Build, Buy, or Wait: The Small Business AI Decision Tree

Build, buy, or wait. Every AI idea you have lands in one of those three buckets. Get the bucket wrong and you burn more money than you ever would picking the wrong tool inside the right bucket. Most of the AI misfires I see come down to this: you default to "buy" when the smart move was to wait, or you sit on "wait" when you could've bought something simple that morning.

So here's how I sort it with clients.

Path 1: Build

Build means you make your own AI workflow inside the business: a prompt template, a Zapier automation, a Claude project tied to your data, a custom GPT, a little Python script that calls an API. You don't need engineers for this. You need someone who'll sit down and set it up on purpose.

Build when all four of these are true:

1. The workflow is unique to your business. Off-the-shelf software won't fit, because your inputs, outputs, or constraints don't match what everyone else does.
2. The tools to wire it up exist today. Claude projects, ChatGPT custom GPTs, Zapier with AI steps, Make with AI nodes, n8n, Cursor. These have dropped the bar way down. If you can build the workflow with these and a few hours of attention, it qualifies.
3. You'll use it more than ten times. Under ten uses, building costs you more per use than just doing it by hand. Over ten, the math flips your way.
4. Someone here will own it. A built workflow with nobody watching it is a dead workflow within ninety days. If no one's accountable, don't build it. Buy something and let a vendor carry the upkeep.

When all four hold, build. It costs you time, usually two to eight hours per workflow, and you walk away with something fitted to exactly how you work.

Examples that should be built

• A weekly summary of your support inbox, categorized by issue type and tagged with sentiment, sent to your operations lead every Friday morning.
• A draft response generator for a specific recurring email type (proposal follow-ups, reschedule requests, refund inquiries) pulled from a corpus of your past responses.
• A meeting prep brief that pulls the customer's last three emails, your CRM notes, and any outstanding open items into a single page before each call.

Each one is small, specific, and shaped by how your business actually runs. Too narrow and too custom for anyone to sell you off a shelf. That's exactly what build is for.

Path 2: Buy

Buy means you pay for a tool or platform that handles a whole category of work, and you go along with that vendor's opinions about how it should be done.

Buy when any three of these are true:

1. The category is mature. At least three solid vendors are competing on features, pricing, and support. Mature categories come with stable pricing and known limitations.
2. Your needs are common. You're solving this the same way a thousand other businesses solve it. Customizing it would be nice to have, not core to the value.
3. You don't want to maintain it. You're paying the vendor to handle updates, security, integrations, support, and infrastructure. That's the whole deal.
4. The total cost, your time included, comes in under building. A $50/month tool that takes 30 minutes to set up almost always beats a free build that takes 8 hours to stand up and 30 minutes a month to babysit.

Where this goes wrong: you treat "buy" as "subscribe." You sign up with no plan for how it plugs in, nobody owning it, and no way to tell if it's working. Do that a few times and you've got the AI subscription bloat that eats most marketing budgets. (See the real cost of your AI tool stack for what that looks like line by line.)

Examples that should be bought

• Email marketing with AI segmentation and send-time optimization. Mature category. Klaviyo, Mailchimp, ActiveCampaign all do this well. Don't build.
• AI-assisted scheduling. Cal.com, Reclaim, Motion. Buy.
• Transcription and meeting summaries. Otter, Fireflies, Granola. Buy.
• Image generation for marketing assets. Midjourney, Adobe Firefly, Canva's AI tools. Buy.

When a thousand businesses need the same thing, vendors compete to do it well.

Path 3: Wait

Wait is the path you probably dodge, because sitting still feels like falling behind. Most of the time it's your smartest move. You let the market spend the money on expensive R&D, you watch a few waves of vendors crash and burn, and you buy once there's a clear winner with a product that holds still.

Wait when any two of these are true:

1. The technology is too new. Three vendors, all under 18 months old, all changing their pitch every quarter. The category hasn't settled. Whoever you pick today gets acquired, pivots, or vanishes inside eighteen months.
2. Your data isn't ready. AI runs on data. If your CRM is a mess, your support tickets aren't tagged, or you don't have a body of content yet, no tool is going to save you. Clean up the data first, then look at AI in a couple of quarters.
3. The use case is more of a daydream than a real job. "Wouldn't it be cool to have an AI agent that..." If you don't already do this by hand today, AI isn't the thing to invent it. Get the manual version working, then automate it.
4. The cost-to-value math doesn't pencil. A $400/month tool that saves you 30 minutes a week is a bad trade for a small business. Wait for the price to drop, the value to grow, or the job to get clearer.

Does waiting feel passive? Sure. You're still doing something. You're saving budget, attention, and your team's energy for the categories where AI pays off right now, and letting the shaky ones sort themselves out.

Examples that should be waited on

• "Autonomous AI agents that run your business." Too new, too unreliable, too oversold. Revisit in twelve months.
• AI-generated video for marketing. The output quality is improving rapidly but the tools change every quarter. Unless video is core to your model, wait six months and see who's left.
• Custom enterprise-grade AI strategy platforms. Most of these are venture-funded vendors looking for early adopters to subsidize their development. Let larger companies pay that bill.

How the three paths interact

A real AI plan uses all three. Most businesses I work with land on a mix:

• Three or four bought tools for the common, settled categories. Email, scheduling, meeting prep, transcription.
• Two or three built workflows for the stuff that's specific to you. Weekly digests, custom drafting, internal triage.
• A wait list of five or ten things that look interesting but aren't ready, checked each quarter.

Running the decision tree

Next time an AI idea comes up, walk it through this:

1. Can a mature vendor with three solid options handle this? If yes, buy. Stop here.
2. Is this unique to my business, and can I wire it together with tools I already have? If yes, build. Block off four to eight hours, name the owner, set a thirty-day review.
3. Otherwise, wait. Put it on the quarterly list with the one thing that would change your mind: a vendor matures, your data gets cleaner, a price drops, a use case gets obvious.

The owners getting the most out of AI this year are the ones who got their build/buy/wait split right early and stuck with it. Get it right and AI takes the low-leverage busywork off your plate, so your hours go to the work that needs you.