AI Consulting vs Doing It Yourself: How to Pick the Right Route

A business owner emailed me last month with a quote from an AI agency attached and one line written above it: could I just do this myself? That is the real question hiding behind AI consulting vs doing it yourself, and almost nobody selling the service will answer it straight, because the honest answer tends to cost them the sale.

I sell this work too, so read me with the same pinch of salt. But I have watched enough owners spend money they did not need to spend, and lose weekends they did not need to lose, that I would rather hand you the version with the sales gloss scraped off. Here is how the two routes actually compare, what each one really costs, and the simple test that tells you which one the task in front of you deserves.

AI consulting or doing it yourself: the short answer

The right choice depends on the task, not your budget or how technical you feel. For a repetitive job that lives inside one or two tools and would be annoying rather than expensive if it broke, build it yourself. For a job that is customer facing, spans several systems, or would cost you real money the day it fails quietly, pay someone who has done it before.

Most owners get this backwards. They assume the hard, important sounding work is what they should hire out, and the small stuff is beneath a consultant. In practice the small stuff is where you win fastest on your own, and the genuinely hard stuff is the only thing worth a fee.

So the phrase AI consulting vs doing it yourself is a bit of a trick. It sounds like you join one camp or the other. You do not. It is a decision you make one task at a time, and most weeks the answer is different.

What each route actually costs you

Doing it yourself is not free, and hiring is not just the invoice. The cost of the DIY route is your time and a few subscriptions. The cost of hiring is the fee plus whatever dependency comes attached to it. Here is the honest comparison, including the third route I will come back to.

The trap on the DIY side is opportunity cost. If your hour is worth 80 pounds and a first automation eats forty of them across three weekends, you have spent over three thousand pounds of your own time to avoid a subscription. That cost is real even though no invoice ever names it.

The trap on the hiring side is the mirror image. A fee that looks clean on the invoice, then a monthly retainer for changes you could make in ten minutes if anyone had bothered to show you how. The software underneath both routes is the cheap part. Claude Code comes included with a Claude Pro plan at around 17 pounds a month, and it is the same tool a good consultant would quietly reach for anyway.

When doing it yourself wins

Build it yourself when the task is boring, repeats often, and lives close to you. That covers most admin, which is exactly the work owners assume is too small to bother with.

Think about the jobs that actually leak your evenings. Tidying a messy customer export before it goes into your CRM. Drafting the same enquiry reply for the ninth time this week. Rebuilding the monthly report from four different places. Categorising invoices so the bookkeeper stops chasing you.

You already understand these, and that understanding is the whole point. You can describe them precisely, and describing a task precisely is most of the work. Claude Code lets you write that description in plain English, watch it show you a plan before it touches anything, and check the result against the original. An afternoon on a task you know well gets you a first working automation. If you want a structured way to spot which of your own tasks is worth starting with, the AI Opportunity Scorecard walks you through it in about two minutes and estimates the hours sitting in your processes.

When paying a consultant is worth it

Pay a consultant when the problem is already scoped and still hard. That is the honest boundary, and it is narrower than the market wants you to believe.

The jobs that earn a fee tend to share a shape. Two systems that were never designed to talk and now must. Messy data that needs real judgement to untangle before anything can be automated. A process whose rules live entirely in one person’s head. Compliance work where being wrong is expensive. Anything customer facing that fails silently and reaches a client before you ever see it.

The test is not how impressive the project sounds on a call. It is whether a mistake would cost you real money, and whether you would still be stuck after a fortnight of honest effort. When both are true, someone who has already made those mistakes on somebody else’s budget is cheaper than making them on yours.

Why most projects fail whichever route you pick

Whether you build it or buy it, the same thing sinks most AI projects: chasing something generic and clever instead of embedding one real task. The route is not what usually kills it.

MIT’s GenAI Divide report in 2025 studied hundreds of enterprise AI efforts and found that 95 percent of pilots delivered no measurable return. The failures were not caused by weak models. They came from slick tools that demoed beautifully and turned brittle the moment they met real work, while the small number that paid off were the ones wired deeply into a single high value workflow.

Read that the right way and it reframes the whole question. A consultant does not save you from the mistake that actually matters, which is picking a vague, ambitious project instead of one painful task and doing it properly. Choosing the right first job is the thing that decides whether any of this works, and no fee can outsource that choice for you.

The third option most owners miss

There is a route between the two that most of the market keeps quiet about: have someone build the first automation with you, not for you. It is the answer I argue for, and not out of modesty.

Here is what it looks like in practice. The accounts stay in your name and your billing. You sit on the screen share and watch the thing get built. The instructions are written down in plain English so the next change does not need a phone call. Your team makes an edit before the session even ends, so you know they can.

An owner who has built one automation this way never again pays a retainer for a change they could make themselves, and they hire far better on the day they genuinely need to. That is the trap the pure done for you model sets, and it is the one I built my own service to avoid. If you want the longer version of why I work like this, it is on my about page.

How to decide in ten minutes

Write the task down first, in one plain sentence, then put two questions to it. This is the fastest honest filter I know, and it keeps the decision on the task where it belongs.

Notice that none of this is about you. Feeling non technical is not a reason to hire out a job a temp could describe in four sentences, and being confident with software is not a reason to take on a compliance build that could bite a client. The task tells you the route, if you let it.

So the honest answer to AI consulting vs doing it yourself is that you will end up doing both, on different jobs, and the real skill is telling them apart. Start with the boring task you already understand, build it yourself, keep the instructions in a file so it survives, and let that first win decide whether you want the next one.

When you hit the job that is genuinely hard, or you would simply rather not start from a blank screen, that is what the AI Opportunity Audit is for. It is a fixed fee session where we take your best task, build it live together, and leave the working thing and the skill with you rather than a monthly bill. Whichever way you go, aim for the same finish line: you owning the thing, not renting it back from anyone, me included.