The Data Problem Nobody Has Fully Solved: Inside Delegated Authority with Paul Templar

Half a million bordereaux processed every year. Roughly half of Lloyd’s Managing Agents as customers. Eight or nine countries of operation. And a founding story that started with two brothers building the UK’s first price comparison website.

Paul Templar didn’t set out to run an insurance data company. He joined VIPR in 2009 as its third employee, working under co-founders Rick and Bob Brown — the same team he’d worked with at Moneynet, the UK’s original price comparison site. Bob Brown had a Lloyd’s broking background — Marsh, North American division, his own brokerage he later sold — and market contacts kept telling him the same thing: you guys understand technology, we have technology problems, come help us.

That conversation became VIPR. Seventeen years later, Paul is CEO of a company that has become close to synonymous with Delegated Authority data management in the London market — and is now working out whether the US market, which everyone assumed would be more sophisticated, is actually years behind.

In Episode 166 of InsurTechTalk, Paul walked me through what Delegated Authority actually is, why bordereaux remain one of insurance’s most stubborn unsolved data problems, and what changed the day he went from CTO to CEO with, in his words, no manual for how to do the job.

About Paul Templar

Paul Templar is the Co-Founder and CEO of VIPR Solutions, a data management platform serving the Delegated Authority market. He joined VIPR in 2009 as the company’s third employee, having previously headed technology at Moneynet, the UK’s first price comparison website, working alongside VIPR co-founders Rick and Bob Brown. Paul served as VIPR’s CTO for 11 years before being promoted to CEO in 2019, when the Brown brothers stepped back toward retirement. VIPR closed its first private equity round with Tenzing in 2020 — during COVID — and has since expanded from the UK into eight or nine countries, including the US, Canada, Bermuda, Switzerland, Spain, Benelux, the Nordics, and Malaysia.

What Is Delegated Authority? A Plain-English Explanation

This is worth walking through carefully, because it is foundational to a large share of specialty and MGA business globally, and most of InsurTechTalk’s US-centric audience has never had it explained cleanly.

The Three Parties in the Lloyd’s Market

• Managing agents — the roughly 56 brands that trade within the Lloyd’s marketplace (Axis, Everest, Canopius, Beazley, Brit, and others). These manage the underwriting syndicates and provide capacity. Paul’s analogy: think of a managing agent as roughly equivalent to a US carrier.
• Lloyd’s brokers — the Gallaghers, Aons, and Beazleys of the world, sitting in the middle as matchmakers. They go to market, find distribution, and place that distribution with the capacity providers.
• Lloyd’s coverholders — essentially MGAs. These are the parties actually retailing or wholesaling the business — direct to consumer or through other channels.

How Binding Authorities Work

A binding authority is a contract that gives a coverholder a defined set of underwriting rules they can operate within — Paul’s example: property business in Florida, no closer than a mile to water, condos no more than three stories, timber frame construction or otherwise, and so on. Within those rules, the coverholder can write business freely. Generally the broker acts as the conduit, and multiple managing agents will each take a slice of the coverage — a lead insurer taking the largest share, followed by a follow market of smaller shares making up the remaining percentage.

What a Bordereau Actually Is

Coverholders send data back into the market — usually monthly — listing claims made, policies written, and premiums collected. That data is the bordereau. The problem: there is no fixed standard for how it arrives. Think of it, in Paul’s words, as spreadsheets — but spreadsheets of wildly varying format, structure, and quality, arriving from potentially hundreds of coverholders with wildly different levels of technical sophistication.

Why Bordereaux Remain Such a Persistent Problem

This is the structural insight worth sitting with: the problem is not really technical. It is about incentive and market structure.

The Core Challenge

• The market has been reluctant to impose hard standards on the third parties supplying data, because those parties range enormously in sophistication
• At one end: fully tech-enabled, API-driven parties who can send system-to-system data effortlessly
• At the other end: coverholders manually handwriting forms and re-keying them into spreadsheets
• Everything in between exists simultaneously, and a managing agent or insurer receiving this data has to handle all of it
• The result is not just variety in format but massive variance in data quality and reliability

What VIPR Actually Does

At its core, VIPR is a data business. The goal for every customer: arrive at a single, high-quality, “golden source of truth” data set that can feed downstream systems, drive underwriting decisions, inform capital deployment, and support claims management.

The core workflow:

• Data collection — helping customers gather bordereaux data as efficiently as possible from third parties, regardless of their format or sophistication level
• Standardization — transforming inconsistent, non-standard data into a structured, usable format
• Contract compliance checking — verifying that coverholders are adhering to the terms of their binding authority: writing the right classes of business, not exceeding limits, taking the correct commission rates
• Issue resolution and storage — once discrepancies are identified and corrected, clean data flows into the golden source of truth that powers the rest of the business

The Product Suite

• Intrali — the flagship, core product: standardizes data, checks it against contract terms, flags and helps resolve issues, and feeds the clean data downstream
• Portal — a web portal that lets coverholders submit data directly, which flows seamlessly into Intrali via API — enabling straight-through processing
• Connect — an integration layer that connects to virtually any data source: monitoring an FTP site, intercepting email attachments, or providing a direct API for more sophisticated MGAs
• Data Cloud — an industry-grade cloud data warehouse built on Snowflake, with a purpose-built data model for reporting and feeding downstream systems
• Insights — the analytics and dashboarding layer sitting over Data Cloud, increasingly incorporating AI to let users chat with their data and build dashboards on the fly
• Onboarding — a due diligence and know-your-customer workflow for vetting new coverholders before they start supplying data

Who VIPR Serves

The common thread across VIPR’s customer base: anyone receiving delegated data from a third party. That includes insurers, reinsurers, fronting carriers, captives, and — increasingly — larger MGAs that are themselves sub-delegating authority and receiving bordereaux data from coverholders beneath them in the chain.

The CTO-to-CEO Transition: No Manual Included

Paul’s path to the top job is unusual on this show — most founders arrive as CEO from day one. Paul spent 11 years building the product before taking the business.

What Actually Changed

• In a small business, Paul was already wearing multiple hats — building software, but also involved in sales, training, and client presentations. His background before VIPR was actually rooted in financial services sales, which gave him more grounding for the transition than a purely technical background would have
• Rick and Bob Brown served as mentors throughout, deliberately helping Paul develop beyond the technical role
• When the Brown brothers decided it was time to step toward retirement and explore private equity partnership to fuel VIPR’s next growth phase, Paul put himself forward for CEO — a transition that happened in 2019
• VIPR closed its first private equity round with Tenzing in 2020, in the middle of COVID — an unusual environment for closing any transaction, let alone a company’s first institutional capital raise
• Paul’s own description of the moment: he was “left with the keys to the kingdom” and “couldn’t find the manual” for how to actually be a CEO

What COVID Meant for VIPR’s Growth

Insurance has traditionally been a relationship-driven, face-to-face industry. COVID stripped that away overnight, forcing the market toward remote work and digital processes it had previously resisted. That shift directly benefited the categories of technology VIPR was building — accelerating adoption of tools that made sense in a world where in-person relationship management was no longer possible.

The US Market: Not What VIPR Expected

This was one of the most counterintuitive parts of the conversation, and worth drawing out clearly for a US audience.

Structurally Similar, More Complex in Practice

• The three-party structure maps reasonably well: a US carrier plays the role of the managing agent, a program administrator sits in the broker’s position, and agents on the ground correspond to coverholders
• The US market is structurally simpler in some ways — typically a single carrier insures a given risk, rather than Lloyd’s multi-carrier lead-and-follow structure, and there’s no multi-currency complexity
• But the range and complexity of data formats VIPR encounters in the US is actually greater than in the UK — where UK submissions are mostly Excel and CSV, US submissions include PDFs, XML, fixed-width text files, and handwritten documentation, spanning a much wider range

The Surprising Discovery

VIPR expected Silicon Valley sophistication to have already solved this problem in the US. Instead, they found many US carriers either building their own bespoke ETL pipelines and data infrastructure in-house, or outsourcing the problem entirely — often without awareness that purpose-built vendor solutions already existed. Paul’s example: an already-stretched IT department building a new data pipeline from scratch every time a new contract comes on board, taking three months each time, with no compounding innovation benefit, versus a vendor solution that continuously improves and is maintained by someone else entirely.

Paul’s colleague Tony’s framing, which Paul cited approvingly: organizations should buy world-class solutions where they already exist, and only build custom technology where nothing suitable is available.

VIPR’s AI Philosophy: Deterministic Outputs, Human in the Loop

This was the most substantively important part of the conversation for anyone evaluating AI vendors in a regulated environment.

The Core Principle

AI is an enabler — not a replacement for deterministic, repeatable, auditable process. In a regulated environment, hallucination and inconsistency are not acceptable failure modes. Paul’s approach: use AI to help build a deterministic process, verify the output with a human in the loop, and then treat that verified output as a repeatable unit going forward.

Concrete example — data mapping: AI assists in mapping a new, unfamiliar data format the first time it’s encountered. A human verifies the mapping is accurate. That verified mapping then becomes the fixed, repeatable rule applied to every subsequent submission from that source — removing the risk of the AI reinterpreting the same data differently each time.

The Ambiguity Problem

Paul’s sharpest illustration: a column in a bordereau might simply be labeled “premium.” A human processor who has worked that specific data feed for three years knows implicitly that this particular column means premium including taxes. An AI model, absent that institutional context, will take the label at face value and process it incorrectly — producing a different, wrong answer with full apparent confidence.

Why VIPR Uses Open-Weight Models, Not Just Frontier APIs

Two deliberate reasons: controlling token cost, and keeping customer data within VIPR’s own boundary rather than sending it to third-party model providers. Paul flagged that many prospects haven’t considered the token economics of AI-enabled solutions at all — assuming AI capability is effectively free, when in reality every frontier-model query carries a real, compounding cost on top of any software license.

Ontology as a Guardrail

VIPR has built a structured ontology around its core data sets — explicit guidance on how data is structured, how elements relate to each other, and what interactions are valid. The result: even broad, loosely worded questions to the system stay within defined guardrails, because the model’s reasoning path is constrained by the ontology rather than left to freely associate. Paul described being able to watch the chain of thought and the underlying queries the system runs to arrive at an answer — a level of auditability that matters enormously in a regulated context.

The Closing Answer: Simplify, Don’t Add More Standards

Paul’s answer to what the industry should be talking about more was direct: how do vendors, insurers, and brokers work together to make the data submission process simpler — not add more competing standards, more egos, more misaligned incentives, but genuinely streamline what has become an unnecessarily complex system.

Key Takeaways

• Delegated Authority is a three-party structure — managing agents (capacity), brokers (matchmakers), and coverholders (MGAs actually writing business) — and bordereaux are the periodic data reports that flow back up that chain
• Bordereaux data quality remains poor not primarily because of technology limitations, but because the market has been reluctant to impose hard standards across wildly varying levels of coverholder sophistication
• The US market’s data format complexity actually exceeds the UK’s, despite assumptions that US insurance technology would be more advanced
• Many US carriers are still building bespoke, non-compounding data pipelines in-house or outsourcing entirely, unaware that purpose-built vendor platforms already solve the problem
• AI’s proper role in a regulated data environment is enabling deterministic, auditable, repeatable processes with a human in the loop — not autonomous judgment calls on ambiguous data
• Token cost and data boundary considerations are underappreciated factors that most prospects haven’t priced into their AI vendor evaluations