Every SaaS-wrapped AI tool looks cheap in year one. The real number shows up when you try to leave -- and by then your workflows, your prompts and your staff's habits are all built around a vendor's proprietary layer. This is the switching cost nobody prices in at signup, and it's usually far larger than the monthly fee that first caught your eye.
Why the exit is more expensive than the entry
A Melbourne bookkeeping practice we spoke with signed up for a $79-a-month AI receipt-categorisation tool in 2024. Two years and roughly $1,900 in fees later, the vendor changed its pricing tier without notice, tripling the monthly cost. Moving off meant retraining three staff on a new tool, re-mapping category rules the vendor had never let them export, and re-teaching six months of learned corrections from scratch. The switch itself cost more in staff hours than the tool had in total subscription fees across its entire life.
This is the pattern: point-solution AI tools rarely let you export the thing that actually has value, which is the accumulated context, the tuned prompts and the corrected outputs. You're not locked in by the software. You're locked in by the training data and workflow habits you built inside someone else's account, and most vendors have no commercial incentive to make that easy to take with you.
What 'owning' your AI actually buys you
Prompts, instructions and corrections live in files you control, not inside a vendor dashboard you can't export from
Switching the underlying model (Claude to something else, or Haiku to Opus for a harder job) doesn't mean starting over
No pricing negotiation lost to a vendor who knows migrating away costs you more than the price rise they're proposing
Staff training transfers, because the workflow lives in a shared document or skill, not a vendor's proprietary interface
A rough way to price your own switching risk
Before signing anything, ask what happens if the price doubles in 18 months. If the honest answer involves retraining staff, rebuilding rules, or losing months of corrected output, that's the real cost of the tool, not the sticker price. For a 15-person Australian services business, that retraining cost typically runs $3,000 to $8,000 in lost billable hours, on top of whatever the new, higher subscription costs on the way out the door.
This doesn't mean every point solution is a trap. A narrow tool that does one job well, with an export function that actually works, can be the right call for a short-term need. The test is whether you could walk away in a month if you had to, and land on your feet with your workflow and your data intact, rather than starting from a blank page.
Building the same workflow so it's portable from day one
The practical fix isn't avoiding SaaS AI tools entirely, it's structuring the workflow so the valuable parts (the instructions, the examples, the corrected outputs) live somewhere you control from the start. A Claude skill or a documented prompt library sitting in your own file system does the same job as a vendor's trained model, without the exit tax if you ever need to move it. Building this way from day one costs roughly the same as building it the locked-in way; the difference only shows up years later, when the bill for staying quietly outweighs the bill for having left it portable.
If you're weighing a new AI subscription against building the same workflow on a platform you control, it's worth costing both paths properly before you sign. Get in touch through /contact and we'll run the comparison with your actual numbers.
What this looks like across a whole toolkit, not just one app
The switching-cost problem gets worse the more point tools a business accumulates. A 20-person firm running five separate AI-adjacent subscriptions -- one for email drafting, one for meeting notes, one for social captions, one for invoice processing, one for customer replies -- faces five separate exit taxes if any one vendor changes terms, raises prices, or shuts down. Each tool individually looks like a small monthly line, but the combined switching risk across five vendors compounds the same way the cost of a single locked-in tool does, just multiplied by however many subscriptions have quietly accumulated on the company card.
A Perth accounting practice we reviewed had exactly this pattern: four AI subscriptions totalling $340 a month, none individually alarming, but two of them had already changed pricing once in eighteen months and offered no data export at all. Consolidating three of the four onto workflows built on a platform the practice controlled cost roughly $9,000 upfront and cut the ongoing subscription total to $89 a month, with the added benefit that none of the resulting workflows carried the same exit risk going forward.
