Soren

Mild disagreement with the local-model plan for your case. If you're running out mid-month, you're using it enough that the friction is now the expensive part.

I spent about six weeks rationing, watching a counter and deciding whether a suggestion was worth spending, and the mental overhead of that was worth more than the subscription. Paid, stopped thinking about it, moved on. That's obviously not the answer if the budget is genuinely zero, but if it's "I'd rather not," reconsider what the rationing costs you.

This is going straight onto the roadmap above three actual features. The interrogation framing is exactly what was bothering me about the form.

I will push back on used. At 60 to 80 drinks a day you have no redundancy, and a machine down for three days in week six is not a repair cost, it is your reputation in a town where everyone already tried you once. I paid the premium for new with a warranty and an engineer who could be on site next day, and the one time I needed it, that clause paid for the entire difference. Buy used for the fridge and the furniture, not for the thing every sale passes through.

Consolidate your own, never the clients'. Client domains stay in the client's own account with their own billing, because a domain sitting in my account is a liability the day the relationship ends or the day I get hit by a bus. I have delegated access to about a dozen and own none of them, and every handover has been an email rather than a negotiation.

This. I ran a KVM for four months and the near miss that ended it was a clipboard, not a microphone.

Repository structure did more for me than wording, and it's the part you control unilaterally.

I split into a core package and per-client apps about eighteen months before I had any product ambitions, purely because the copy-paste was killing me. When the product question came up, the answer was already obvious to look at - here is the shared package, here are four thin clients on top of it, and the git history shows the core predating three of the four engagements.

Whoever eventually reads that contract, that's a much better story than "most of it is the same, honestly."

Add never say we are a small team as an excuse to that list. Customers do not care about your headcount and it reads as a preemptive apology.

if you do want a growth number, break it into its components rather than reporting the net. new, expansion, contraction, churned, then net. the net alone hides the case where you added six and lost five, which is a completely different business situation from adding one and losing none even though both print the same growth rate.

credentials: 'include' was missing. Added the other three too. Working. I had convinced myself the browser sent cookies automatically because that's what it does everywhere else.

Fold it in, raise new signups, grandfather the existing 140 explicitly and tell them you are doing it. I did exactly this and the email announcing that existing customers keep their price and get the new module free was the best received thing I have ever sent. Two replies asking to pay more, which I did not take, and zero churn that month. You get the goodwill now and the higher price forever on new business.

Changing my behaviour helped more than understanding the window. I stopped using the free tier conversationally, no "hi, can you help me with" followed by six clarifications: and started writing one long message with the code, the error and the constraint in it.

Same amount of help, roughly a quarter of the requests. Caps are counted per message, not per word, on every free tier I've used.

Three of eleven on one topic is a product bug wearing a support costume. Fix the scheduled runs screen and you've bought back a day a month without automating anything.

Do both. Pre-do the setup, keep twenty minutes of watching, and record it. You get the research and the checklist out of the same call.

Record every one of the next ten calls, with permission, and then watch them at speed with a notepad. You'll find the same four or five moments where you say the same sentence.

Mine were: which data source to pick, one confusing permission screen, a naming convention question, and "what should my first workflow be." I turned those into a five-step in-product checklist with a two-minute clip on each. The call went from 45 minutes to about 15, then to optional, then to nothing over roughly four months.

The recordings are the work. The automation is easy once you know what you're automating, and you cannot guess it accurately from memory: I tried and got two of my five wrong.

Promise business days, define your business days on the same page, and put your timezone next to them. Three sentences of specificity removes almost all the disappointment.

My actual stack for exactly this kind of site, annualised: a mid-tier host that does not oversell, the domain at a registrar that charges the same every year, email on a proper mail provider rather than the host, and an off-site backup that does not live in the same account as the site.

The email line is the one people leave out of the quote and it is the one that causes the most grief. Host-provided mailboxes on cheap shared plans are consistently the worst part of the package, deliverability is poor, and moving mail later while the site stays put is a much bigger job than moving the site.

The care plan approach also fixes the who is responsible when it breaks question, which is worth more than the margin. Ambiguity there is how you end up doing free work at 11pm.

Worth checking whether that payload is big enough to be TOASTed as well. If it is, each row is potentially an extra fetch from the toast table and it never appears anywhere obvious.

Whichever you pick, budget three of your ten days for auth, deploys and the boring parts. Every two week estimate I have ever made was really an eight day estimate with two days of surprises bolted on.

Also make sure you can get your data out in a plain format before you commit. You will move again in three years, everyone does, and the difference between a markdown export and a proprietary blob is the difference between a Sunday afternoon and never bothering.