There is a line in most interoperability budgets that nobody remembers approving. It has a sober name, something like “mappings and transformations”, and it grows a little every year. When I ask where it comes from, the answer is usually a list of small projects, each perfectly reasonable on its own: two systems that needed to talk, a partner that joined late, a model that was updated on one side and not the other.
That line is the visible part of something much larger, and it has a cause that sits years upstream, in decisions that looked free when they were taken.
Every modelling decision has a price
A modeller facing a new concept has two broad options. One is to look at what the peers already use and build on it, even if it fits a little less neatly or takes longer to understand. The other is to model the concept in the way that suits the project in front of them. The second option is almost always faster, and it often produces a cleaner local model. On the day of the decision, it looks like the better choice.
The cost appears later, and somewhere else. The day that system has to exchange data with a peer that chose differently, someone has to build a bridge between the two models, test it and keep it working. The modeller who made the original choice has usually moved on by then. The cost lands on another team, another budget, sometimes another organisation.
Economists have a name for a cost that one party creates and another pays: an externality. Divergence in a shared ecosystem behaves exactly like one, which is why it is so persistent. The person deciding never sees the bill.
A debt that compounds
Ward Cunningham coined the term technical debt in 1992 to describe how shortcuts taken today become interest paid tomorrow. Semantic divergence creates a close relative, which I find useful to call alignment debt. Each diverging choice is a loan taken out on behalf of the ecosystem, and the ecosystem pays the interest every time two parties need to understand each other.
Unlike most loans, this one compounds with the size of the ecosystem. Picture five companies in a medicines supply chain, each recording storage conditions in its own way. Every pair that exchanges data needs a mapping, so one diverging concept can require up to ten. Assume, for illustration, two days of work to analyse, build and test each mapping, half a day a year to keep it working as either side changes, and a blended rate of €85 an hour. One diverging concept then costs about €13,600 to bridge and €3,400 a year to maintain, roughly €30,600 over five years. If they also diverge on the excursion, the batch status, the recall reason and the remaining shelf life, the bill grows in step:
| Diverging concepts | To build | Each year after | Over five years |
|---|---|---|---|
| 1 | €13,600 | €3,400 | €30,600 |
| 2 | €27,200 | €6,800 | €61,200 |
| 3 | €40,800 | €10,200 | €91,800 |
| 4 | €54,400 | €13,600 | €122,400 |
| 5 | €68,000 | €17,000 | €153,000 |
Agreeing on one shared model instead, at three days of work per company, would cost about €10,200 once. At ten companies, each diverging concept needs up to forty-five mappings. The budget line in my opening is the interest on that debt.
Why the debt stays invisible
The cost is spread thinly across many budgets and many years, so no single owner sees the total, and programme reports count standards adopted and vocabularies published, which records the assets and leaves the liabilities off the page.
The deeper reason is timing. The only moment when divergence is cheap to avoid is the moment of decision, and that is precisely when the modeller has the least information. Alone with a deadline and an empty model, they cannot see what their peers have already chosen, so they cannot see the price of choosing differently. Nobody would sign a loan with the amount left blank, yet that is what an ecosystem asks of its modellers every day.
Paying it down, one decision at a time
The encouraging part is that the same mechanism works in reverse. Each modelling decision that follows what the peers have converged on avoids a future mapping. It also leaves one more piece of evidence behind, so the next modeller facing the same choice sees a clearer picture and is more likely to follow. Convergence, once it starts, makes itself easier.
Paying down alignment debt can therefore start small. The price has to be visible at the moment of decision, and the governance culture has to be one that treats a knowing divergence as a decision to be justified, with its cost written down, the way a responsible organisation treats any other debt.
Questions to put to your own ecosystem
- What do we spend each year on mappings and transformations between parties? Add it up across budgets; the total is usually a surprise.
- Which of those mappings exist because two parties modelled the same thing differently? Separate them from mappings that bridge genuinely different needs.
- Who owns each mapping, and who updates it when either side changes?
- Which divergences were chosen knowingly, and where is the reason recorded?
- What can a modeller see about their peers’ choices before deciding? That is where the next loan will be signed.
Making the cost visible
Divergence will never disappear, and some of it is healthy: real differences in need deserve different models. The aim is to make it a choice with a known price, taken by people who can see what their peers have already built. Making that price visible, at the moment decisions are taken, is the problem we built EVIDENT around at METALEVELS.
If you would like to put a number on your own ecosystem’s alignment debt, start a conversation with us.