Custom AI vs SaaS — what works for a small team

This is the conversation I have three times a month. Owner has heard the AI drumbeat. Owner isn't sure if they need to buy something or build something. Owner asks: what would you do?

Here's the framework we actually use — and the maths that goes with it.

01Two questions before you decide anything

Question 1: is the task generic or specific to your business?

Generic tasks: transcription, grammar-check, image resize, image search, code assistant. These are commodities. Off-the-shelf almost always wins on cost and speed-to-live.

Specific tasks: chasing invoices in your voice, drafting quotes in your language, prioritising leads by your book of business, writing the field debrief in your tradie's way of speaking. These need context that only exists inside your business. Custom almost always wins here.

Question 2: will you run it for more than 18 months?

If yes, custom's TCO usually beats a subscription. If the answer is 'we're not sure, we're testing', buy off-the-shelf until you're sure.

02The 18-month TCO comparison

For a typical Brisbane SMB workflow — say, an invoice-chase assistant for a business with 200-500 active clients — here's roughly what we've seen:

• SaaS route: $49–$149/user/month, 3-4 seats, plus onboarding time. Over 18 months: roughly $4,000–$11,000. Value: fits ~70% of workflows, ~30% of chases still need manual editing.
• Custom route: $8,000–$14,000 one-time build, hosted on your own infrastructure, no per-user fees. Over 18 months: same or less than SaaS. Value: fits ~95% of workflows, near-zero manual editing after the first month.

The bit that gets missed: the SaaS number keeps ticking forever, and it goes up when the vendor raises prices (they do, on average 8-12% a year). The custom number is one-time. In year three, the gap widens dramatically.

03Where custom loses

Custom isn't automatically the right answer. Where it loses:

• You need it running by next week (build takes 3–8 weeks).
• The task is genuinely generic and doesn't touch your business data.
• You're a team of one and won't get 15+ hours a month back from it.
• The vendor's SaaS has network effects (e.g. Xero has bank feeds you literally can't replicate cheaply).

For those cases we tell people to buy off-the-shelf and move on. Custom is a discipline you spend where it pays back — not everywhere.

04Where SaaS loses

Where SaaS quietly costs more than the number on the invoice:

• Per-user pricing that scales linearly with team growth.
• Feature bloat you didn't ask for, that changes the UX every quarter.
• Vendor lock-in — your data lives in their format.
• Integrations that break when either side updates.
• AI features grafted onto a product that wasn't designed for them (most 'AI in [tool you already use]' features).

05The hybrid answer (usually right)

The best-run small businesses we work with don't pick one or the other. They pick per-workflow. Their pattern:

• Xero for money — SaaS, always. Bank feeds and tax rules are non-negotiable.
• HighLevel or HubSpot for lead capture — SaaS, cheapest tier.
• Custom for the operational glue — reading Xero + CRM + inbox, producing the daily brief and drafting the chases.

That's the pattern that pays back. Buy the platforms; build the glue.

06How to decide, quickly

Three-minute test.

• Write the task in one sentence.
• If a competent developer could implement it in a page of pseudocode: custom is a live option.
• If it requires 300 API integrations and a bank feed: it's a SaaS play.
• If the answer is 'I'm not sure what I'm asking for': you're not ready for either yet — get a Leak Check first.

07Related reading

For the pricing side, see how much does custom AI cost a Brisbane SMB. For why off-the-shelf keeps failing here, see why bought AI tools don't fit Aussie SMBs.