Start typing to search this publication.
techcirkle
Open menu

Subscribe to techcirkle

Get new posts delivered straight to your inbox.

Why Insurance Integration Is Half Your Project

Insurance platforms connect to rating providers, payments, reinsurance, brokers and regulators. Each brings its own protocol, contract and failure mode.

techcirkle
5 min read
Cover image for Why Insurance Integration Is Half Your Project
Insurance technology and data integration concept
Insurance technology and data integration concept

Every insurance software plan I have reviewed underestimates integration. Not occasionally — consistently, and in the same direction, by roughly the same margin.

The plan allocates perhaps twenty per cent of effort to "integrations", listed as a single line. The actual figure lands closer to half. The gap is where the year of slippage comes from.

It is worth enumerating the surface, because the estimate is only wrong because the list is never written down completely.

The actual surface

Rating and enrichment providers. Hazard data, credit references, vehicle and property databases, sanctions and PEP screening. Each is a separate commercial contract, a separate SLA, a separate latency profile, and a separate failure behaviour. Some return in 80ms, some in eight seconds, and your quote journey has to accommodate both without appearing broken.

The subtle cost here is not the integration itself but the degradation design: what does your quote do when the hazard provider is down? Decline? Quote with a loading? Queue for later? That is a commercial decision requiring an underwriter, and it needs making for every provider.

Payments and premium finance. Straightforward until you hit reconciliation. A payment that partially covers an instalment, a direct debit that fails after the policy has incepted, a premium finance agreement that pays the carrier in full while the customer defaults to the lender. Each has a defined correct behaviour and none of them are in the happy path.

Reinsurance and bordereaux. Reports produced in formats defined by the reinsurer rather than by you, usually per treaty, frequently as spreadsheets with layouts that vary between counterparties and change without much notice. This is the workstream most often discovered late, because nobody in the technology conversation owns the reinsurance relationship.

Broker and comparison platform connections. Often over legacy protocols, with idiosyncratic dialects per counterparty. Two brokers implementing the same standard will differ in ways only discovered in testing, and sometimes only in production.

Regulatory reporting. Fixed schemas, immovable deadlines, and no tolerance for approximation. The schema is published, which makes it look easy; the difficulty is that it requires data your operational systems were not designed to retain.

Document generation and archival. Sounds trivial. Is not, once you account for versioned policy wordings, jurisdiction-specific requirements, and the obligation to reproduce exactly the document a customer received on a past date — which means storing the rendered artefact, not the template and the data.

Why the estimate is always wrong

Insurance agent working through documentation with a client
Insurance agent working through documentation with a client

Three structural reasons, all of which are avoidable if named early.

Each integration is estimated as an API integration. It is not. It is a commercial relationship with a technical component. There is contract negotiation, a security review from the counterparty, test credentials that take three weeks to arrive, a certification process, and a support relationship. The code is often the smallest part.

The failure paths outnumber the happy paths. For every provider you need behaviour for: timeout, malformed response, authentication failure, rate limit, planned maintenance, and — the awkward one — plausible but wrong data. That last case requires business rules, not error handling.

Counterparty timelines are not yours. You can staff your side. You cannot make a reinsurer's IT team prioritise your certification testing, and that dependency does not compress with more engineers. This is the single most common cause of a slipped go-live date.

What to do about it

Enumerate before estimating. Write down every external party with a data relationship, including the ones nobody thinks of — the ad-hoc report a regulator requested, the spreadsheet a broker emails monthly, the file a loss adjuster uploads. The complete list is always longer than the plan's list.

Estimate per counterparty, not per protocol. Two brokers on the same standard are two integrations, because their interpretations differ and their test cycles are independent.

Start counterparty-dependent work first, even when it is not on the critical path internally. Credentials, security reviews and certification slots have lead times you do not control, and starting them in month one costs nothing.

Design degradation explicitly, with the business. For every external dependency, decide with an underwriter what happens when it is unavailable. Engineers making that call alone will choose technically defensible behaviour that is commercially wrong.

Build one integration end to end early — including its failure paths, monitoring and reconciliation — before estimating the rest. The first one calibrates every subsequent estimate, and the calibration is usually upward.

The one that catches everyone

Reinsurance bordereaux, consistently.

It arrives late because it is not customer-facing, no one demonstrates it, and the person who understands the treaty structure is in finance rather than technology. Then someone realises the reports are contractually required, the format is defined by a counterparty, and the data model does not retain what the treaty needs apportioned.

Add it to the plan on day one, and get the treaty documents in front of whoever is designing the data model. It is a fifteen-minute conversation that routinely saves a month.

Full guide — the six functional areas, build versus buy, bitemporal data modelling and core replacement mechanics: Insurance Software Solutions. We also scope insurance platform programmes.

Frequently Asked Questions

How much of an insurance software project is integration?

Roughly half, against plans that typically allocate a fifth. The gap accounts for most of the slippage in these programmes, and it exists mainly because the list of counterparties is never enumerated completely before estimating.

Why are integration estimates wrong so consistently?

Because each one is priced as an API integration when it is a commercial relationship with a technical component — contract terms, counterparty security review, test credentials, certification. The code is frequently the smallest part.

What is the most commonly missed workstream?

Reinsurance bordereaux. It is not customer-facing, nobody demos it, and the treaty knowledge sits in finance rather than technology — so it surfaces late, when the data model already fails to retain what the treaty needs apportioned.

How should we estimate integrations?

Per counterparty rather than per protocol, since two parties implementing the same standard differ in interpretation and have independent test cycles. Build one end to end first, including failure paths and reconciliation, then use it to calibrate the rest.

What about failure handling?

Design it with the business, not alone. For each external dependency decide with an underwriter what happens on timeout, malformed response or plausible-but-wrong data. Engineers choosing independently will pick technically sound behaviour that is commercially wrong.

How do we avoid counterparty-driven delays?

Start the work you do not control in month one — credentials, security reviews, certification slots. Those lead times do not compress with more engineers, and they are the most common cause of a missed go-live date.

Subscribe to techcirkle