Nine out of ten executives think they could switch AI providers within a month. In a Zapier survey of 542 US decision-makers published in April 2026, 89% said exactly that. Yet among the two-thirds who had actually tried to move, 58% said the switch failed outright or took far more effort than expected.
That gap between confidence and reality has a name: AI vendor lock-in. It sounds like a big-company problem. It is not. If your shop has spent months training one chatbot, building custom assistants inside one app, or wiring one AI tool into your WhatsApp and spreadsheets, you are already building a dependency.
This guide explains what AI vendor lock-in is, how it quietly happens, why October 2026 is a good moment to think about it, and the simple habits that keep a small business free to change its mind.
What is AI vendor lock-in?
AI vendor lock-in is when leaving one AI provider becomes so costly, slow or risky that you stay even when a better or cheaper option exists. The "lock" is rarely a contract. It is your data, prompts, custom assistants and connected workflows living in a format that only one company's tools can read.
Think of it like renting a shop and spending HK$200,000 on built-in fittings. The lease may let you leave after a year, but walking away means abandoning the shelving, the counter and the signage. Nobody forced you to stay. The fit-out did.
AI works the same way. The monthly subscription is easy to cancel. What is hard to move is everything you built on top of it: hundreds of saved instructions, a year of chat history, a customer-service bot trained on your price list, and automations that only talk to one AI.
How does AI vendor lock-in happen to small businesses?
Lock-in builds up gradually through convenience. Each step, such as saving a custom assistant, letting the tool remember your preferences, or connecting it to your email, saves time today. After 12 months those small conveniences add up to a setup that would take weeks to rebuild somewhere else.
For a typical Hong Kong SME, lock-in usually grows in five layers:
--- Saved instructions: custom assistants, templates and "skills" that hold your tone, price rules and FAQs.
--- Memory and history: what the AI has learned about you and your customers from past chats.
--- Uploaded knowledge: product catalogues, policies and contracts stored inside one platform.
--- Connections: links to Gmail, Google Drive, Microsoft 365, WhatsApp or your booking system.
--- Team habits: staff who only know one tool's buttons and shortcuts.
None of these is bad on its own. The risk appears when all five sit with one provider and none of them is written down anywhere else.
Why does AI vendor lock-in matter right now?
AI providers change products far faster than traditional software vendors. In 2026 alone, popular features were renamed, paywalled or retired with a few months' notice. When the provider changes direction, a locked-in business has to follow, whether or not the change suits it.
The clearest recent example is Google replacing Gemini "Gems" (custom assistants) with a new feature called Skills. According to Android Authority's report of Google's timeline, Skills began rolling out in Workspace on 5 October 2026 and in the Gemini app on 13 October. For business and enterprise accounts, Gems can no longer be created, edited or used from 1 March 2027.
The same report notes that Skills do not sync between the Gemini app and Workspace, so a Skill needed in both places must be built twice. Workspace automations with an "Ask a Gem" step will also stop working on that date.
This is not unique to Google. OpenAI has retired older models and custom GPT features with short notice, as we covered in our guides to custom GPT retirement and AI model deprecation. Every provider updates its line-up several times a year.
What are the real costs of being locked in to one AI provider?
The main costs are price exposure, disruption and lost bargaining power. When you cannot leave, you accept whatever price rise, usage cap or feature change arrives. The Zapier survey found 74% of executives said losing their main AI vendor would disrupt daily operations, and only 6% could stop without interruption.
The same survey listed the concerns executives worry about most:
--- 46% cite data migration challenges.
--- 46% cite over-dependence on a single vendor.
--- 44% cite declining service quality.
--- 41% cite sudden price hikes.
--- 37% cite a single point of failure, such as an outage.
For a small business, these translate into very practical pain. A 15-person trading company whose quotation process depends on one AI tool loses half a day of output every time that tool has an outage. A clinic that built its booking replies inside one chatbot must rewrite every answer if the price doubles and it decides to move.
Note that Zapier sells tools that connect different AI models, so it has an interest in the conclusion. The survey method is published (fielded 30 January to 6 February 2026, margin of error about ±4%), and the pattern matches what many owners describe in practice.
Is AI vendor lock-in always a bad thing?
No. Some dependence is the price of getting value quickly. Picking one main AI tool and learning it well is usually smarter for a 10-person company than juggling four. The goal is not zero dependence. The goal is knowing what you depend on and keeping a copy of the parts that matter.
Three common misconceptions are worth clearing up:
Misconception 1: "Monthly billing means no lock-in." A no-contract plan makes the subscription easy to cancel, but your custom assistants, memory and files still live inside it. Short contracts reduce financial lock-in, not data lock-in.
Misconception 2: "Using several AI tools solves it." Spreading work across three tools without a written record of your instructions simply creates three smaller locks. Diversity helps only when the core knowledge sits in a place you control.
Misconception 3: "Lock-in is an IT problem." It is a business continuity problem. The person who should care most is the owner who would lose sales if the tool stopped working tomorrow.
How can a small business avoid AI vendor lock-in?
Keep the valuable parts of your AI setup in plain, portable formats you own, and test a backup tool before you need it. Five habits cover most of the risk for an SME, and none of them requires an IT team or extra software spending.
1. Write your instructions down outside the tool.
Keep every custom assistant's instructions, tone rules and FAQ answers in a shared Word or Google document. If the tool disappears, you paste the text into the next one. This single habit turns weeks of rebuilding into an afternoon.
2. Store your knowledge files in your own drive.
Upload copies of your price list, policies and product sheets to the AI tool, but keep the master version in your company drive. Never let the only copy live inside a chatbot.
3. Export your history every quarter.
Most major assistants let you download your chat history and data from their settings page. Some now go further: Anthropic's Claude added a memory import tool in 2026 that lets users bring stored context across from ChatGPT or Gemini by copying and pasting it. Know where the export button is before you need it.
4. Test a second tool twice a year.
Take your three most important tasks, such as a quotation, a customer reply and a weekly summary, and run them through a second AI tool using your written instructions. If the results are close, you have a working backup. If not, you know what to fix.
5. Ask four questions before you commit.
--- Can I export my data, prompts and history in a standard format?
--- How much notice do you give before retiring a feature or model?
--- What happens to my files and settings if I cancel?
--- Does the plan renew automatically, and what is the notice period?
What does a lock-in check look like for a Hong Kong SME?
A lock-in check is a 30-minute review of every AI tool your business relies on, recording what it does, what you have built inside it and how you would move it. Most owners finish it with a short list of two or three items to copy out and one backup tool to test.
Imagine a 12-person interior design studio in Kwun Tong. It uses one AI assistant to draft client proposals, a second to write Instagram captions and a chatbot to answer website enquiries. Its check might look like this:
--- Proposal assistant: 40 saved instructions inside the app, no copy elsewhere. Action: export to a shared document this week.
--- Caption tool: nothing custom built. Low risk. No action.
--- Website chatbot: trained on 60 FAQ answers and a price sheet. Action: keep the master FAQ in the company drive and confirm the vendor's export option.
The whole review takes less time than a staff meeting. The payoff arrives the day a provider announces a price change or retires a feature, because the studio can compare options calmly instead of scrambling.
Frequently asked questions about AI vendor lock-in
Can I move my ChatGPT history to another AI tool?
Partly. ChatGPT lets you export your data as a downloadable file, and some rivals can import stored memories by copy and paste. Full chat history usually cannot be merged into another assistant's memory, but it can be uploaded as a reference document.
Will my custom assistants keep working if the provider changes its product?
Not always. Google's Gems-to-Skills change shows that custom assistants can be migrated, renamed or switched off on the provider's timetable. Keeping the instructions in your own document is the only way to guarantee you can rebuild them.
Are open-source AI models a way to avoid lock-in?
They reduce dependence on a single company because the model itself can be hosted by different providers. However, they still need setup and someone to maintain them, so most small businesses treat them as an option to keep in mind rather than a starting point.
How often should a small business review its AI tools?
Twice a year is enough for most SMEs, plus an extra check whenever a provider announces a price change, a plan change or a feature retirement.
What is the key takeaway on AI vendor lock-in?
AI vendor lock-in is not caused by bad contracts. It comes from building valuable things inside one tool without keeping a copy. Writing instructions down, keeping files in your own drive and testing a backup twice a year lets a small business enjoy AI's speed without losing the freedom to switch.
The providers will keep changing their products. Gems will become Skills, models will be retired and prices will move. The businesses that stay calm through those changes are the ones that treated their AI setup as their own property from day one.
Choosing AI should feel like hiring a good partner, not signing up for a one-way door. We understand AI. UD stands with you.
Reviewed by the UD AI team.
Want to know which of your AI tools you depend on most, and how easily you could switch? Take the free AI Ready Check and we will walk you through it step by step, from mapping your current tools to setting up a backup plan that keeps your options open.
