Postcards that wait for the weather

Most marketing runs on the company's schedule. A newsletter goes out on Tuesday. A campaign launches in September. A sales list gets worked when somebody has time to work it.

Some offers have a better clock.

For a standby-generator company, that clock is the weather. A household on a private well can lose more than lights during an outage. No power means no pump, which means no taps, no shower, no water for the heating system, and sometimes a frozen house if the owners are away. The days after a storm are when the value of backup power is easiest to understand because the problem is still sitting on the kitchen counter: spoiled food, dead batteries, a basement that almost flooded, a family schedule rearranged around a dark house.

I built a pipeline around that moment.

It watches public outage and weather feeds across the service area. It scores each event by type, severity, geography, and the number of relevant homes nearby. Severe thunderstorm warnings, high-wind warnings, winter storms, and major outages can all count. Heat advisories, broad watches, and noisy low-value alerts usually disappear before anyone sees them.

The audience is already mapped before the storm arrives. Public well-completion records identify homes that likely depend on a private drinking-water well. Those records are cleaned into a small local database with address, town, coordinates, well type, drilling year, and source metadata. The campaign can filter to the exact prospect segment: drinking-water wells, recent enough records to trust, inside the service area, outside the suppression list.

That preparation matters because the event window is short. After an outage clears the threshold, the system can match the outage geometry against the address pool, remove current customers and recent contacts, deduplicate the list, estimate the cost, assign a tracking number, publish a campaign landing page, and submit the mail batch.

No one needs to draw a polygon at midnight. No one needs to export a CSV while the phones are already ringing.

The trigger is conservative

Weather data is loud. The system has to be good at silence.

Each alert runs through tiered rules. Critical events always surface. Major warnings need enough relevant homes in scope. Watch-level alerts need both scale and a narrow enough geography to be meaningful. The pipeline groups similar alerts by event type and calendar day, then applies a daily cap so one active storm line cannot create a pile of duplicate starts.

The goal is commercial judgment, encoded plainly. A tornado warning touching the service area deserves attention. A generic special weather statement gets ignored. A winter storm warning over a county full of private wells means something different from the same warning over an area with no target households in the database.

That judgment is visible in the logs. Every skipped event has a reason: wrong alert type, too few homes in scope, too many counties, already seen today. When the system acts, the run is auditable after the fact.

The list is local

Good automation usually starts with unglamorous data work.

For this campaign, the useful audience was sharper than "homeowners": households where a generator solves the immediate problem of keeping a private well running. Public records made that audience available, but only after cleanup. The source data had to be pulled from state systems, normalized, filtered, mapped, and made searchable by geography.

Once that base exists, a campaign can be assembled from the actual service territory instead of a rented demographic segment. ZIP codes work. Drawn polygons work. Outage polygons work. The same recipient builder can support a pre-season mailing, a storm-follow-up mailing, or a hand-picked local campaign around one town.

Suppressions are part of the audience, too. Existing customers, quoted prospects, and recently contacted households are filtered out before anything goes to print. The system treats restraint as part of relevance.

The mail piece carries its own attribution

The postcard is generated per recipient. The greeting falls back safely when an assessor record looks like a trust, LLC, or business. Town names are formatted for natural copy. The QR code carries a recipient tag through a short redirect, so scans can be attributed without printing a brittle final URL. The printed URL stays stable, while the destination can change behind the redirect.

Each campaign also gets its own local phone number. Calls forward normally, but inbound activity belongs to the campaign that produced it. That makes the physical mail behave more like a digital channel: each batch has cost, audience, creative, destination, phone number, send date, and response trail in one place.

The send path is built for real operations. The landing page publishes before the cards go out, so the QR destination is live when the mail arrives. Postcards are submitted with idempotency keys, so a retry skips duplicate cards for the same recipient. If a few records fail because an address is incomplete, the successful rows stay mailed and the failed rows can be corrected.

Why physical mail fits the moment

A storm-triggered email would be faster, assuming the company had clean email addresses and permission to use them. Paid ads could move quickly, assuming the platform could target the affected households with enough precision and enough inventory at the right moment.

Direct mail is slower, but it has a useful shape here. It arrives after the immediate disruption, when people have had time to cool down and start thinking about prevention. It reaches the home as a physical reminder, with enough space to speak to the neighborhood and the specific problem the household just experienced.

The automation makes the message quieter, more specific, and better timed.

That operating pattern is the part I care about. A real event occurs in the world. The system understands whether the event matters, who it matters to, what should happen next, and where the guardrails are. The visible output might be a postcard, a call list, a landing page, a repair reminder, a customer check-in, or a sales brief. The deeper pattern is the same: listen for demand signals, narrow the audience, respect the exclusions, create the asset, record the decision, and hand the next step to the right channel.

For businesses with local service areas, high-consideration purchases, seasonal risk, or a strong "I should really deal with this" moment, that kind of automation can feel less like marketing and more like good timing.