One AI supplier means one set of prices, terms and roadmaps. The Portability Ladder tests four things: method, data, workflow and account control. Diversify by being able to move, not by buying twice.
Ask a small organisation which AI tool it depends on and the answer is usually one name. That is understandable. One tool means one login, one training session, one invoice. But it also means that pricing, terms, features and availability are decided by someone else, and the organisation's working methods have quietly become a function of that supplier's roadmap.
Boards are used to asking about concentration risk in funding and in banking. Very few ask it about AI. We think they should, and we think the answer is neither to avoid the dominant supplier nor to run four in parallel. It is to make dependence a conscious, priced and reversible decision.
Lock-in is rarely a contract clause. It is the accumulated habit, data and workflow that make leaving more expensive than staying.
Concentration is easier to manage once it is unbundled. An organisation that "uses one AI vendor" is usually dependent on it in four different ways, and each has a different exit cost.
We use a four-rung ladder to assess how easily an organisation could move. Each rung is a question that can be answered honestly in an afternoon.
Rung one: could we restate the method? Is every recurring AI task written down as a method a colleague could follow with any comparable tool? If not, the method lives in one person's head and one tool's memory.
Rung two: could we take the data? Have we tested an export, not merely confirmed that an export button exists? A test is the only evidence.
Rung three: could we rebuild the workflow? How many routine processes would break if the tool were unavailable for a week, and is there a manual fallback?
Rung four: do we hold the keys? Are accounts, billing and administrator rights held by the organisation rather than by an individual employee or a personal card?
An organisation that clears all four rungs has an exit plan, even if it never uses it. One that clears none is not necessarily making a mistake, but it is making a bet it may not know about.
Diversification does not mean using more tools. It means being able to use a different one within a week.
The instinct is to buy a second licence for everyone. That is the most expensive way to reduce risk, and it rarely works, because an unused second tool decays quickly.
A cheaper design is asymmetric. Keep one primary tool for daily work. Nominate one alternate and give a small group, perhaps the champions in each department, a low-cost account. Twice a year, ask that group to run a defined set of representative tasks on the alternate and compare results with the primary against the same quality criteria. The cost is a few hours and a handful of licences. The benefit is an evidence-based answer to a question that otherwise gets asked only during a crisis.
Two further habits keep the cost down. First, keep your test tasks and quality criteria in a shared document that is tool-neutral. Second, avoid building deep integrations for tools you have not yet confirmed you will keep. Shallow, replaceable connections are worth more than clever ones.
Before adopting or renewing any AI tool, we suggest four questions. What happens to our data if we leave, and in what format can we take it? How much notice will we receive of price or term changes? Where is data processed and who else can access it? And what would it take to move our documented method to a comparable product? A supplier who answers plainly is showing you something about the relationship. One who cannot is showing you something else.
Not every organisation should diversify. For a very small team, the overhead of a second tool may exceed the risk it removes. Concentration is a reasonable choice when three conditions hold: the method is documented, the data has been exported at least once, and the organisation holds the account. Those three conditions make a single supplier a preference rather than a dependency.
The ladder is practical because it can be run quickly. Bring together the person who administers your accounts, one champion from a busy department and someone from finance. Spend an hour listing every AI tool in use, including those individuals adopted on their own. For each, walk up the ladder and mark where the answer is yes, no or unknown.
The unknowns are the most valuable output. They are the places where dependence exists and nobody has looked at it. Typical examples include personal accounts holding work documents, workflows that rely on a single person's saved settings, and terms accepted years ago that nobody has reread. Assign each unknown to a named owner with a date, and revisit at the next quarterly review.
Portability is not free. Writing methods down takes time, test exports take time, and running a semi-annual comparison takes time. We would estimate the recurring effort as measured in hours per quarter for a small organisation, not days, but each organisation should measure its own.
Set that cost against the alternative. A forced move under pressure, because of a price change, a policy change or an outage, is usually far more expensive than the same move planned. The comparison is not between diversifying and doing nothing. It is between paying a small amount regularly and paying a large amount once, at a moment not of your choosing.
The goal is not to avoid dependence but to keep it optional: a documented method, a tested export and an account you control turn a single supplier into a choice rather than a trap. You can build that for the price of an afternoon and a few licences.
If this is the question on your desk, a thirty-minute call tells you whether the service fits, or that you do not need us yet.