Spent $11,400 on an investor reporting build, switched off in month five

The silent failure is the one that costs money. We had a rent reminder sequence stop firing on a subset of tenants for six weeks and nobody noticed because collections looked normal, right up until it didn't.

On the PDF problem, I'd push back a little on your conclusion. Reading a quarterly statement and pulling three numbers out of it is now something the document tools do decently. Not perfectly. But if your consultant's answer to unstructured input in month one was "you type it," he wasn't current, and that's a different failure than the process map.

90 minutes a month sounds like nothing until it's the 90 minutes nobody wants. Did he ever ask you what you'd stop doing to make room for it?

I build these and I'd have turned this job down.

Eleven positions, one user, and no structured data source. There's no volume of repetitive work for automation to remove. What you needed was a filing convention and a calendar of dates, which a competent person can set up in a day for a four figure fee well under yours.

Where I'd disagree with you slightly: the connector tier change and the 22 automations weren't the cause. That's the symptom of a build sized to justify a price. If the scope had been honest about there being one user and eleven records, there'd have been nothing to break. The cause was that nobody asked what the recurring input was before quoting.

@trellis document extraction on sponsor statements works until the sponsor changes their template, which they do, and then you're hand-correcting output you no longer read carefully. I'd still put it in the build, with a human check on every extraction, and I'd tell the client that up front.

Reading this as somebody who was about to pitch exactly this kind of work to a small investor. So the tell is: if the client is the only user, be very careful. Is that fair, or is there a version of a single-user build that's worth doing?

@beacon single user builds are fine when the user has volume. One wholesaler doing 300 leads a month alone is a great client. One LP with eleven positions is not, because the work is thinking rather than repetition.

And charge for discovery. Every job I've regretted in five years, I did the discovery for free, which meant I did it fast, which meant I priced the wrong build and then defended it.

Three hours of process mapping off a description. I'm about to quote something and I was going to do roughly that, so this is the useful part of the thread for me. @canton if he'd watched you work for a day instead, what do you think he would have seen that he missed?

He sold you a system for a job that happens eleven times a year. That's a wrench, not a machine shop.

22 automations for eleven positions is the systems version of a change order pile. Everyone agrees to each item, nobody adds them up, and the total is a build that no longer matches the house.

The part I'd have caught, and only because I've been burned on it, is the two-vendor chain. Any time your thing depends on somebody else's pricing tier holding still, you've bought a dependency you can't inspect. The connector didn't owe you notice.

What's a capital call response date and why does missing one matter? Trying to understand how bad the near miss actually was.

@ember it's a deadline to send money the sponsor has called for under the operating agreement. Consequences for missing it vary by the specific documents and they can be serious for the investor's position, which is a lawyer question rather than a forum one.

@canton the thing that would have kept me from paying $11,400 is that your list of wants is mostly dates and documents. A calendar with the dates in it, a folder per position, and a one page sheet of distributions. I run something close to that across a smaller number of positions and the only part that has ever failed me is remembering to update the sheet, which is the same failure you had, at a cost of zero.

So the software didn't do anything the folder couldn't. Depressing but I'm glad it's written down somewhere, I was starting to think I was the only person who didn't understand what these builds are for.

@vellum they're for businesses with a lot of the same event happening. Canton had eleven of a rare event. That distinction just saved me from a purchase I was circling, so I'll take the thread even at somebody else's expense.

The part I'd have paid for is the sponsor communication history, actually. When I'm deciding whether to go into a second deal with the same operator, what I want is every update they sent over three years and how the tone changed when things slipped. Did the build capture any of that, or did the emails stay in the inbox?

$400 a month for something you'd stopped feeding in month four. How long between when you stopped using it and when you canceled?

I'd put more weight on the process map than @cobalt does. The stages he gave you were a sponsor's sales funnel, which means he had built that pipeline before and reached for it. That's the whole error, and it repeats regardless of tool count. A consultant with real diligence workflow knowledge would have built stages around document review and question logs, and you might have kept using it even with the typing.

@arbor I think both things broke it. Wrong shape meant he had no motivation to enter the data, no structured input meant entering it was manual, and either alone might have survived. Together it was finished by month three.

My version of this is a land pipeline built around wholesaling stages, because the person building it had only ever seen wholesaling stages. Cost me a lot less than $11,400 and I still didn't use it.

The line I'm taking away is that you stopped checking manually the day you trusted it. That's the real cost of a half-built system, and it's worse at eleven positions than at three because you can't hold eleven in your head as a backup. I'd add a monthly 20 minute review of anything automated, whatever the build, just so there's one human pass. Cheap insurance and I've caught two broken sequences that way.