Often you can — and you should try first. For $20–$25 per user a month (ChatGPT Business) plus $12–$70 a month (Make or Zapier), a small business can automate a lot without hiring anyone. Bring in a consultant when the workflow crosses systems, when a wrong answer costs money or compliance, or when you have already tried and it didn’t stick.
Honest answer: a lot of what small businesses want from AI does not need a consultant. A ChatGPT Business seat is $20–$25 per user per month, Zapier starts at $19.99 a month, Make at $12, and between them a motivated owner can automate a generic, single-system task in an afternoon. The cases that do justify paying someone $8,000–$35,000 are specific — and you can usually tell which side of the line you are on in five questions. The table shows the two options with real prices.
| DIY: ChatGPT + Zapier / Make | Consultant (Cerebrum as the example) | |
|---|---|---|
| What it costs | ChatGPT Business $20/user/mo annual ($25 monthly), min. 2 seats · Zapier free 100 tasks/mo, Professional from $19.99/mo (750 tasks) · Make Core $12/mo (10,000 ops) | Audit $2,500–$5,000 (1–2 wks); build $8,000–$35,000 (3–6 wks), fixed price |
| Good for | One system, a generic task, low volume, output a person checks anyway: drafting replies, summarising, moving a form entry to a sheet, categorising | Workflows that span 2+ systems, exceptions that are the job, money or compliance in the loop, volume where accuracy is real money |
| Who maintains it | Whoever built the Zap. When they leave or the app changes, it breaks quietly | Documented, handed off, 30-day check-in, retainer if you want one |
| Typical failure mode | Works in the demo; nobody owns it; exceptions pile up in someone’s inbox | Wrong problem chosen — which is why we start with an audit |
| Time to value | Days | 4–8 weeks including the audit |
People are already doing it. Microsoft’s 2024 survey of 31,000 workers found 75% of knowledge workers use generative AI at work and 78% of them brought their own tools. The U.S. Chamber found 58% of small businesses used AI in 2025, up from 23% two years earlier — overwhelmingly without a consultant. If your task lives inside one system, is roughly the same in every business like yours, and a person will read the output before it goes anywhere, buy the seat and try it for two weeks. Drafting the first reply to a common email, summarising a call, turning a form submission into a row and a Slack message, first-pass categorisation of transactions: none of these needs us, and OpenAI states ChatGPT Business does not train on your workspace data by default, which removes the most common objection.
When the workflow crosses systems. Email to your property-management software to a spreadsheet to an owner report is four hops; each is a place a no-code chain silently fails, and MIT’s 2025 review found only 5% of custom enterprise AI tools reach production. When the exceptions are the job. The self-employed borrower, the multi-location claim, the PO with three prices — the 20% of cases that cause 80% of the pain are exactly what a generic prompt gets wrong. When the data lives in heads or inconsistent PDFs. RAND lists insufficient high-quality data among the five root causes of AI failure; no amount of prompt-writing fixes an input that isn’t there. If two of those three describe your workflow, DIY will produce a demo, not a system.
Zapier’s own 2021 survey found 94% of SMB employees perform repetitive tasks and about 70% said automation made them more productive — but automation that isn’t owned decays. The person who built the Zap changes roles; the CRM renames a field; the AI step starts returning a slightly different format and nobody notices until a customer does. Consultants earn their fee less by knowing the tools than by designing who reviews what, what happens when the input is weird, and how the system is documented so it survives staff turnover. BCG’s research on 1,000 executives put the split at 10% algorithms, 20% technology and data, 70% people and processes. Most DIY effort goes into the 10%.
| Your answers | What to do |
|---|---|
| Yes to 4 | DIY, and test it for two weeks. |
| Yes to 1–3 and no to 4 | That is a build, and worth pricing. |
| No to 1 or 3 | Wait — fix the process or start capturing the data, and revisit in 60–90 days. |
Question 5 doesn’t change the answer so much as decide where the human sign-off goes; anything with money or compliance attached gets one regardless of who builds it.
Three things, none of them writing prompts. First, they find the right problem: RAND’s practitioners named miscommunication about the project’s purpose as the number-one cause of failure, and an outside audit is the cheapest fix. Second, they redesign the workflow rather than bolting AI onto it — McKinsey’s 2025 survey found high performers are nearly three times as likely to have fundamentally redesigned workflows. Third, they design the guardrails: which cases route to a person, what the system does when the input is malformed, and how the team takes it over. If a consultant’s proposal is mostly about which model to use, you are paying for the 10%.
We say it regularly. If your workflow is generic and single-system, the DIY stack above is the answer and we will tell you on the call. If the process is broken — two people describe it differently — fixing the handoff first costs nothing and often recovers more time than the automation would have. If the volume is two hours a month, a checklist wins. The buy, build or wait decision is the whole point of the operations audit, and we put buy and wait recommendations in writing next to the build ones. A fair question for anyone advising you on AI: what is the last thing you told a client not to build?
For single-system, generic tasks with a person reviewing the output — yes, and you should. It stops working when the workflow spans several systems, when exceptions dominate, or when nobody owns the automation.
Roughly $40–$100 for a small team: ChatGPT Business at $20–$25 per user (two-seat minimum) plus Zapier from $19.99 or Make from $12.
When the workflow eats 8–10+ hours a week, crosses systems, and a wrong answer costs money or compliance. At that point a fixed-price build typically pays back in 6–18 months.
A good one won’t. Ask for the system to be built inside the software you already run, documented, and handed over — and get the buy/wait recommendations in writing.
These take a while to write. One email when the next goes up, nothing else.
Book a free 30-minute operations call. No pitch, just an honest conversation about where AI can help.