DIY or Hire a Consultant? A Decision Tree
There’s no one-line answer to this question, but there is a concrete four-question decision tree: is the task a one-off or a recurring job, how costly is a mistake, how many hours a week do you realistically have and is the learning curve worth it, and does the task stand alone or does it connect multiple systems together. Answer these four questions in order, and your own decision usually becomes clear on its own.
The framework below comes out of running 43 Google Ads accounts and AI agents on my own — not theory, but what actually worked when I tried it myself versus when hiring a consultant turned out cheaper.
I’ve covered this topic before in narrative form, in a piece about when free tools are enough; here I’m turning the same question into a concrete decision tool: how far can free AI tools take you.
Question One: Is the Task a One-Off or Recurring?
This is the first fork: is the task in front of you a one-time thing (say, trying an AI image tool for a single campaign, or a one-off data cleanup), or does it repeat on a regular basis (a report prepared every week, a WhatsApp assistant answering messages every day)?
If your answer is “one-off”: try it yourself. The time you invest in the learning curve won’t need to be repeated, so the cost of that investment is low. Even if your first attempt is rough, the damage is limited.
If your answer is “recurring”: move to the next question — because in a recurring task, both the risk of mistakes and the time cost compound.
Question Two: How Costly Is a Mistake?
The second question is about the size of the risk: what do you lose if something goes wrong here? If an internal report’s formatting breaks, or a chatbot gives one weird answer, the loss is small and takes minutes to fix. But if the mistake is wrong information reaching a customer, a lost booking record, or deleted customer data, the cost is far heavier.
If the risk is low: try it yourself — the cost of a mistake is already part of the learning process.
If the risk is high: consider bringing in a consultant, at least for the initial setup — someone experienced has likely already seen this exact mistake and knows how to prevent it.
Question Three: How Many Hours a Week Do You Have, and Is the Learning Curve Worth It?
The third question is about time: how many hours a week can you realistically dedicate to learning this, and is that learning a lasting skill for you, or a one-time detour?
If you can set aside three to four hours a week and you’ll reuse this skill later, the learning curve works like an investment in yourself. But if most of your day is already filled with your own work — sales, production, customers — the real cost isn’t the consultant’s fee, it’s what those hours would have cost you by pulling you away from your own work. In that case, buying an experienced person’s time is often cheaper.
Question Four: Does the Task Stand Alone, or Does It Connect Multiple Systems?
The last question is about complexity: is the task limited to a single tool (say, feeding one piece of text into ChatGPT), or does it connect multiple systems together (linking a CRM to WhatsApp, a calendar to a payment system, an ad account to a reporting tool)?
With a standalone tool, the worst case is that you lose some time. With multi-system integration, moving forward without knowing in advance how one system affects another — especially around authorization, data formats, and error handling — usually leads to quiet problems; by the time you notice, the process may already have been running wrong for weeks.
If you answered “consultant” to two or more of these four questions, your decision is already fairly clear — leaving at least the setup phase to someone experienced and running the rest yourself is a reasonable middle path. If you answered “myself” to all four, you genuinely don’t need a consultant for this — the “Field Note” below is exactly that kind of case.
A Field Note
A few months ago, a prospective client wanted to talk to me about an AI setup to generate product descriptions for a batch of 40 items. Once we went through the questions, the picture was this: the task was a one-off (just those 40 products going live that day), a mistake was low-cost (a bad description could simply be regenerated), they had a few hours a week to spare, and the task was limited to a single tool — it wasn’t connected to any third system. I told them: “You don’t need this — just upload the product photos to ChatGPT and run the same prompt 40 times.” The same week, another conversation had the exact opposite picture: the client wanted an automation linking their booking calendar to WhatsApp and their accounting software — three systems, a task repeating every day, and a mistake meaning a double booking or a wrong invoice. There I said: “Don’t build this yourself — at least let me build the first version.” Same four questions, two different answers — the difference wasn’t the client, it was the nature of the task itself.
You Can Ask Me to Do This
If you ran the decision tree and landed on “consultant,” you can bring this to me. Scope varies with the size of the task, but it’s usually remote, hourly work spanning 2 hours to 2 weeks — I handle the setup, and you can take it from there if you prefer. Full service page: AI consulting. To get started, reach out via the contact page.
Frequently Asked Questions
I can’t give a clear answer to some of the four questions — what should I do?
Answer the question you’re unsure about by assuming “high risk.” The point of the decision tree isn’t to make you feel comfortable — it’s to make the risk you’re overlooking visible; uncertainty usually means “proceed carefully,” not “don’t think about it at all.”
Is hiring a consultant always more expensive?
No. It can look that way if you just compare the fee, but the real cost comparison isn’t the fee — it’s the value of your time plus the cost of the risk of a mistake. For a simple, low-risk, one-off task, your own time is almost always cheaper; for a complex, recurring, high-risk task, an experienced person’s time is usually cheaper.
Can I try it myself first and hire a consultant later if things get messy?
Yes, and for low-risk, one-off tasks that’s a reasonable strategy. But with multi-system integration, “try first, fix later” can be risky: some mistakes — a misconfigured data connection, for instance — take longer to clean up retroactively than they would have taken to set up correctly from the start.
Does this decision tree only apply to AI tasks?
No, the framework is general — the same four questions apply to any technical task (a website, an ad account, an automation). I’ve focused on AI tasks here because it’s the decision I run into most often these days, but the logic is universal.