Every week another inbox offer promises to run your marketing, your admin or your bookkeeping with AI while you get on with the actual job. The offers are not fake. Some of them are very good. But the honest answer to "should I pay for done for you AI services or do it myself" is not the same for every business, and it is not the same for every task inside one business.
This guide sets out where the line usually falls, using published UK survey data rather than vendor claims.
What does "done for you AI" actually mean?
The phrase covers a very wide range. At one end sits a freelancer who writes better prompts than you and sends back finished drafts. At the other sits a full build where quotes, invoices and follow-ups move between your systems without anyone touching them. Both get marketed with the same three words, which is why the first question to any provider is not "how much" but "what exactly arrives, and who owns it afterwards".
How common is this already among UK businesses?
That gap matters more than the headline. The ONS tracks business AI use through its Business Insights and Conditions Survey, and the pattern across recent waves is consistent: the share of firms using at least one AI technology keeps climbing, while the average number of technologies per adopter has barely moved. Adoption is spreading sideways, not deepening. Use is also far higher in large businesses than in small ones, and lowest in trades such as construction.
So the competitive picture is not "everyone is miles ahead". It is "lots of people have opened a chatbot once and stopped". That is genuinely good news if you are late, because doing one process properly puts you ahead of most of the firms already counted as adopters.
How much time is actually at stake?
The same research points to owner capacity, rather than demand, as one of the biggest brakes on growth, and to a sizeable minority of owners working far beyond a standard week. Treat the exact percentages as indicative and the direction as reliable: admin volume, not ambition, is what most owners are short of.
That is the number to hold in your head when a provider quotes you a monthly fee. The question is not "is this cheap" but "does this reliably return more hours than it costs me, at my own charge-out rate". If you want a task-level breakdown of where those hours usually hide, the walkthrough on where AI cuts admin hours in a small business is a useful starting map.
What does the DIY route really cost?
For an anchor on the software line, Claude's published pricing lists Pro at 20 US dollars a month billed monthly, or 17 US dollars a month on the annual plan, with a free tier available. Check it before you budget, since vendor pricing moves. Comparable consumer plans from other vendors sit in a similar band. Against roughly 11 hours a week of admin, that subscription is not the decision.
The decision is your learning curve. Budget realistically: a weekend to get fluent with one tool, then two to four weeks of part-time fiddling per process before the output is consistent enough to trust unsupervised. If that sounds fine, DIY is the right call and you will end up with a system you actually understand.
Which tasks are genuinely worth doing yourself?
That covers a lot of a normal week: summarising meetings, drafting replies to routine enquiries, tidying supplier emails, turning a voice note into a job sheet, first drafts of social posts. It also covers most review chasing. The process for automating Google review requests in about 30 minutes is a fair example of something no business should pay a retainer for.
Keep these in-house for a second reason too. Doing them yourself teaches you what AI is actually good at, which makes you far harder to oversell later.
Which tasks are worth handing to someone else?
Three categories come up repeatedly for UK small businesses:
- Anything crossing systems. Quote to job to invoice to chase, moving across three different pieces of software, is where DIY builds tend to stall and where a professional build earns its fee.
- Anything with a compliance edge. VAT, payroll adjacent work, client data under UK GDPR. Not because AI cannot help, but because the review process around it needs designing by someone who has done it before.
- Anything published under your name at volume, where quality and consistency both matter and a bad month is visible to customers.
The comparison between an AI admin assistant and a virtual assistant is worth reading alongside this, because for some owners the honest answer is that neither AI route wins and a part-time human is the cheaper fix.
Does paying for AI content risk your search visibility?
Read Google's guidance on generative AI content before you sign any content retainer. Nothing in it prohibits AI. What it targets is the pattern of mass-produced pages with nothing original in them, and the policy applies the same way to human-written filler.
The practical filter for a content provider is simple. Ask where the first-hand material comes from. If the answer involves interviewing you, using your job records, your pricing, your customer questions and your photographs, the AI is doing drafting work and that is fine. If the answer is a keyword list and a publishing schedule, you are buying the exact thing Google describes. The cost comparison in AI SEO content versus hiring a writer covers the money side of that choice in more detail.
How do you compare two offers that both say "done for you"?
| Question | Good answer | Warning sign |
|---|---|---|
| Who owns the accounts? | Your business name, your billing, your logins | Provider holds everything |
| Who owns the prompts and templates? | Handed over in a document you keep | "That is our proprietary IP" |
| What if a tool changes? | Named as included maintenance | Silence, or billed as new work |
| Can you exit? | System keeps running, you get a handover | It stops when they stop |
| Can you start small? | A paid pilot on one process | Twelve-month retainer only |
Two more habits are worth adopting. Ask to see real output from a comparable UK business rather than a case study slide, and ask what the provider decided not to automate and why. Anyone who says everything can be automated has not run a system in production for long.
What does a sensible first move look like?
Concretely, over the next month:
- Write down every repeating admin task and the minutes it costs each week. Against an admin load of roughly 11 hours a week, many owners find three tasks account for over half of it.
- Take the single most repetitive one and attempt it yourself with a free tier for two weeks. Keep notes on where it fails.
- If it works, keep it and move to the next task. If it half works and the failures are all about connecting systems, that is your candidate for paying someone.
- Whichever route you take, write the process down. The documentation, not the tool, is the asset. It survives a vendor change, a price rise and a staff departure.
The uncomfortable truth in the ONS numbers is that most UK businesses using AI are using it thinly, fewer than two technologies each. Depth on one process beats breadth across ten, whoever does the building.
What is the honest summary?
Neither route is a status decision. The businesses getting real value are not the ones who bought the most, they are the ones who picked a single expensive-in-hours process, made it work properly, wrote it down, and only then moved on to the next one.
Done — FAQ
What does a done-for-you AI service usually include?
There is no standard definition, which is exactly why you should ask for the deliverable list in writing before paying anything. Most credible offers include four things: a discovery pass that maps the tasks eating your week, a build phase where prompts, templates and connections between your existing tools are set up, a handover where someone in your business is shown how the system runs, and ongoing maintenance when a tool changes its interface or pricing. What is often missing, and what you should insist is named, is who owns the accounts and the prompts at the end. If the provider holds the logins and the instruction files, you are renting your own process. Ask for accounts in your business name and a copy of every prompt and template in a document you control.
Is DIY AI actually cheaper than paying someone?
On the software line, yes, and it is not close. A single consumer subscription is modest money. Claude's published pricing at the time of writing lists Pro at 20 US dollars a month billed monthly, or 17 US dollars a month on the annual plan, and the free tier is enough to test whether a task suits AI at all. The cost that decides the question is your time. Survey work on UK small firms, including the American Express SME Business Barometer, has put owner admin and finance time at roughly ten hours a week. If learning and building eats several of those hours a week for two or three months, price that honestly against your own charge-out rate. DIY is genuinely cheaper when the task is bounded and you enjoy the tinkering. It stops being cheaper when the build keeps getting pushed to Sunday evening and never finishes.
Will using AI to write content damage my Google rankings?
Not because it is AI. Google's own guidance on generative AI content says that using AI tools to generate many pages without adding value for users may violate its spam policy on scaled content abuse, and it asks publishers to focus on accuracy, quality and relevance, especially when content is generated automatically. The trigger is thin, mass-produced pages, not the tool that made them. In practice that means AI is safe for structure, research and first drafts, and dangerous when it is used to publish twenty near-identical location pages nobody reads. If a provider promises a fixed number of posts per month with no named author, no first-hand detail and no review step, that is the pattern Google describes.
How do I know whether a task is a good candidate for AI at all?
Three tests, applied before you spend anything. First, does the task repeat at least weekly, with the same shape each time? One-off work rarely repays a build. Second, can you write down what a correct output looks like in a paragraph? If you cannot describe the standard, no system can hit it. Third, what happens if it is wrong? Chasing an unpaid invoice with slightly clumsy wording is recoverable. A wrong VAT figure or a wrong quote is not. Start with high-frequency, low-blast-radius work such as meeting summaries, review requests and first-draft replies. Keep human sign-off on anything that touches money, contracts or a client's personal data.
What should I ask a provider before signing anything?
Ask who owns the accounts, prompts and data at the end of the contract, and get the answer in writing. Ask what happens when an underlying tool changes its pricing or removes a feature, because that will happen within the contract term. Ask for a named example of the same build running in another UK business, and ask to see the actual output rather than a case study slide. Ask what the exit looks like: can your team keep running the system if the provider disappears, or does it stop working. Finally, ask for a small paid pilot on one process rather than a twelve-month retainer. A provider confident in the work will price a pilot. One that refuses is selling access, not outcomes.
