The Reconciliation Gap Holding Insurance Back

An interview with Tony Shek, by Sean Murphy

An insurance carrier receives a single payment from a broker, half a million pounds, and then has to work out which of five thousand individual policies it actually covers. The sum almost never matches cleanly. There is a balance left over, but nobody can say with certainty which policy has been paid and which has not. Multiply that uncertainty across a year, and the unmatched amounts accumulate quietly in the background until, at one large carrier Tony Shek has worked with, two and a half million pounds simply had to be written off. The money was real. The problem was that nobody could trace it.

Shek leads the insurance vertical at AutoRek, a firm that builds data management and reconciliation software for financial services. He is careful to point out that the write-off he describes sat inside a business with billions in revenue, so as a proportion it was small. As an absolute figure it is the kind of number that makes a finance director sit up.

And it captures, more vividly than any survey, the problem he spends his days trying to solve. Insurers are sitting on payments they cannot match to policies, and the cost of that blind spot is both a cash flow drag and a credit risk.

Insurance has long been seen as a step behind the rest of financial services when it comes to technology, and Shek offers two explanations. The first is cultural. The whole business is built on caution, and the people inside it know it.

Ask any insurance professional and they will tell you, cheerfully enough, that they are a conservative bunch. Risk is the product, so prudence runs through everything. The second reason is more practical, and in Shek's view it is the one that really matters. A single insurer will typically be running between six and ten separate systems at once, the residue of years of migrations and acquisitions that were never fully integrated because integration was expensive and there was always a more urgent priority. The result is a landscape of disconnected data that resists standardisation, and that disconnection is what holds modernisation back.

This is the gap that AutoRek's 2026 insurance report puts numbers to. It found that 82 percent of insurers believe AI will define the industry's near future, while only 14 percent have actually embedded it into their financial operations. Shek is quick to say the appetite is genuine and the use case is obvious to the people living with the pain. What gets in the way is the foundation. Legacy systems are part of it, but so is the simpler matter of how data is held. Underwriting, pricing and claims teams are often still working in spreadsheets, in formats that do not talk to one another. He gives a small example that illustrates a large problem. Buy one policy through a managing general agent and another through a broker, both with the same carrier, and your name might be recorded slightly differently each time, with a different address attached. To the system you are now two people. Scale that across millions of records and the difficulty of building anything intelligent on top becomes clear.

The delays this causes are not trivial. The same report found settlement cycles stretching past sixty days for nearly half of insurers, and Shek traces those waits back to the same root. When a carrier cannot quickly match what it has received against what it is owed, payment slows, and the checking, verification and approval that has to happen manually in the meantime eats time the business does not have. Most firms cope with spreadsheets and the mapping functions inside Excel, which is workable but slow, and which leaves exactly the kind of gap that lets a missing payment go unnoticed for a year.

On where insurers should begin, Shek is less of a salesman than might be expected. He does not lead with his own product. The first thing a firm needs to do, he argues, is understand its own constraints, because some of the legacy systems involved are mainframes that have been running for decades and have no capability to connect to anything modern. Only once a company knows what it is dealing with can it work out how much automation is realistically achievable without ripping out the core. Sometimes the answer is end to end automation. Often it is not, and a more honest target might be seventy percent, which still saves considerable time and improves the integrity of the data dramatically. The point is to start from where you are rather than from where a vendor would like you to be.

AutoRek's own answer to the efficiency question is ARIA, the AI layer it added to its platform in 2025, which Shek says has been received well. Where the underlying software matches and maps on a rule based footing, ARIA sits on top of it as something closer to an in-app expert, and Shek describes three personas a user can call on.

The software does not replace judgement but it removes the drudgery around it.

The difference ARIA makes is scale and intelligence, a reasoning layer rather than something matching on a standard reference, applied across millions of transactions at a time.

AutoRek now serves many Tier 1 and Tier 2 insurers across roughly fourteen different use cases, with its strongest presence in the UK, Europe and the US and a longer reach into South Africa, Australia and Asia. The breadth is the point Shek wants to leave, that reconciliation is not one problem but many, threaded through claims, commission payments and regulatory reporting alike. The two and a half million pounds written off at the start was never really about one broken payment. It was about a business that could not see itself clearly, and the work Shek describes is, in the end, about giving insurers that visibility back, one matched policy at a time.