Scorpion AI Tooling has been trading for a little over a month. The first invoice went out on 14 July 2026, and the company is registered in England and Wales as Scorpion AI Tooling Ltd, company number 17336409, based in Newcastle-under-Lyme.
This is not a launch announcement. It is an account of what the first month actually contained, with the specifics attached, because a supplier nobody has heard of asking you to trust them ought to show some working. Most of what follows is unglamorous, and some of it is me getting things wrong.
Getting to the point where you can take money
There is a gap between deciding to start a company and being able to invoice somebody, and it is filled almost entirely with admin. Incorporation at Companies House. A business bank account. A registered office. An email address on the company domain rather than a personal one. Registration with the Information Commissioner's Office, because the business handles other people's data. Terms and conditions, a privacy notice, and a contract template that says who owns the work at the end of it.
None of that wins a customer. All of it has to exist before the first one arrives, because the alternative is improvising a contract while somebody is waiting to sign it.
If you are weighing up a supplier you have not used before, these are cheap things to check and they tell you something. A company number you can look up on the Companies House register. An email address on their own domain. A written contract offered before the work starts rather than after. None of it guarantees good work, but the absence of it is worth noticing.
The first month in numbers
As of 20 August 2026:
- Five businesses on the client list, three of them invoiced so far.
- Ten invoices raised. Two were voided and replaced rather than deleted, one of them because I raised the same invoice twice in the same afternoon.
- Most of what has been invoiced has been paid. The rest is still inside its payment terms, and nothing has gone past a due date yet.
Those are small numbers, and they are the real ones rather than a rounded version of them. A company a month old is a company a month old. The work in progress is worth more than what has actually been collected, but work in progress is not money, and I have tried to keep the two apart in my own head as much as in the accounts.
The pattern I find most instructive sits behind those figures. I researched a great many more businesses than I ever picked up the phone to. Research is comfortable and phone calls are not, and a well-organised list is very good at feeling like progress while producing none of it.
If you are early in something similar, the measure worth watching is not how much you have prepared. It is how many real conversations you have had. The first grows on its own. The second only moves when you make it move.
The things that went wrong
Three worth recording, because the useful part of each one is what it changed. All three were corrected at the time, and none of them cost a client anything.
A quote carried an out-of-date company number. The number had been updated in one place and not in the others, and an old one was still sitting in a document template. The same quote had a second fault: the monthly retainer line rendered as £0.00, because the generator was treating a recurring charge as a one-off and could not total it. I spotted both and reissued the quote in corrected form, superseding the first, before it had been accepted or acted on. The fix was not a resolution to be more careful next time. It was to correct the document generator so a recurring charge cannot be presented as a one-off, and to put a check in place so that when a company detail changes, every document carrying it is found and updated rather than only the one in front of me.
An invoice was raised twice. Same client, same amount, minutes apart. Both already carried invoice numbers, which cannot simply be deleted, so one was voided the same day with the reason recorded on the face of it, and only the correct one stands. Voided rather than removed matters: a gap in an invoice sequence is a question an accountant is entitled to ask, and "it is a duplicate, here is the void" is a better answer than a missing number. Nothing was ever billed twice.
I drafted contracts before confirming who I was contracting with. Two agreements reached a complete draft with the client's registered company details still bracketed as unconfirmed. Neither was sent in that state, and neither was signed in that state. A contract names a legal entity, and guessing at the suffix or the registered office makes the document worthless however good the rest of it is. Both were completed from the public register before going anywhere, and confirming the entity is now the first step of drafting rather than the last.
The one I am most glad I caught
Early on I put a testimonial form together. It asked whether the feedback could be used publicly, but it asked as a tick-box near the end, which is a very easy thing to skip past on the way to signing. A signature at the bottom of a generous page is not consent to everything printed above it, and a form that lets somebody sign without answering the question has not really asked it.
So I rewrote it. It now asks the identification question directly and makes it a choice rather than a box: how the person wants to be named, whether their business can be linked, or whether they would rather the feedback stayed internal. Where the answer was not already explicit, I went back and asked the person directly rather than working it out from the paperwork, and nothing has gone on this website until they confirmed exactly how they wanted to appear. Where somebody has not yet answered, their feedback stays unpublished.
The wider point is that consent has a scope, and the scope is whatever the person actually agreed to rather than whatever the paperwork technically permits. That applies to a testimonial. It applies rather more seriously to customer data going anywhere near an AI tool, which is most of what this business does.
If you collect testimonials, reviews or case study material, ask the permission question separately from the content question, and keep the answer somewhere you can find it later. A signature at the bottom of a page is not consent to everything above it.
Publishing the research rather than describing it
Across July and August I ran a technical health check on the websites of 88 independent businesses in one area of Staffordshire — whether the site loads, whether it works on a phone, whether it is served securely, whether a search engine or an AI assistant can make sense of it.
The findings are written up in a separate post, and the full anonymised dataset is published alongside it under a Creative Commons licence, with the per-criterion scores included so the totals can be checked rather than taken on trust. Sixteen of the 88 had no working web address at the time of the check, which was not the result I expected to find.
Publishing the data was a deliberate decision. Anybody can write that they surveyed some businesses and found something alarming. Publishing the rows means somebody can disagree with the method and show why, and a claim that can be checked is worth more than one that cannot.
Where things stand
Five businesses on the client list, work in progress on several of them, and a research list considerably longer than my record of having done anything with it. The company is self-funded, priced close to what the work costs to deliver, and small enough that the person who scopes a project is the person who builds it.
If there is one thing I would carry into the second month, it is that observation about preparation. The parts of this that felt like progress and the parts that produced income were not the same parts, and the distance between them was larger than I expected.