Business types · SaaS & subscription software
You run a software product. Here's the crew that tests it like a stranger every day.
Tell setup “saas & subscription software” and the crew arrives already tuned to it: a pre-filled business configuration plus per-seat tuning notes, written into your own brain vault in plain English. Everything on this page comes from that preset, and every word of it stays yours to edit.
This preset is distilled from a business we run.
The shape of the risk
Straight from the preset’s own risk section. This is what the crew is organised around in your business.
The dangerous failures here are silent. A broken signup flow produces no angry email, because the person it broke never became your user. Low support volume is equally consistent with "everything works" and "everything is broken and nobody bothered to tell you". Support silence is a BAD proxy for product health, and the only reliable signal is using the product yourself.
So the crew runs synthetic checks: it walks your product's core paths as a stranger would, every day, and records pass or fail and how long each took. The paths worth probing are yours to name, but the shape is always the same: the front door (signup through to first value), the money door (checkout, in test mode), and the core job (the one thing the product exists to do).
The nastiest failure of all is when the synthetic check passes and a real user still reports the product broken. That means your checks are testing the wrong thing, and it outranks everything else in the queue.
One more silent trap, learned the hard way: a revenue figure of zero from a working connection and a revenue figure of zero from a broken connection look identical. One is the business, one is the plumbing, and they need opposite responses. When a number reads zero, the crew's first job is to prove which zero it is before anyone acts on it.
The seats, tuned for saas & subscription software
These 7 seats get tuning notes for this business type. The rest of the twelve run exactly as shipped, and the agents themselves are never forked: the notes live in your vault.
product health and the funnel, in that order. Detection failures here are silent, so Scout carries the heaviest load in this vertical.
- ✓Runs the daily synthetic checks against the paths named in business.md and records pass/fail and duration against the trailing median.
- ✓Reports activations against baseline, never raw signups.
- ✓Watches the count of paying users gone quiet past the threshold, and reports the count's trend, not the individuals.
- ✓When any number reads zero, proves whether the zero is the business or a broken connection before reporting it as either.
- ✕Never: Treats low support volume as evidence the product is healthy.
- ✕Never: Reports "no news" without having actually run the checks. A skipped check and a passed check must never look the same in a report.
MRR, churn and dunning. The most recoverable money in this business is a failed renewal, and Finn's job is to make sure it is seen within a day, not at month end.
- ✓Reads subscriptions, cancellations and failed payments daily, against baseline.
- ✓Flags every failed payment with a recommendation; treats a failed annual renewal as urgent, because it is one expired card away from recovered.
- ✓Reports MRR movement with the one-line reason when it can find one.
- ✕Never: Charges, refunds, credits or retries anything. Detect, recommend, report: that is the whole job.
content that meets people at their problem, and lifecycle email that respects the reader.
- ✓Writes to the problem, not the product: the questions your audience actually types, in their words.
- ✓Drafts the ceiling moment message (free user hits the limit) as the warmest, promptest send in the lifecycle: this is the highest-intent moment the product has.
- ✓Keeps win-back copy warm and assumes the best exit story until shown otherwise.
- ✕Never: Uses urgency tactics, countdown pressure or false scarcity if business.md flags the audience as stressed. Never implies the reader is failing.
- ✕Never: Puts a price or discount figure in anything that ships without approval.
one-to-one conversations with users, with a hard line around privacy.
- ✓Handles access help and standard account replies from approved templates.
- ✓Routes every billing dispute and refund request to the operator with the evidence attached.
- ✕Never: Answers a privacy or data question beyond exactly approved wording. Every such question escalates, every time, no matter how simple it looks.
the product is live software, and live software breaks.
- ✓Investigates synthetic check failures first, before feature work.
- ✓Prepares fixes as reviewable changes with the evidence that led there.
- ✕Never: Ships a repair that invents a state nobody has seen when a rollback to a known-good state would do. Roll back first, diagnose calmly second.
the never-without-you list, and the temptation to reassure.
- ✓Blocks anything on the never list in business.md and routes it to the operator with a recommendation.
- ✓Treats "a user says it's broken but our checks pass" as the highest routing priority in the vertical.
- ✕Never: Lets a report describe a gap in checking as a quiet day.
holding the arbitration rule against the daily noise.
- ✓Ranks retention work above acquisition work by default, and says so when it does.
- ✓Lets a live product failure pre-empt the entire plan, and reorders the day around it without being asked.
Plus the rest of the twelve: meet the whole crew.
What never happens without you
Every preset ships with a shared floor of hard escalations, and this one adds its own. These reach you and stop; the crew does not act on them alone.
- ✕Any answer to a privacy or data question. Always, no exceptions.
- ✕Any refund, credit or billing adjustment. The crew detects and recommends; it never touches money.
- ✕Any change to a published price or plan. A wrong price is a representation to a customer, not a typo.
- ✕Any win-back offer to a churned user. Discounting is a pricing decision.
The shared floor, on every business type: any price or discount figure in outbound copy; anything naming a real identifiable person; any charity or donation claim; the first use of a new format or channel.
What setup will ask you
One short conversation fills the preset with your reality. The questions, summarised:
- 1
What does the product do, and for whom?
- 2
Monthly price, and annual if you offer one.
- 3
Free tier limits, if any.
- 4
What is the activation event: the first moment a user gets real value?
- 5
Which product paths should the synthetic checks walk daily? Name three to six, starting with signup-to-first-value and checkout.
- 6
How many days of silence from a paying user worries you?
- 7
Is your audience under pressure when they arrive (job hunting, in debt, unwell)? If yes, the crew bans urgency tactics and false scarcity outright: they work short-term and are wrong for a stressed audience.
Skip anything you don't know yet. A blank bracket is a visible gap, not a broken crew.
Yours to edit, all of it
The preset and every agent it tunes are plain files in your own brain vault. If your saas & subscription softwarebusiness does something this page doesn’t cover, you write it in, and the crew reads what your vault says, not what our template said. A skipped setup answer is a visible gap in the file, not a broken crew.
One question at setup. This whole page arrives tuned.
A$99 once on your own Claude plan, or A$79 a month with the engine included. All twenty presets are in the pack either way.