What the system produces, and the two tracks in it
It produces two flows that need different design. A low-frequency broadcast that keeps you in mind across the whole list, and a high-value individual track that produces actual conversations. Businesses that build only the first get an open rate; businesses that build only the second get twenty messages and then stop.
The broadcast track is deliberately unambitious — quarterly, useful rather than promotional, and designed so that missing an issue costs nothing. Its job is to keep you from becoming a business somebody used to work with.
The individual track is where the revenue comes from. A small number of one-to-one messages per month, specific, asking for nothing, from a person. Ten a month covers a hundred and twenty relationships a year, which for most founder-led businesses is the entire history within two years.
The architecture, part one: the list and its states
Stage one is extraction and enrichment. Every past client out of invoicing and into the CRM as a record with the fields the system needs: what you did, when it ended, who the actual human was, why the engagement ended, and the last contact date. The why-it-ended field is the one nobody captures and the one that determines which track a record belongs in.
Stage two is state assignment, and four states cover it. Dormant-good: the work went well and it simply ended. Dormant-unknown: no record of how it went. Dormant-poor: it ended badly, and this state exists so those records are excluded rather than accidentally messaged. Reactivated: currently in a live conversation.
Stage three is the ongoing entry rule, which is what makes this a system rather than a one-time cleanup. When an engagement closes, the client automatically becomes dormant-good or dormant-unknown, with the closing owner setting the state. Without this stage, you rebuild the same backlog and run the same one-off campaign in three years.
Stage four is ownership. Each dormant record has a named person, and for most small businesses that is whoever ran the engagement. Unowned lists do not get worked, and this is the field most likely to be skipped because it feels administrative.
The architecture, part two: the two tracks and the handoff
Stage five is the broadcast, on a quarterly trigger. One genuinely useful thing — something you learned, something that changed in their world, an answer to a question clients keep asking. The design constraint is that it must be producible in under two hours, because anything more expensive gets skipped in a busy quarter and the rhythm is the whole product.
Stage six is the individual queue, and this is the stage that needs engineering. The system surfaces a small number of records each month — prioritized by relationship quality, time since contact, and any external trigger worth referencing — and puts them in front of their owner as a task with the context attached: what you did for them, when, and what they said at the time.
Stage seven is the message, which is not automated and must not be. Automate the reminder, but write the message like a friend. The system's job is that the person comes up and that the context is there; the ninety seconds of writing is the part that makes it work, and a templated version is exactly the thing everybody recognizes and ignores.
Stage eight is the hard rule about the first contact after a long silence: it asks for nothing. No offer, no pitch, no availability. A pitch after two years of silence retroactively converts the whole relationship into a transaction, and it is the single most common way reactivation programmes damage the asset they were meant to use.
Stage nine is the handoff. A reply that indicates interest moves the record out of dormant and into the normal pipeline, with its history attached — at which point the standard qualification and proposal process takes over rather than a special one.
The failure edges
The first: leaving the list in accounting. Everything else can be perfect and the system still will not run, because nothing there generates a next action and nobody owns the records.
The second: opening with a pitch. Converts poorly and, worse, spends the goodwill. The people most likely to have bought again are the ones most disappointed to discover the first contact in two years was a sales message.
The third: a broadcast that is too expensive to sustain. A newsletter designed to be impressive produces two issues. One designed to be producible in two hours is still running in year three, and the rhythm is what the whole thing is for.
The fourth: no dormant-poor state. Every list has relationships that ended badly, and messaging them cheerfully is an avoidable and memorable mistake.
The fifth: no ongoing entry rule. Without it this is a cleanup rather than a system, and you will do it again from scratch in three years with a bigger backlog.
The sixth: over-automating the individual track. The moment the personal messages become a sequence, they read as one, and the response rate collapses to broadcast levels while the relationship cost goes up.
The prioritization, and what a good month looks like
The individual queue needs an ordering or it defaults to alphabetical, which wastes the highest-value contacts on whichever month you were most diligent. Three inputs work: how the engagement went, how long since contact, and whether anything external gives you a genuine reason to reach out — a funding announcement, a hire, a change in their market.
That last input is worth instrumenting even crudely, because it changes the message from a check-in to a specific observation, and specificity is most of the difference in response rate. A simple monitor on your dormant-good list for public news is cheap to build and turns a generic queue into a set of genuine reasons.
A good month is ten sent, two or three replies, and perhaps one conversation. That ratio is worth stating because founders abandon reactivation when the first ten messages produce two replies and they read it as failure. It is not — it is roughly the expected rate, and the compounding matters more than any single month, because the twelfth message to somebody who has heard from you three times lands differently from the first.
What done looks like, and what it takes to build
Done is ten personal messages a month going out without anyone deciding to, a quarterly useful thing that costs two hours, and a steady trickle of returning work that nobody ran a campaign to get.
The checklist: past clients extracted into the CRM with engagement, end date, human, reason and last contact; four states with an ongoing entry rule at engagement close; a named owner per record; a quarterly broadcast with a hard production budget; a monthly surfaced queue with context attached; a first-contact rule that asks for nothing; and a handoff into the normal pipeline.
The build is one to two weeks, and the extraction is the awkward part — invoicing records frequently carry a company name and nothing else, so reconstructing who the human was is manual for the historical backlog. It is worth doing once; the ongoing entry rule means it never has to be done again.
The prerequisite is a CRM that can hold a state and generate a task. If your pipeline lives in a spreadsheet, the surfaced queue has nowhere to surface into, and the individual track degrades into a reminder to remember.
Move the list out of accounting into something that generates next actions, then run two tracks: a cheap quarterly broadcast and ten personal messages a month that ask for nothing.