Skip to content

Revenue you can account for, and a report that proposes

Canopus is an autonomous growth manager that credits new trials and subscriptions to the campaign or the lifecycle sequence that last reached the person, within fourteen days, and writes a report about it: observations, conclusions, proposed actions.

The number comes with its method attached: last touch, fourteen days, measured against the equivalent period immediately before. It is called an estimate on the screen that shows it, because that is what it is, and the report that follows ends in the slots planned on the calendar and, when the results argue for it, a revision of the strategy you accept or decline. Every attributed euro lands on a contact in the CRM, so the number traces back to a person.

How does a campaign or a sequence get credited with a customer?

  1. Every send and every click is written down with a contact on it

    A campaign email or a sequence email goes out from your own domain and the send is recorded against the person who received it. A click on a tracked link is recorded the same way. Clicks and conversions are the measures; delivery and bounce data arrive with domain sending.

  2. Conversions come from your billing, not from a pixel

    A trial start and a first paying subscription are read from Stripe and dated by Stripe, not by the day a sync noticed them. That matters for every customer who existed before the connector did, and it is what makes the question answerable at all.

  3. The two are joined on their own, and the baseline sits beside the result

    A job that runs on its own looks back twenty-eight days, credits each conversion to the most recent message within fourteen days of it, and writes one row per campaign or sequence per person. The attribution screen then shows the same measurement over the thirty days before, and the difference between them.

What does this replace?

  • A spreadsheet where you paste campaign dates next to Stripe sign-up dates.

    The join runs on its own, by contact, with a stated window and a stated rule.

    Replaces: The attribution sheet
  • Guessing whether a good fortnight was the campaign or just a good fortnight.

    The equivalent previous period is measured too, and the difference is labelled an estimate.

    Replaces: Guessing
  • Sitting down with the numbers to work out what happened and what to try next.

    When a campaign closes, the report is written: observations, conclusions, and two or three proposed actions.

    Replaces: The post-mortem by hand
  • Forgetting by March what you learned from a campaign you ran in January.

    Every finished campaign becomes a playbook entry that later agents are given.

    Replaces: A notes file

What can a last-touch number honestly claim?

Last touch means this: a trial or a new subscription is credited to the most recent campaign or sequence message the person received or clicked, provided it arrived within fourteen days and before the conversion. The email that says a trial ends on Thursday is usually that message, which is why sequences are credited like campaigns. Nothing else is credited. A conversion with no message in that window is credited to nobody at all, which is the answer that keeps the total honest and the one a model that always finds a cause would never give. A message sent the day after somebody paid is never counted as its cause. When a send and a click land in the same second, which happens when someone opens an email the moment it arrives, the click wins the tie, because a person who clicked did more than a person who was merely written to.

The faults of that rule are known and they are printed on the screen rather than buried in a tooltip. Somebody who read three emails and subscribed after the third gives all the credit to the third: the first two did work that does not appear anywhere. Somebody who was going to subscribe anyway and happened to read an email the day before is counted as converted because of it, which biases the figure upward. Somebody who converted on another device, or forwarded the email to the colleague who actually paid, is not counted at all. Modelled multi-touch attribution would need far more data than a company with two hundred customers has, and would trade an understandable bias for an unfalsifiable one.

The baseline exists because of that upward bias. Beside the attributed figure, the same measurement is taken over the equivalent period immediately before, of the same length and not overlapping it: thirty days on the attribution screen, so the thirty days before that. The lift is the difference, floored at zero, because a period worse than the one before it means no visible lift rather than revenue destroyed. It is deliberately crude arithmetic with no seasonality correction and no control group, and the word on the page is "estimate", not "measurement". Fourteen days is the window because it is long enough to cover the gap between reading an email and deciding to pay, and short enough that a coincidence a month later is not swept in.

Attribution is recomputed rather than observed as it happens. A conversion here is not an event anything can hook: it is a field a Stripe sync changed. The only honest design is to look back over the window again and again, so a job that runs on its own reads twenty-eight days of touches and conversions, twice the attribution window, to be sure a conversion recorded yesterday can still find a message from a fortnight ago. Every write is keyed on the conversion kind, the campaign or sequence and the person, and a conversion is credited once, so running the job twice writes the same rows and a corrected sync duplicates nothing. Clicks and conversions are the measures; delivery and bounce data arrive with domain sending. The join is first by contact identity. UTM tags are put on every campaign link, and when your product or your lead form passes them to Canopus with a sign-up, a person who was not yet in your CRM is credited to the campaign or sequence whose link they arrived from. An arrival from anything else, a post or an ad, competes under the same rule as its channel: an older email does not take a customer that a more recent post brought. Beside the campaigns, a table by source counts what the people of each channel went on to do, trials, customers and revenue, whether or not an email ever reached them.

When a campaign closes, three things run in order: what finished is concluded, the report is written, then the next batch of posts is asked for. The report is observations and conclusions: a summary of three or four sentences, up to three findings, up to four numbers. What to do next is not a list at the foot of it. The analyst plans the coming slots on the calendar, each naming a channel, a format, the segment it addresses and the angle it is betting on, and every slot comes back as a piece waiting for your approval. When the results argue against the strategy itself, an objective drifting off its trajectory, an experiment concluded against a hypothesis, the slots of a channel skipped again and again, a reviewer proposes a revision: a diff of the strategy in force, argued on that evidence, that you accept or decline. Declining is as prominent as accepting, deliberately: the acceptance rate of those revisions is one of the two numbers this product is judged on, and it is computed over answered revisions only, because a revision nobody has looked at is not a refusal. A screen that made refusal feel like failure would make that rate meaningless within a few loops.

Every campaign that finishes becomes one row in a playbook, and the verdict is computed rather than asked for. Below thirty recipients a campaign is inconclusive whatever its numbers, which is the most common outcome and is itself the finding: a model asked what a campaign to forty people taught will always find something, and a playbook of confident lessons drawn from noise would poison every agent that read it. A campaign that produced a paying customer is kept regardless of sample, because a customer is not a rate. Published posts you reported results for are concluded the same way, with a higher impression floor, since a figure copied from a platform dashboard by hand is a good-faith estimate. Only the conclusive lessons are passed on, and every producing agent gets them: the analyst writing the report, the email producer drafting a sequence, the content producer writing a post. That is what makes the product get better at your company rather than merely consistent.

Questions founders ask

See your own numbers first

Connect Stripe and your real revenue is on screen in five minutes, before an agent has said a word. Create your account, answer the questions, and your strategy gets written.