Why £18k of AI Spend Goes to Waste (And What to Do Instead)

Do not automate what you cannot describe

Nate B Jones puts it simply: "Do not automate what you cannot describe." That is the whole insight in a nutshell.

Founders have spent decades building businesses by doing the work, not describing it. They never had to write down how the proposal process works, because they are the proposal process. They never mapped the customer onboarding flow, because they are the onboarding flow. Now we have the ability to automate almost anything, and vendors are trying to implement something that exists only as instinctive habit in one person's head.

The way I see it: the unit of AI investment is not your department. It is a single workflow. If you cannot describe that workflow in plain English, you are not ready to invest in it. You need to be able to articulate it first. You need to be able to measure the impact.

Five levers, once the work is described

Nate B Jones offers 5 levers that translate well to bootstrapped businesses, once you have done the describing.

Automate is for work that repeats, follows a clear pattern, and where the exceptions are recognisable. Follow-up emails after a discovery call. Weekly reporting that pulls from the same 4 data sources. Invoice reminders, booking confirmations. The acid test: could you write the instructions for someone who had never done it before and expect them to get it right 9 times out of 10? If yes, this is automation territory.

Build is for work that is company-specific, with too many edge cases for anything off-the-shelf. The proposal process that takes you 3 hours because every client gets a bespoke scope. The sales deck you know is just a feature list, not a business case, and you know it needs rebuilding, but there is no template that fits. The secret sauce of your business probably lives here, and that is exactly why a generic tool will not touch it.

Buy is for needs where the market has a mature solution. Scheduling meetings, payment processing, a basic CRM, document signing. The rule here is to buy primitives, not platforms. You want the smallest tool that solves the specific job. The moment a vendor starts explaining how their platform "also does" 14 other things, you are paying for complexity you will never configure, let alone use.

Hire is for gaps that need trust, framing, or standard-setting. Not for a mythical candidate who does content, paid, SEO, brand, and analytics. That person does not exist, even if your JD says otherwise. Hire for the specific gap: someone who can own marketing execution so the founder stops being the bottleneck. When sales blames product and product blames sales, what is usually missing is a person with the authority and clarity to set the standard.

Wait is not laziness. It is a valid tactic. Defer implementing lower-priority workflows when the market is still maturing, or when your change management capacity is already stretched. That shiny new AI platform everyone is talking about at conferences? Wait. Not because it is bad, but because adopting it right now might cost more in disruption than it saves, and it might be redundant in 18 months' time.

Two questions that tell you which lever to pull

How specific is this workflow to your business? And how mature is the market for solving it?

High specificity plus low market maturity usually means "build" or "hire." High maturity plus low specificity means "buy" or "automate." When both are low, you wait. When specificity is high and the market has not caught up, you protect the workflow rather than throwing money at it prematurely.

Most founders I work with find that a single pass through their top 6 or 7 workflows produces a mix of all 5 levers. This is normal and healthy. It means you are being deliberate instead of treating every problem as a nail because someone sold you a very expensive hammer.

The Monday exercise

Here is what I would actually do if I were reading this article instead of writing it.

Pick your highest-friction workflow. The one that breaks when you go on holiday. The thing that only you can do, or that only you do, which is a different problem entirely. Then answer 5 questions about it.

1. What information comes in? What triggers this workflow, what data or context does it need, and where does that information currently live?
2. What is the job being done? Not the activity. The outcome. What does this workflow produce that matters to the business?
3. How does the work flow, step by step? Write it down as if you were explaining it to someone starting on Monday. You can use an LLM to interview you, or record a Loom walkthrough.
4. What tools or people are involved? Which software, which team members, which external partners touch this workflow? Where are the handoffs?
5. How do you know if the output was good? What does a good result look like, and how quickly do you know whether you got one?

If you can answer all 5 in plain English, you have done something most founders have not. You have described the work. Now you can make a deliberate choice about which lever to pull. Automate it, build something bespoke, buy a tool, hire someone, or wait.

If you cannot answer all 5, that is not a failure. That is your first investment: describing it. And honestly, it is probably the highest-return hour you will spend this quarter.

Processes do not need to be perfect, but they do need to exist

If you do not have time for this exercise at all, that tells you something about your capacity and resourcing, and that probably matters more than any AI tool. What I see as a common theme among founders is that they are too busy running the workflow to step back and look at it. That busyness is a self-enforced trap. The workflow keeps running, keeps depending on you, and keeps preventing the investment decision that would free you from it.

Every AI investment you make without a specific workflow definition is a guess. Some guesses work out, but most do not. The 40% of projects that Gartner predicts will be cancelled by 2027 are the ones that missed this step. They got excited about the possibility and bought the tool before describing the work. Or they hired the person before they defined the gap.

The ones that succeed will probably look a bit boring. They will have described the work, made deliberate choices about which lever to pull, and invested in things that actually matter. Founders who have been thinking about how to set up their AI tools properly already know that the setup is where the real value lives, not the prompt.

The founders who do well with AI are not the ones who automate the most. They are the ones who automate the right things and invest the time they free up into the work that only they can do.

We run a half-day workshop called the Discovery Diagnostic that is literally this exercise at scale: 23 decisions about how your business actually works, mapped against where the real opportunities sit. "I know I need marketing, but I don't know what good looks like" is probably the most common thing people say before they book one. If what you have read here lands, that is the conversation worth having.

And if it does not land, do the 5 questions anyway. They cost nothing but an hour, and you might be surprised what you find when you finally describe the work you have been doing on instinct for years.