20.07–26.07: $258 and 29.06–05.07: $290. Judge the lasting effect by the comparisons above.

Four weeks after the switch to the new ad accounts (31.08.2026–27.09.2026): spend $8,144, 21 sales, cost per client $388, return on ad spend 2.29. The last two weeks are still maturing, and in them the center was no longer using Sales oversight and the AI agent.

4. What else changed between the periods

• At the end of August, the budget was moved from two old ad accounts to two others.
• The budget was cut deliberately, based on data about which ads bring payments.
• There were fewer people in marketing and on the sales team.
• From mid-September the center stopped using Sales oversight and the AI agent.

So the difference between the periods cannot be attributed entirely to NETTO. This is a “before and after” measurement at one center, not an experiment.

5. How we calculated

6. How NETTO finds sales that tracking missed Verified

June–August 2026, data as of 25.08.2026: tracking by tags in deals saw 138 sales from ads; after matching by phone there are 170. Tracking missed 32 of 170 (19%). Return on ad spend for these three months across all ad accounts: 1.51 by tracking, 1.81 after recounting (this is a different period and a different set of accounts than in sections 1 and 2).

1. Lead from ads. The deal in the CRM has an ad ID, or the phone number came from a Meta lead form.
2. Contact. We take the phone number of the deal’s contact. The calculation uses an encrypted number (hash); the numbers themselves never appear in reports.
3. Sale. A deal with the status “won”.
4. Matching. If a won deal has no ad tag, we look up its phone number in the history of ad leads: any ad lead from that number counts if it came no later than one day after the day the deal was created. There is no limit on how far back.
5. Attribution. In this recount, a number’s first purchase went to that number’s first ad, and a repeat purchase (for example, a second child) to the last one; out of 32 cases, the two rules gave different results once. In the NETTO dashboard since 25.08.2026 there is one rule: a sale without a tag goes to the latest ad lead from that number and takes that lead’s date.

Where these sales come from: a sales rep created a deal by hand for a number that came from ads; a second child from the same family; the deal was created by the WhatsApp integration or the call bot without a tag (17 sales out of 32).

Limitations. We managed to link 98% of sales: 6 of 301 deals have no phone number. Chats from “click to WhatsApp” ads without a saved number cannot be linked. The first-touch rule also credits ads with sales where the last touch was the call bot (3–4 cases out of 17), so by last touch the gain would be slightly smaller. Duplicates: a sale is counted once, even if the number has several ad leads.

7. The center’s funnel: where leads get lost

25.05–22.08.2026, leads from Meta lead forms, 2,987 leads, data as of 23.08.2026.

8. Reaching the lead on the day of the lead Observation

Leads from 14–28.07.2026, data as of 24.08.2026. If the person was reached on the day of the lead, they bought 8.7% of the time (288 leads); if later, 4.1% (173); if never reached, 2.1% (95).

This is a correlation in the data, not an experiment: people who pick up right away also buy more often. The speed of the first call by itself barely affected bookings: a call within 5 minutes gave 36% bookings, 5–15 minutes 45%, after 10 hours 43% (831 leads). So NETTO does not promise that an early call alone will raise sales. NETTO makes sure every lead is reached on the day of the lead, and shows overdue calls.

9. Night leads and the AI agent Hypothesis

What was measured Observation: 32% of leads come in between 22:00 and 08:00. They waited on average (median) 629 minutes for a reply and bought 3.3% of the time vs 5.7% for daytime leads. But night leads that were called in the morning booked at the same rate as daytime ones and still bought less often. So the difference lies largely in the people themselves, not only in reply speed.

Hypothesis: the AI agent replies at night right away and removes the wait until morning. How we test it: we compare booking, attendance and payment for night leads with and without the agent’s reply. The AI agent has been working at the center since August 2026, and the volume is still small (9 deals in August). There are no conclusions about money yet, and we don’t draw any.

10. Checking the NETTO dashboard against original sources Verified

Spend in the NETTO dashboard matched the Meta ad account to within 0.0000% over three periods. Cash register totals matched the CRM with a deviation from −1.01% to +0.28%. Every number in the dashboard expands into a list of deals with links to the CRM.

Summary

Want the same numbers for your center? On the review call you open your CRM, and we show which ads bring payments.

NETTO · Sole proprietor «BYR Media» · Astana · WhatsApp +7 777 935 05 08 · Case study and numbers · Terms and data · For AI assistants