A platform team spends about 11% of its engineering capacity staying portable - an internal interface over the managed queue, nightly exports of vendor-held data, and an annual paper exit. Finance asks what that buys. What has the team actually purchased, and when is paying for it irrational?

What is being tested

Whether you can price optionality instead of asserting it. "Avoid lock-in" is a preference. "We pay about 2.2 engineer-years a year for an option worth X" is a position finance can accept or reject, and senior engineers are expected to produce the second sentence.

What the premium buys

Two concrete things, and they are worth different amounts.

Negotiating position. A vendor's pricing power is bounded by what it would cost you to leave within your notice period, typically 30 to 90 days. A capability you could replace in three weeks gets market pricing at renewal. One with a nine-month exit gets whatever the vendor decides, because the alternative cannot be funded on notice.

A shorter exit if you are forced out. Acquisition, a licence change, a regional withdrawal or a pricing model change are not hypothetical in a three-year horizon. The nightly exports remove the data half of the exit, which is usually the half that cannot be compressed by hiring.

What it pays

11% of a 20-engineer team is about 2.2 engineer-years a year. At a loaded rate near $200000 that is roughly $440000 of capacity. The abstraction also costs capability: an internal interface over a vendor converges on the lowest common denominator, so you pay for features you have forbidden yourself to use, and every vendor upgrade becomes a change in two places.

Now compare. If the contract is $600000 a year and the hedge holds a 30% renewal uplift, it returns about $180000 against $440000 spent. The hedge loses money. At a $4M contract the same 30% is $1.2M and the hedge is obviously worth it. The ratio that decides it is annual contract value against the engineering cost of portability, not any principle about lock-in.

The decision rule

Buy portability where the capability is commoditised and the contract is large: object storage, SQL, queues, telemetry pipelines with OpenTelemetry already in them. Do not buy it where you chose the vendor for the non-portable part, because the abstraction deletes the reason you bought it.

Where you decline the hedge, write the exit cost in the decision record in engineer-months and revisit it annually, since it grows on its own. One cost has recently fallen: since 2024 both of the largest providers waive data-transfer-out charges for customers migrating off, so the transfer line that used to feature in these arguments is largely gone. Engineering time and the dual-run overlap are what exits cost now.

When this is the wrong answer

For a four-engineer team whose constraint is shipping at all, 11% of capacity on reversibility is capacity the product cannot spare, and the correct answer is a documented, accepted lock-in. The hedge is also wrong where migration is implausible for non-technical reasons: a regulated workload certified against one provider's controls will not move because an abstraction exists.