Straight answers.

The AI is not. The mistakes are. Licences cost tens of pounds per person per month. What runs up a large bill is buying a platform before you have a use case, paying for seats nobody opens, and discovering nine months in that the data was never fit for it. My own fees are on the services page in full: a first piece of built work is two days at £2,000, and most builds land between £4,000 and £8,000. The point of paying for a review first is that it costs less than the wrong build.

A thousand pounds a day, and there is no VAT to add, so the quoted number is the number invoiced. The review is a fixed £1,250 and comes straight off the first invoice for work commissioned within ninety days. Built work starts at two days, quoted as a fixed number of days before it starts, never open-ended and never revised upward without a new quote. On-call support is £500 a month, cancellable on thirty days notice by either of us. Any fixed price can be paid in three monthly instalments at no extra cost, and there is no travel, no expenses and no disbursements, because everything is remote.

Probably not in any way that reaches your accounts. Around 16 percent of UK firms have deployed AI deliberately against a defined purpose. Around 70 percent have some in the building because it arrived bundled with software they already had. The gap between those numbers is where the value sits, and licences will not carry you across it.

Everyone's is, including the businesses you assume have it sorted. Preparation is commonly 40 to 60 percent of the effort on any AI project, which is exactly why the answer is not a two-year data programme. Pick one narrow area where the data is good enough, prove the value, and let that fund the tidying up.

It depends entirely on the tool and the contract, and most organisations have never checked. Some tiers train on your inputs and some do not. Some keep data in the UK or EU and some do not. It is answerable in an afternoon. There is a page on this site with the ten questions to put to any AI vendor, and you are welcome to use it without me.

Neither, and I will not pretend otherwise. What I can do is find where your exposure sits, tell you which of it is routine and which of it needs a specialist, and make sure you are not paying a solicitor to discover basic facts about your own systems. Part of the job is telling you when the next call should be to someone else.

That depends more on how you announce it than on what you build. Badly handled rollouts produce quiet resistance rather than open objection, and a meaningful share of employees admit to having worked against one. Better to have the honest conversation early than watch a good tool die of silence.

Too small, no. The cost of entry has collapsed and small firms often move fastest because there is nobody to ask for permission. Too late, also no. Most organisations are still in the shallow end, and being second with something that works beats being first with something that does not.

Because the work arrives in writing, and writing does not need a meeting. It is findings, recommendations, runbooks and recorded walkthroughs, which can be checked, forwarded, disagreed with in the margin, and read again in eighteen months when somebody asks why a decision was made. A workshop cannot do any of that. Nothing in your business stops, nobody sits in a room, and there is no travel on the invoice.

Then you have had your money's worth. A clear no, written down with reasons that survive the next board meeting, beats a project that limps along for a year. It happens often enough that you should know before you commission anything.

Phil Swindin, and only Phil Swindin. There is no team behind the website, no junior who turns up after the pitch, and no partner who sells the work and then disappears. Around thirty years in enterprise architecture, security and transformation across banking, defence, healthcare and manufacturing. CISSP since 2009. There is a fuller account of that on the about page.