Skip to main content

9 min readPublished November 8, 2026Talavishiraaj Groups
Technology & Engineering

Interoperability is usually scoped as a standards problem and delivered as a semantics problem. Two systems can both speak FHIR and still disagree about what a patient episode is, when it started, and which identifier is the real one. The interface is days of work. Reconciling the meaning is the project.

In short
  1. 01A shared standard means systems can exchange messages. It does not mean they agree what the messages mean.
  2. 02Identity reconciliation is the most underestimated part of almost every healthcare integration.
  3. 03Decide the authoritative source per field before building, not during testing.
  4. 04Workflow matters more than throughput: data arriving outside the moment of care is not used.
  5. 05Governance is not paperwork here. Who may see what, and who answers when the system is wrong, has to be settled before go live.

What is healthcare interoperability?

The ability of clinical and administrative systems to exchange information and use it. Both halves are load bearing, and projects consistently scope the first and discover the second.

Exchange is largely solved. Standards exist, most modern systems speak them, and moving a message from one place to another is well understood engineering.

Use is where the work is. A receiving system has to interpret what arrived in the same way the sending system meant it, and that agreement does not come with the standard.

If there are standards, why is this still hard?

Because a standard governs structure, not meaning. Two systems can both emit a conformant message and still disagree about when an episode of care began, whether a cancelled appointment counts, or which of three identifiers refers to the same person.

Those disagreements are invisible in testing with clean sample data and appear immediately with real records, which is why integration projects so often look finished and then stall at the point of trusting the output.

The implication for scoping is uncomfortable but useful: the interface work is predictable and the semantic work is not, so a plan that budgets mostly for the interface has budgeted for the easy half.

What is the most underestimated part?

Identity. Deciding that this record and that record describe the same person, when the sources disagree on spelling, date of birth, address history and which number is authoritative.

Every organisation has an answer to this that works well enough for its own purposes and breaks on contact with another organisation’s answer. Reconciling them is slow, needs clinical input rather than only technical input, and cannot be deferred, because everything downstream depends on it being right.

A project that has not named who owns identity reconciliation has not started.

What has to be settled before writing an interface?

Which system is authoritative for each field, field by field rather than system by system. It is normal for one system to be definitive about demographics and another about clinical events, and the integration has to encode that rather than assume a winner.

What happens when they conflict. Silent precedence rules are the most common source of quiet, long lived data problems, because nothing visibly fails.

What the data is for. An integration built to populate a report has different accuracy and latency requirements from one informing a decision at the bedside, and conflating them produces a system that is over engineered for one use and unsafe for the other.

How does workflow decide whether it works?

More than throughput does. Information that arrives correctly but outside the moment someone is making a decision is not used, and an integration judged on messages per second can be a complete success by its own metrics and change nothing clinically.

The test is where the output lands. If using it means leaving the system someone already has open, adoption will be poor regardless of data quality.

This is also why pilots mislead. A pilot with motivated participants will use a worse integration than a full rollout will tolerate.

How should this be scoped and governed?

In stages that each end with something usable, rather than one programme that delivers at the end. Integration work has a long tail of exceptions, and a plan with no intermediate value has no way to stop sensibly.

With governance settled first: who may see what, under what lawful basis, with what audit trail, and who answers when the system produces something wrong. In this sector that is not administrative overhead, it is a precondition.

And with the running cost stated. Integrations have a permanent relationship with systems that are versioned, renamed and replaced without consulting them, so maintenance is not optional and should appear in the business case rather than arriving in year two as a surprise.

Where this work sits
Technology & Engineering →

We build and integrate the production systems a business runs on: internal platforms, customer facing applications, integrations, and the data layer underneath them.

Weighing a decision like this one?

Describe the situation and we will tell you how we would approach it.