In a market where pricing and players shift every six months, the question is not theoretical. It never arises when choosing, always when you would like to leave.
Definition
Vendor lock-in describes the difficulty of changing supplier once your organisation depends on them.
It does not come from a contract but from accumulation: data stored with them, scenarios built in their interface, habits formed by your team.
Dependence is not a problem in itself. You depend on your bank and your electricity supplier without worrying. It becomes a risk when it is strong, invisible, and concentrated on a fragile player or one whose pricing can change without notice.
The three forms, lightest to heaviest
| Form | Exit cost | Example |
|---|---|---|
| Habits | Low: some training | Changing email provider |
| Construction | Medium: rebuild everything | Your automation scenarios |
| Data | High: sometimes impossible | A client history with no export |
The third line should guide your choices. A tool that does not let you export your data in a readable format holds you, whatever happens. It is the first criterion to check, before features.
Limiting the risk without going without
Check the export before committing. Not the promise of an export, the actual export. Download a test file on day one, not on the day you leave.
Keep your data elsewhere. Your scenarios can live in a platform, your reference data should live somewhere you control.
Document in plain words. An Automation scenario described in a document can be rebuilt elsewhere. A scenario that exists only in an interface is lost with it.
Prefer standards. A tool that speaks API and Webhook is easier to replace than a closed one. On the AI side, MCP plays exactly that role.
Keep a technical way out. A tool like n8n, offering Self-hosting, leaves you an option even if the vendor changes policy.
Beware of dependence built without a decision. Nobody chooses to depend on a tool: you add one scenario, then ten, then forty, and one day discover that leaving would take three weeks of work.
Frequently asked questions
Should closed tools be avoided?
No, they are often the simplest and most polished. You simply need to know what leaving would cost, and accept that cost knowingly rather than discover it.
Is dependence on an AI model strong?
Less than people think for everyday use: a well-built Prompt transfers between models with adjustments. It becomes strong as soon as you do Fine-tuning, which is specific to one model.
Is it better to spread across several providers?
Rarely for a small business: you multiply complexity and subscriptions for a risk that stays theoretical. Better one provider with a verified way out.
How do you assess this risk before choosing?
By asking a simple question: if this tool doubled its prices tomorrow, what would I do? Our n8n course builds that question into tool selection, with the export criteria to check.