Build, Buy or Configure: Choosing the Right AI Route for an SME

Every AI conversation reaches the same fork. Do you buy a product that already does the job, wire existing tools together, or have something built for you? Each route suits a different situation, and picking wrongly is expensive in either direction. Here is a plain way to decide.

The three routes

Buy. Take a product designed for your problem and use it as designed. An AI note-taker for meetings, a customer-support add-on in your helpdesk, the AI features already in Microsoft 365 or Google Workspace. You get speed and someone else's maintenance. You give up control and accept their way of doing things.

Configure. Combine tools you already have or can subscribe to, using automation platforms such as Zapier, Make or n8n, plus a model like ChatGPT, Claude or Gemini for the thinking steps. You shape the flow to your process without writing much code. You depend on the platforms and you own the maintenance of the connections.

Build. Commission custom software or an agent designed around your process. You get a precise fit, ownership and the freedom to change it. You also take on cost, a longer path to something useful, and an ongoing duty to maintain it.

Questions that decide it

Run your idea through these in order.

1. Does a product already do 80 percent of this? If yes, buy it. Accept the missing 20 percent unless that gap is the reason the project exists. Many owners over-build because they cannot tolerate small imperfections.

2. Is the task about connecting things you already use? If the real problem is data moving between a form, a CRM and Xero, configuring is nearly always the right route. This is the territory of boring automations, and it is cheap to try and easy to change.

3. Is the process a source of advantage for you? A bespoke quoting method, a unique service workflow or a proprietary data set can justify building, because a generic product would flatten what makes you different. A standard task like payroll does not.

4. How often will it run and how many people will use it? Build costs spread thinly across heavy use. A tool used by one person monthly rarely warrants custom work.

5. Who will look after it? Every option has upkeep. Bought tools change pricing and features. Configured flows break when a connected app updates. Built software needs fixes, hosting and security patches. If nobody in the business will own it and you do not want an ongoing arrangement with an agency, favour the simplest option.

6. How sensitive is the data? For personal, health or financial information, ask where data goes, who can see it and how it is deleted. Check current terms of any product. Sometimes control is the reason to build, and sometimes a reputable vendor is safer than your own code. Take legal advice for specifics.

A rough guide

Situation	Usually best
Common problem, many products exist	Buy
Connecting existing systems, rules are clear	Configure
Unique process that drives your margin	Build
Unsure the idea works at all	Configure a small version first
Needs to be a customer-facing product	Build, after validating the idea

The last two rows deserve emphasis. If you are unsure, build the cheapest version that can fail quickly. A configured prototype that teaches you the process is better than a polished build of the wrong thing.

Common traps

• Building because it feels more serious. Custom work has status. It is not automatically better.
• Buying five tools that overlap. You end up with licences nobody uses. The seat-buying problem applies here too.
• Ignoring switching cost. Cheap and quick today can be painful to leave later. Ask how you would export your data.
• Forgetting the human side. None of the routes works if staff do not change how they work.
• Skipping the brief. A fuzzy request leads to a poor fit on any route. See how to brief an agency.

Getting an outside view

A short conversation with someone who has no product to sell you can save a large detour. 21Labs, a Sydney agency serving businesses across Australia, helps owners choose the route through AI strategy, and builds when building is justified. Discovery is free, and every enquiry gets a reply within 24 hours.