Product1 distinct publisher3 min readUpdated
The wildfire alerting nonprofit is generalizing to a second hazard while its own headcount and revenue double. Its CEO calls that growth a little too wild for comfort.
The Product Desk · Product desk

Compiled by The Product DeskSomething wrong?How this is made
Watch Duty, the Santa Rosa, California nonprofit whose wildfire alerts became a household name during the Los Angeles fires, is expanding into flood monitoring across all 50 states, CEO John Mills said in an interview with Rapid Response, the Fast Company podcast hosted by former editor-in-chief Bob Safian [1][2][3][6]. It is doing that in the same stretch in which it doubled both headcount and revenue, growth Mills describes as "a little too wild for comfort," and it is doing it without investors or a cap table [4][5].
The fire product works because its inputs are specific. Mills says the system scrapes and mines signals every minute of every day, including firehouse tone-outs, where the siren goes off and the apparatus bay door opens automatically, and wildfire cameras carrying AI detection [12]. A hit arrives as a latitude and longitude with a fire type; staff resolve it to a place, find that region's Slack channel, and ping it, all inside a few minutes [13]. That is how the Spokane fires reached them, and within minutes they knew there was more than one fire, that there was wind on it, and that it was near structures, which turned it into an all-hands event [15]. Mills claims the pipeline puts Watch Duty level with or ahead of firefighters on first knowledge [11]. Fast Company describes the Spokane fire as the most destructive in Washington state history [9], and Safian, citing the numbers to Mills, said the fires there were apparently purposely set, burned more than 10,000 acres, destroyed hundreds of homes, and forced 65,000 people to evacuate [10].
The discipline around that speed is the part worth copying. Every fire is treated as life-and-death because nobody yet knows what it is [14]. After every event the team runs an after action report, and Mills says the standard operating procedures come out harder and more durable each time, with the underlying system unchanged [17]. His stated ranking is explicit: actionable and accurate outrank speed, and he will take being one minute later in exchange for more precision [18]. That ordering was a sales problem before it was a virtue, something the organization had to prove to government agencies early on: "just because we're speedy doesn't mean we're hasty" [19].
Which is exactly why floods are a harder lift than a feature announcement suggests. Every signal Mills describes is fire-specific, and the abridged transcript does not describe a flood equivalent [12][21]. The after-action flywheel that hardened the wildfire SOPs starts at zero for water, and it starts during a season Mills says will only get worse, with British Columbia already under a province-wide state of emergency and a team that sleeps in shifts once its communities are hit [7][8][16][17]. Doubling revenue alongside headcount also means revenue per employee is roughly flat, so the second hazard is being funded by more people rather than more output per person [20].
Watch for whether flood alerts ship with their own named signal sources and their own after action cadence, or ride the fire SOPs; whether agencies extend to flood the credibility Watch Duty spent years earning on fire [19]; and whether the accuracy-over-speed rule survives a hazard where the sirens do not go off [18]. The published excerpt gives no launch date, no revenue figure, and no headcount [21].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
CEO John Mills discussed the decision to expand Watch Duty into flood monitoring across all 50 states.
Since the L.A. wildfires, Watch Duty has doubled its head count and revenue, growth Mills describes as a little too wild for comfort.
Fast Company describes what it takes to scale a life-saving operation without investors, a cap table, or the luxury of getting it wrong.
Mills says alerting signals include a firehouse being toned out, meaning the siren goes off and the apparatus bay door automatically opens, and wildfire cameras with AI detections on them; Watch Duty scrapes and mines lots of different signals 24 hours a day every minute.
When a signal hits, Watch Duty's system pings with a latitude and longitude and a report such as a tone out for a vegetation fire; staff look it up, identify the place, find that channel in Slack, ping it, and everybody springs into action, all in a matter of minutes.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Single-source, self-reported
All operational and growth claims come from one publisher's abridged podcast transcript in which the CEO is the only substantive speaker. There are no figures, no audited financials, no latency measurements, no agency confirmation, and the excerpt is truncated mid-sentence. The concrete process descriptions (AAR cadence, dispatch workflow, accuracy-over-speed rule) are internally consistent and specific, which lifts the score above the floor, but nothing here is independently checkable from the supplied material.
Real wildfire deployment, unproven flood rollout
The wildfire product shows genuine in-the-field adoption: active operation across the Spokane and British Columbia events, a described working relationship (including early friction) with government agencies, and organizational doubling attributed to post-L.A. demand. Against that, the flood expansion is an announcement with no launch date, no coverage detail, and no users; and no download, alert-volume, or agency-count metric appears anywhere in the supplied material, so measured adoption stays below the midpoint.
Mildly overstated
The framing runs somewhat ahead of what is shown: '50 states' flood coverage is asserted with no launch date or flood data sources, and 'we know before the firefighters know' is an unverified self-report presented as capability. The overstatement is modest rather than severe because the substance is process detail rather than superlatives, the growth claim is hedged as 'a little too wild for comfort,' and Mills explicitly subordinates speed to accuracy and admits Watch Duty was not ready for the Texas floods.
Promotional format, aligned interests
Both parties have visible incentives pointing the same way. The article is an abridged transcript of Rapid Response, a show hosted by Fast Company's former editor-in-chief and produced by the Masters of Scale team, ending in a subscribe call to action, so the publisher is promoting its own property. The subject is a donor-funded nonprofit whose revenue depends on public salience and which is announcing a new product line, and the format contains no adversarial questioning or outside voices. Incentive alignment is high without evidence of fabrication, so this is a distortion-risk signal rather than a finding of inaccuracy.
Low-moderate
Confidence is limited by cluster structure rather than internal contradiction: one publisher, one source, one speaker, and a transcript truncated mid-sentence. What the source says is clear and consistently reported, so the descriptive claims about the operating model can be relied on as attributed statements; the quantitative and capability claims cannot be relied on at all until corroborated.
product
Upwork says 23% of its clients are pulling work back from AI. Treat it as a reversal rate.1 distinct publisher
product
Two years of jobs data invert the AI risk rankings: clerks shrink, managers grow1 distinct publisher
product
OpenAI shipped a teen ChatGPT. The over-65 cohort doubled to 23% and got nothing.1 distinct publisher
product
White House lets vetted firms hack back and leaves liability blank for 60 days1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 17, 2026