Published 21 November 2026 by the Scriptguru team

Two years ago, automating almost any part of a business meant hiring someone. Today a lot of it does not. ChatGPT, Claude and Microsoft Copilot summarise meetings, draft replies and answer questions from your own documents. Zapier, Make and Power Automate connect popular apps with a few clicks. Shopify's own assistant can now build store automations from a sentence typed in plain English.

So the honest question for any business is no longer "can this be automated?" but "can we do it ourselves, or do we need help?" This guide sets out where the line usually falls, so you can decide before you spend money with anyone, including us.

What you can usually do yourself

If a task lives inside one tool, or between two popular tools that already talk to each other, start with what you already pay for.

  • Writing and summarising. Meeting notes, first drafts of emails and proposals, summaries of long documents. Any of the main AI assistants does this well.
  • Answering questions from your documents. Copilot, ChatGPT and Claude can all search files you give them access to. For a policy handbook or a product catalogue, that is often enough.
  • Simple app-to-app steps. "When a form is submitted, add a row to a spreadsheet and post in Teams." Zapier, Make and Power Automate exist for exactly this, with ready-made templates.
  • Approvals inside one system. A leave request or a document sign-off inside Microsoft 365 is a standard Power Automate template.

Someone in your team who is comfortable with spreadsheets can set most of these up in an afternoon. Paying a developer to build them is usually money wasted.

Where it stops working

The trouble starts when the work crosses several systems, involves judgement, or has to keep running without anyone watching it. These are the signs you have reached that point.

1. The data has to move between systems that do not connect well

A supplier invoice arrives as a PDF, has to be matched against a purchase order in one system, and posted to accounts in another. An order comes in by email and has to be entered, line by line, into an order or stock system. Off-the-shelf connectors often cover the popular half of the journey and not the rest, especially with accounting software, industry systems or anything older. Closing that gap means writing a proper connection, in code.

2. Exceptions matter more than the normal case

Automating the invoice that matches perfectly is easy. The value is in what happens when the total is $42 above the order, the supplier is new, or a line is missing. A good system holds those for a person with the reason attached, and handles everything else on its own. A quick automation tends to either stop silently or push the wrong number through. Designing the exceptions is most of the real work.

3. It has to keep running for years

Software you connect to changes. Login rules change, providers change what their systems accept, AI models are replaced. A workflow that someone built on a Friday afternoon can fail weeks later without anyone noticing, until a customer complains or a month-end report is wrong. If the process matters, someone needs to monitor it, be told when it fails and fix it. That is a job, not a one-off setup.

4. Nobody owns it

Many businesses already have a handful of automations built by one enthusiastic person. When that person is on holiday or leaves, nobody else knows how they work. If you cannot say who would fix it tomorrow, it is not really part of how the business runs.

5. The data is sensitive

Customer details, payroll, medical or financial information need more care: who can see it, where it is stored, which AI service it is sent to and what is logged. That is worth getting right once, properly, rather than across a dozen personal accounts.

A useful rule of thumb. If a task touches one or two systems and nobody would notice for a day if it stopped, do it yourself. If it touches three or more systems, involves money or customers, or would cause real trouble if it failed quietly, it is worth having it built and looked after properly.

Why many AI projects stall

It is worth being clear-eyed about this. Gartner has predicted that over 40% of "agentic" AI projects will be cancelled by the end of 2027, mostly because of cost, unclear value or weak controls. In our experience the pattern is familiar: a demo works on a clean example, then meets real data, real exceptions and nobody responsible for keeping it running.

The projects that last tend to share three traits. They start from one process with a measurable cost, such as hours spent retyping invoices each week. They keep a person in control of anything that matters. And someone is responsible for them after launch.

A simple way to decide

SituationUsually the right move
Drafting, summarising, answering questions from documentsUse the AI subscription you already have
A simple step between two popular appsZapier, Make or Power Automate, set up in-house
Work that crosses three or more systems, or involves money or customersHave it built properly, with exceptions and monitoring
Your process does not fit any tool you can buyConsider your own system, built around how you work

If you do bring someone in

Whoever you choose, ask for these before you sign anything:

  • A fixed price for a defined scope, after a short paid assessment, rather than an open-ended hourly estimate.
  • Everything built in your accounts, with the code and logins handed over to you.
  • Testing against your own real examples, including the awkward ones, before anything goes live.
  • A clear answer to "who fixes it when it breaks, and how quickly?"

That is how we work. But the more useful point is this: if most of what you want sits in the "do it yourself" rows above, you probably do not need us, and we will tell you so on the first call.