A marketing automation platform: desktop → Telegram → web

Summer 2026: spend and reports. Meta spend is written to the tracker in each ad account's own time zone. One profile had accounts in two time zones: the day boundary was about 10 hours off, and part of a day's spend landed on the next day. Counting only unique clicks when writing removed a 10–15% overstatement. Each write is a full-day total, so running it again is safe: it runs 96 times a day, and yesterday is recalculated after midnight while the network sends its late hours. Reports from sign-up to first payment run for each campaign in the second network. Spend and installs come from that network; sign-ups, payments and revenue come from the tracker. The two are matched by a six-character campaign ID in the campaign name. The report shows cost per install, conversion at each step, ROI and the break-even point. The settled rate (sign-ups older than 6 hours) showed the same offer converting differently in different apps, so the app is checked before the offer. Payments arrive over 24–48 hours, so the report flags the first-day rate as understated.

Tracing clicks to ads. A filter on fbclid prefixes in the event collector dropped 492 of 672 clicks a day: a sale reached the Conversions API without its click ID, and of 4 sales a day the ad account saw 2. Checking the whole prefix family raised the share to 672/672. Then the prefix list was removed altogether, since such a list goes out of date with the next format change. 48 sales from a 7-day window were sent again with the same fixed event ID, and all 48 were accepted. Several campaigns ran one post with one link, and the tracker merged their clicks into one bucket. The adid field inside fbclid matched the real ad ID in 0 of 137 checked clicks. So the lookup table is my own, built from clicks where Meta adds the IDs itself: 262 pairs, 22 ads, 0 collisions, and an independent check of 361 out of 361. A directory of ads, ad sets and campaigns covers 10 ad accounts (912 ads, 760 ad sets, 244 campaigns) with 0 integrity errors. The collector remembers the last record it processed, so repeated runs don't double-count; before that, every run doubled its counters. The analytics bot doubled tracker figures for campaigns with the same name and counted sales from only one stream: 10 against 13 in the tracker, which triggered a false stop signal. Now all streams are added up, and the report matches the tracker to the cent. 34 tests on real samples, all offline.

The rules engine. Five possible recommendations for each funnel (a tracker campaign and its Meta campaigns), from an early stop to scaling up. The engine suggests and the operator decides: it writes nothing to the tracker or the ad accounts. It reads three sources. The tracker is the only source of money and funnel data. Meta provides ranking, live spend and account health. The partner network is a cross-check on sales: when it reports more sales than the tracker, a stop is blocked. The thresholds were derived from 92 campaign-days and matched the operator's rules of thumb: they are the points where ROI crosses zero, measured by the price of a unique click and of a landing-page click. Rebuilding the numbers from individual clicks matched the tracker report in 1,704 of 1,704 day-and-campaign cells. A 90-day backtest without look-ahead showed that the first version would have stopped the flagship funnel, which went on to bring 139% of the account's profit. So a rule was added never to stop a profitable funnel, the minimum sales for a cost-per-action decision rose from 3 to 8, and the worst-price-zone rule now needs at least 2 sales; it had issued 9 of 17 recommendations with zero sales. Over 90 days the rules affect 0.8–1.8% of spend, and the report says so plainly. 49 tests and a daily Telegram report. The bot's campaign cards use the same engine and show the reason for each recommendation.

The analytics bot and the token store. Meta profiles are added and switched from the chat, and a token is checked and deleted from the conversation straight away. Pause and resume work with any of the operator's tokens, and Meta rate limits trigger retries after 1, 2 and 5 seconds. The web token store checks scope and expiry, shows a token only after you re-enter your password, can archive and restore, locks itself down automatically, and logs the devices used to sign in. An offer can no longer be created with an empty or placeholder link. That check was added after 33 offers were found pointing to a placeholder.