The monthly control loop

Measuring once isn't control. The plan needs a habit: a short review every month, with rules decided before the numbers arrive. The monthly control loop has three steps:

1. Compare results with the plan, channel by channel, on cost per paying family.
2. Find the cause of any gap: what changed, and where in the chain people dropped out.
3. Decide, from four moves only: keep going, fix the ad (or whatever piece of work is off), move budget to a channel that's doing better, or re-forecast if the plan's numbers were wrong.

Then write the decision down with one owner next to it.

The order matters. Deciding before finding the cause just moves the problem: cut a channel whose ad was wrong, and you lose a channel that worked with the right ad. Finding causes without comparing first means chasing whatever looks odd.

At Klasna the review is one line per channel (plan, actual, cost per paying family, the rule it triggers, the cause, the decision and its owner), half an hour in the first week of the month.

The first step of the control loop compares results with the plan, and the number that matters is cost per paying family, not views or spend.

cost per paying family = spend ÷ paying families.

October 2026, days 1–21. Spend is on plan everywhere; paying families aren't:

• paid social: 85,000 ÷ 14 = 6,071.43 UAH, against a plan of 82,000 ÷ 27 = 3,037.04;
• search: 54,000 ÷ 22 = 2,454.55;
• school partnerships: 20,000 ÷ 11 = 1,818.18.

Paid social costs twice its plan per family; the other two are at or better than plan. A report that showed only spend ("on plan") would have hidden this completely.

Compare each channel with its own plan and with the rules agreed in September: scale a channel by a quarter when its cost per paying family stays under 4,000 UAH for four weeks; cut it to the test level when it stays over 6,000 for three weeks.

The slips: comparing spend with spend (it can be perfectly on plan while results collapse), and judging a channel on one strong or weak week.

Paid social has stayed over 6,000 per paying family for three weeks, so the rule agreed in September says: cut it to the test level. But the second step of the loop is the cause, because a cut without a cause just moves the problem.

Uliana checked the enquiries. On 2 October the channel switched to a new video, and since then only 40% of enquiries come from parents, against 70% before; the rest are teens. The ad now reaches the user, not the payer, and Klasna never aims ads at minors anyway.

So the decision has three parts: follow the rule and cut spend to the test level, fix the ad by going back to one made for parents, and check the parent share weekly. If the old level returns, the scale rule can bring the budget back.

What not to do: keep spending because the video has the most views (views don't pay); stop paid social for good (it worked at about 3,000 a family with the old ad); or raise its October goal so it looks closer to plan, which hides the problem instead of fixing it.

The third step of the control loop picks one of four moves, and the cause decides which:

• Keep: the gap is noise. A one-week dip during school holidays that returns to plan is a reason to note it and wait, not to act.
• Fix the ad: one channel jumps over 6,000 per family after a new ad reached the wrong audience. Fix it, and cut spend to the test level as the rule says.
• Move budget: a channel does better than planned for long enough. Search staying under 4,000 per paying family for four weeks triggers the scale rule agreed in advance.
• Re-forecast: the plan's numbers were wrong, not the work. If October's targets were copied from spring conversion rates, every channel misses, and the fix is a new forecast for next month built on autumn rates.

The test for re-forecasting: does every channel miss in the same direction, for a reason that lies in the plan? If only one channel misses, the plan is probably fine and the work isn't.

The slip: re-forecasting to make a bad month look on target. That turns the loop into a way of hiding problems.

Scale and stop rules say in advance when a channel gets more money and when it's cut back. Writing them before the month starts means decisions don't depend on mood, on who argues loudest, or on how attached someone is to their campaign.

Klasna's rules, agreed in September:

• scale a channel by a quarter when its cost per paying family stays under 4,000 UAH for four weeks;
• cut it to the test level when it stays over 6,000 for three weeks.

Each rule has a threshold and a time window. The window matters as much as the number: a single week is usually noise. With 20 or so paying families a channel a month, three or four families in one week can swing the cost per family by thousands. One great week under 4,000 doesn't trigger the scale rule; four weeks do.

Why the gap between 4,000 and 6,000: in between, a channel just keeps going while you watch it. Without that middle zone, every small wobble would trigger a move.

The slip: changing the rules mid-month when they give an answer someone dislikes. Then they're not rules any more.

An AI tool can save time in the monthly review, as long as it stays a drafter and a flagger, never the decider.

Sound uses:

• draft the summary from channel totals, with a person checking every number against the export;
• flag channels that crossed a scale or stop rule;
• suggest possible causes, which the team then checks in the data. A suggestion is a hypothesis, not a finding.

Unsound uses:

• letting the tool move budgets between channels on its own. Budget moves stay with an owner who signs them, because someone has to answer for them;
• pasting the full customer list with names and phone numbers to find causes. Personal data stays out of AI tools without a lawful basis and an agreement with the provider, and even then only the data the task needs; channel totals usually answer the question without any names.

The pattern is the same as anywhere in marketing: the tool makes the draft faster; a person ties every number back to the source and signs the decisions.

The control loop works for one person too, with hours as the budget. Compare results per hour with the plan, then look at what happened next.

Dana, the Klasna English tutor with her own practice for adults, planned three hours a week of marketing for November, twelve hours, mostly talks and referrals. Her November:

• two community talks, 6 hours: 2 enquiries;
• asking her students and friends for referrals, 2 hours: 1 enquiry;
• posting, 4 hours: nothing.

Talks and referrals: 8 ÷ 3 = 2.67 hours per enquiry, and all three people booked an intro call for December. Posting: 4 hours, no enquiry.

The cause of the gap is habit, not the channel: she slipped back into posting, which wasn't in the plan. One month is a small sample, so she decides on the split, not the whole approach: keep the talks and move the posting hours back to them.

And the referrals come from her own students and friends: her contract rule still holds, she never approaches Klasna's families or learners.

A stop rule matters most when things go well. Dana wrote two rules in October: if two talks bring no student, switch community; and when her three free places fill, stop outreach.

All three November enquiries booked an intro call for December. If all three start, her places are full, and her rule says stop. She writes that down now, before December's numbers can talk her out of it: a good month makes it tempting to keep going "just in case", and then she has more enquiries than slots and has to turn people away.

Stopping outreach isn't stopping forever. When a place frees up, the loop starts again with the channel that worked: talks and referrals, at about 2.67 hours an enquiry.

The other rule doesn't apply here: her talks did bring enquiries, so there's no reason to switch community. And adding hours to posting would repeat November's gap.

How it fits together: compare against the plan, find the cause, decide from four moves, with rules written in advance and AI as a checked drafter, at Klasna's scale and at Dana's. The final case puts all eight rows of the 2027 plan on one page.