Here’s a number that stopped me cold: 58%.
That’s the share of organizations that attempted to migrate between AI platforms and said the process either failed outright or required significantly more effort than expected — according to a Zapier survey of 542 U.S. executives with active AI contracts. And the kicker? Nearly 90% of those same executives believed they could switch vendors in under four weeks. 41% thought they could do it in 2 to 5 business days.
That gap — between what leadership thinks is possible and what actually happens — is exactly the problem. And it’s not an IT problem. It’s a vendor risk problem. One that belongs on the desk of every finance and accounting team that has started embedding AI into their work.
We’ve Seen This Movie Before
Think back to the enterprise software wave of the 90s and 2000s. Companies bought into SAP and Oracle because the integrations were seamless and the pitch was compelling. Within a few years, those platforms were woven into finance, HR, and supply chain. The moment anyone tried to leave, they found out the hard way: the system wasn’t just running their data — it had become their institution’s memory. Switching costs were enormous. Years of effort, millions of dollars, hundreds of consultants. Some companies are still mid-migration today.
AI is following the exact same script. It’s just moving faster.
The Alteryx Playbook
If you work in accounting or finance, you probably know Alteryx. For years it was the go-to no-code ETL and workflow automation tool. Teams built their entire reporting infrastructure on it. They trained staff, documented processes — and then didn’t document the workarounds and adaptations that piled up over time.
Then two things happened. Prices started climbing at every annual renewal. And in 2024, Alteryx was taken private in a $4.4 billion deal by Clearlake Capital and Insight Partners. From what I’ve heard from clients who are still in the ecosystem, renewals have become increasingly contentious — more bundling, more pressure, less flexibility. Once you’re deep in the platform, your negotiating position isn’t what it used to be.
The Zapier report asked: what happens when the AI you depend on gets acquired by a private equity firm that strips it for parts? That’s not a hypothetical. We just watched it happen.
Three Things Happening Right Now
Migration is harder than anyone expects. AI implementations require vendor-specific APIs, proprietary tooling, custom training data, and workflow integrations that don’t transfer cleanly. But beyond the technical layer, there’s the institutional layer — the custom prompts, the edge cases, the undocumented adaptations your team made because they were “temporary.” That stuff is sitting inside your current vendor’s ecosystem. It is not portable.
Prices are rising now, materially. OpenAI raised pricing on its flagship model significantly earlier this year. Anthropic moved its enterprise edition from fixed pricing to usage-based in April 2026 — analysts think that could double or triple costs for heavy users. GitHub Copilot has closed new individual subscriptions and dropped access to its most powerful models. The all-you-can-eat era is over. Token-based pricing is the new normal.
Open source isn’t the safe harbor everyone thought. Meta’s Llama was supposed to be the escape hatch — run it on your own hardware, avoid the vendors entirely. Except it was never truly open source, and Meta is now repositioning toward proprietary solutions. Teams that built on Llama thinking they were free from lock-in are discovering the same dependency problem from a different angle.
What This Means for Finance and Accounting
Two things need to change in how your team thinks about AI.
First, AI vendor dependency needs to be in your vendor risk framework — right alongside supplier concentration and customer concentration. Most organizations can answer what their revenue concentration looks like by top-10 customers. Very few can answer what it would cost, in time and money, to exit their AI platform in 90 days. That question belongs in your risk register. And it’s a finance question, not an IT question.
Second, if your team has signed off on AI budgets built on flat-rate or fixed pricing assumptions, those models need to be revisited. AI doesn’t behave like SaaS. With SaaS, costs tend to fall as you scale. With AI inference, the opposite is true — every query has a real cost, every agent run costs tokens, and the more you use it, the more you pay. Build in a 30 to 50% cost increase as a baseline planning assumption over the next 24 months.
Key Takeaways
- Map your AI dependencies now. Which workflows are AI-dependent? Which vendors? What undocumented customizations exist?
- Run a switching cost estimate. What would it actually cost — time, money, productivity — to migrate to a different provider? That number needs to exist.
- Revisit your AI cost models. Fixed pricing is going away. Plan for usage-based costs that scale upward with consumption.
- Treat AI vendor concentration like supplier concentration. Single-vendor dependency is a risk item. Even a small secondary deployment preserves optionality and negotiating leverage.
- The Alteryx playbook is the AI playbook. Adoption → deep integration → price increases → PE acquisition. Build exit optionality before you need it, not after.
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: [lesson link]
Leave a Reply