A morning email built off daily filings: 22 paid seats, one 11 day outage

$180 a seat for something that saves a subscriber an hour a day is underpriced and I think you know that. The question is whether you raise on existing seats or only new ones. I did the second thing five years ago and still have four customers paying 2019 rates who cost me the same to serve as everyone else. Grandfathering feels generous in the moment and it becomes an annoyance you carry forever.

Did you have anyone paying before you wrote the scraper, or did you build first? I'm sitting on a half built thing and trying to figure out whether I'm about to waste four months.

11 days cost you a quarter of your subscribers. What's the plan if a county goes down for six weeks, or permanently moves behind a paid API? That feels like the failure mode that ends the whole thing, and four counties isn't much spread.

I subscribe to two products in this shape and cancelled a third. The third one failed on dedup, same filing arriving as new because the county reposted it with a different case format. Once you don't trust the email you stop opening it. How are you keying records for dedup, on the document number or something you build yourself?

The unit economics here look fine and the risk sits somewhere else. $3,960 MRR against $310 direct is a 92% gross margin, which is normal for a scraper business and misleading, because your real cost is the six hours a week of parser maintenance. Price that at any reasonable rate for your own time and margin lands closer to 60%, and it gets worse as you add counties, since maintenance load scales with the number of distinct portals rather than with revenue.

That's the thing that decides whether this expands. Every new county adds a parser that breaks independently. Four counties at six hours a week suggests roughly ninety minutes per county per week, so twelve counties is your entire working time before you've sold a single additional seat. Platforms in this space usually solve it by standardizing ingestion so a new county is configuration instead of code, or by paying for a vendor feed where one exists. Both cost money up front and both are the difference between a job and an asset.

The 25% churn from an 11 day outage is the number to design around. @windlass is asking the right question. A cached last-known-good state, a manual fallback you can actually sustain, and telling subscribers on day one that the feed is down all reduce the damage. Subscribers forgive an outage they were told about far more than one they discover.

On defensibility, the scraper isn't it, since anyone can point one at a public index. Your filter rules from 60 days of manual review, the dedup keys, and the subscriber list are harder to copy.

What this depends on: whether your counties keep publishing at current access levels. That's a policy decision you don't control, and it differs by state and office, so read the terms again when they change.

I'd pay $400 a month for this if the phone numbers were verified instead of passed through raw. Half of what I get from skip trace is disconnected and my caller eats the time. That's a product on top of your product.

Does the filter catch anything on land? Tax delinquency on vacant parcels is what I'd want, and in my county that's posted by the treasurer rather than the recorder, so I'd guess it's a different scraper entirely.

Can someone explain what document types you're actually watching? I understand the idea of watching filings but I don't know what a filing tells you about whether someone wants to sell.

What did you build it in? Asking as someone who has never written a scraper and is wondering how far out of reach this is.

Your five cancellations over missing phone numbers is the same wall I'm looking at from the other side. I built the calling operation first and it's eating me alive at $61 per answered call, and you shipped data and let the buyer call. Yours is clearly the cheaper end of the business to run.

What was churn like in months five through twelve once the phone number complaint went away? I want to know whether a data feed holds subscribers long term or whether people cancel once they've mined the county.

22 seats across four counties, how much overlap is there? If six of your subscribers are all working the same county they're calling the same owner within an hour of each other, and eventually the best one figures out he'd rather have exclusivity than a shared list. Have you thought about capping seats per county, or is that giving up revenue you'd rather have?

Six hours a week fixing parsers is the number that would stop me. That's not passive at all, and it's the kind of work that has to happen whether you feel like it or not, because a broken morning email is visible to every subscriber the same day.

How often does the owner name match to assessor records go wrong? Common names, trusts, LLCs. Seems like the sort of thing that's 95% right and the 5% is what makes someone cancel.

Running your numbers: $3,960 minus $310 is $3,650 a month, call it $43,800 a year, against six hours a week of your labor which is roughly 300 hours. That's $146 an hour before you count the sales and support you didn't mention. Solid hourly rate, not obviously a scalable asset yet.

And I'd argue the 22 seats is more fragile than the outage story suggests. One county is what fraction of the value? If it's the biggest one and it goes away, does a subscriber in that county have any reason to keep paying? I'd want that concentration written down.

Two of the counties I care about post nothing usable online and you go to a terminal in the building. I've paid a courier to check weekly. Did you look at whether a person on the ground can feed the same pipeline, or did you just cut those counties?

I went and read the terms of use for the recorder in my county after your last paragraph. Bulk downloading is addressed in one sentence and commercial redistribution in another, and neither is clear enough for me to act on without asking someone. Anyone building this should read the actual county page rather than assume public means unrestricted, and understand it differs by state and by office. @warrant, did your attorney give you a written opinion or a conversation? I'd want the paper if I were selling access.

The 60 days of checking every record by hand is what I keep coming back to. Everything good in the post seems to come from that and it's the least technical thing you did.