Sam

On the testing question specifically: at 400 visits a month you cannot run a copy test and you should stop feeling bad about it.

A rough rule for how many visitors per variant you need is sixteen times p times one-minus-p, divided by the square of the difference you want to detect. If you convert at 2% and you want to be confident about a move to 3%, that is somewhere around three to four thousand visitors per variant. You would need most of a year and by then you would have changed six other things.

So do qualitative until you have volume. Six five-second tests told you something real in an afternoon. Do twelve more with different people, change one thing, do twelve more. That loop is faster and more honest than a test you cannot power.

Mild disagreement with the whole exercise. Five second tests measure what a stranger with no problem remembers, and your buyer is not a stranger with no problem, they are someone who has just searched for a solution at eleven at night. Those two people read a page completely differently.

Run the test with people who actually have the problem, even five of them, and the results will be less dramatic and more useful. When I did the generic version I got "it does automation" too, and it panicked me into writing a headline so specific it turned off everyone except one job title.

The cost that never appears in these threads is beans and time. Good beans at home are maybe £20 a kilo, which at 18 grams a shot is roughly 36p a drink, plus milk, plus about four minutes a morning including cleaning. Still enormously cheaper than £3.20, but the four minutes is the thing that actually decides whether people keep it up.

Build the escape hatch either way, on day one. Store every notification body and every purchase token in your own database, raw, forever. It is one table.

Then leaving any vendor is a migration you have already done the hard part of, and the decision stops being a lock-in question at all. This also happens to be the thing that saves you when the vendor has an outage on a launch day.

The 'things are going well so I'll wait' feeling is the entire reason bands exist. Rebalancing always feels wrong: you're selling the thing that's working. Setting a mechanical rule and following it removes the monthly decision, which is the actual product here.

And ask for a notice period in writing. Even 30 days changes the risk profile of putting all your hours in one place.

Change one variable at a time and give it four weeks. Otherwise in three months you will have changed everything and learned nothing about which change worked.

Do not attempt a big merge. Pick the fork you like least, diff it against whichever one you want as the base, and classify every single difference into three buckets: configuration, genuinely different behaviour, or accident. In my experience the split is roughly 70% accident, 25% configuration, 5% genuinely different, and the accidents are things like a hardcoded date format and a renamed column that existed for no reason anyone could remember.

Then: accidents get deleted, configuration becomes rows in a per-tenant settings table, real behaviour goes behind a flag. Move that one client onto the base with their flags on, watch it for two weeks, then do the next one. Took me about four months at a few hours a week and nobody noticed a single cutover.

The stock problem is the real problem and it is worth solving separately from the brand problem. Whatever you land on, find out whether the importer has a direct online store rather than relying on two local shops, because the shops are the ones running dry, not the supply chain. I switched to ordering directly in three-bag lots and the price per kilo came down as well.

This is what I did with a similar rate. Also check for early repayment penalties before you overpay - some loans charge and it changes the maths.

One thing nobody mentions is what happens if you retire early and then want to go back for a year. I did, at 54, and the annoying part was not tax, it was that I had already started a payment arrangement and had to be careful not to break it. Whatever you pick, ask what it costs to reverse before you start it, and take the answer to an accountant who has done this for other people rather than to a forum.

Small thing that saved me a lot of squinting: name groups after the audience, not the feature - (public), (signed-in), (admin). When you have eleven of anything the names carry more weight than the structure.

Good point. One holds about 500 and the other about 3,600, so there's more difference than I assumed from the top ten.

It's 0.07 vs 0.22 on about £26k, so roughly £39 a year. Sounds like stop-buying is the right call rather than selling.

Your instinct is right. The account is tied to your legal entity, your identity verification and your tax details - either the buyer acquires the entity (which drags in every other liability you have), or you do an asset sale and migrate the data.

What migration actually means: there is a supported path for copying the sensitive card data between accounts - customers, cards, payment methods, bank accounts - and it preserves the customer ids, which is a bigger deal than it sounds because it means the buyer's database can keep referencing the same ids. What it does not copy is charges, invoices, subscriptions, products, prices, coupons or events. Subscriptions have to be recreated on the new account from your own records, and anything that is past due cannot be recreated cleanly because you cannot create a subscription whose billing date is in the past.

So the runbook is: copy the payment data, recreate products and prices, recreate subscriptions with the correct next billing dates, repoint webhooks, and reconcile line by line.

Do 60 and stop billing on day one of the 60.

The notice period costs you nothing if there is no money moving. Nobody in the history of software has been angry about too much notice; plenty of people have been furious about being charged for a lame duck. Two months of free service also buys a lot of goodwill from people who might otherwise be loud about it.

Can confirm. Took about six weeks of just being useful before the first enquiry, then three in a fortnight.

Open a separate bank account for it today. Free ones exist. It costs nothing and it means next year's version of this question doesn't happen, because the account statement is the record.

If it's inside a tax-sheltered account, none of the tax tracking matters and you can ignore all of this. Worth checking which account it's in before you build a spreadsheet.