68% of P&C Carriers Run Three or More Policy Systems. Almost None Get an Apples-to-Apples View.

Bruce Broussard says nobody chooses insurance — insurance chooses you. His path started with an internship arranged through a neighbor’s father while he was a computer science student at LSU, which turned into 15 years running the IT department of a midsize international insurer, followed by 15 years at IBM, 12 of them running the company’s global insurance data and analytics practice. That work included helping develop the IIA and IW data products and donating them to ACORD to help establish the standards the industry still runs on.

When IBM wasn’t interested in industry-specific software — only enterprise-level platforms — Bruce moved to Insurity, one of the three largest P&C insurance software vendors, and built what he says was the first vendor-driven data solution in the category, at a time when neither Guidewire nor Duck Creek had one. Six and a half years later, he left to start Percipience with AJ Keller, a business partner of over 20 years he met when IBM acquired PwC.

In Episode 122 of InsurTechTalk, Bruce and I covered why so few carriers actually get a unified view of their own data, what documenting a machine learning model for a regulator actually requires, and the real difference in how a top-tier carrier versus a regional one decides to invest in data infrastructure.

About Bruce Broussard

Bruce Broussard is the Managing Director at Percipience, a company he co-founded with AJ Keller to address gaps in insurance data infrastructure. Percipience’s core product, Data Magnifier, is a data platform — not a data provider — designed to help carriers, MGAs, and MGUs integrate, assimilate, and derive value from the many disparate systems and third-party data sources that make up their operations. Bruce spent 15 years running IT for a midsize international insurer, 15 years at IBM leading its global insurance data and analytics practice, and six and a half years at Insurity building one of the first vendor-driven insurance data solutions on the market.

Why “Just Add Data” Isn’t the Hard Part

Bruce’s framing of the industry’s core challenge cuts against the popular narrative that insurance simply needs more data.

• Insurance companies are, in his words, drowning in data — the harder problem is separating what’s actually useful from what isn’t, and building the discipline to derive value from what’s kept
• Insurance has historically applied technology effectively but not innovatively — the industry’s core competency has always been measuring risk, not building software
• Data is genuinely the primary asset of an insurance company in a way it isn’t for a manufacturing business — there’s no unit cost curve to bend, only information and risk management to get right

The tension he sees constantly in conversations with carriers: spend too much time chasing what’s coming next and you never drive value today; spend too much time optimizing today’s process and you fall behind as the landscape shifts. Very few organizations, in his experience, find the right balance.

The Tier-Two Carrier That Rebuilt Its Whole Process Around Data

Bruce gave a detailed, anonymized example of a genuinely transformative shift — an 80-plus-year-old tier-two carrier that recognized a disproportionate share of its losses came from a small fraction of its book.

• The company committed to becoming “data-driven” not as a slogan but as an operational restructuring: managing risk selection at underwriting and claims handling on the back end, informed continuously by data rather than periodically
• The mind shift Bruce considers the real unlock: moving data from something that appears in a report or dashboard after the fact — a tool for management to assess past performance — to something fed directly into the front end, informing underwriters, actuaries, product development, and claims managers as they work
• He hears this ambition articulated constantly at conferences and in sales conversations, but sees genuine operational execution on it far less often — it’s a significant organizational commitment, not a software purchase

What Data Actually Goes Into a Quote

Beyond what a policyholder enters directly — the basics for something like an auto or home policy — carriers layer in substantially more, and the sourcing has gotten more contested over time.

• Traditional inputs like credit scores are increasingly restricted state by state, forcing carriers toward more creative sourcing
• Social media and other public information genuinely gets used — Bruce’s example: a workers’ comp claimant photographed doing a slam dunk while on a back-injury claim
• The real challenge isn’t availability of data — there’s an enormous amount publicly accessible — it’s determining what’s credible, fair, and legal to use, a bar that gets harder to clear as scrutiny increases

Documenting the Algorithm, Not Just the Rules

This was one of the sharper technical threads in the conversation. I pushed Bruce on whether documentation requirements — historically applied to simple rules-based underwriting logic — now extend to machine learning models themselves.

His answer: yes, and this is a genuine shift from how the industry operated even a few years ago.

• Percipience’s approach is to run large volumes of test cases through a model specifically to check that outcomes aren’t patterned in an inappropriate way — checking for unintended discrimination by race, geography, or other protected characteristics that a model was never explicitly told to use
• The core problem is that many of these models arrive at decisions through pathways that aren’t fully visible or interpretable after the fact
• His illustrative example: a hiring model trained on the résumés of an existing engineering or physics teaching staff will overwhelmingly reflect a bias toward male candidates in STEM roles, purely because that’s the pattern in the training data — and will penalize an equally qualified female candidate for not matching that profile, despite gender never being an explicit input

The takeaway: documentation now has to demonstrate not just what rules a model follows, but that it was tested against exactly this kind of latent bias — because that testing is what actually holds up under a regulator’s or a court’s scrutiny.

Why Integration Is the Real Bottleneck

I asked Bruce what the biggest challenge is for consumers of core systems — Guidewire, Duck Creek, Majesco — when trying to integrate data across them. His answer reframed the problem away from data models and toward context.

• Every core system has its own proprietary data model, but a data structure is, in his words, “eminently conquerable” — the harder problem is understanding the context in which data exists inside that structure, like how each system handles out-of-sequence transactions differently
• Getting an apples-to-apples comparison across systems that structure the same underlying business event differently is genuinely difficult, not a matter of simple mapping
• Citing a Celent study: more than two-thirds of US P&C carriers run three or more policy systems simultaneously — meaning most carriers are never working from a single, unified application in the first place

Percipience’s current flagship engagement is instructive at scale: integrating 27 different core systems — policy, underwriting, billing, claims, accounting, finance, and two separate investment systems, plus multiple MGA program feeds — into Data Magnifier for a single large client, with the first phase targeted for a 10-month delivery window. Bruce noted they’re about two and a half months from delivering that first phase at the time of recording. Percipience’s next release, planned for January, is aimed specifically at reducing the time and complexity of this kind of integration through more automated, AI-driven tooling.

How Tier-One and Regional Carriers Decide Differently

Bruce’s segmentation of Percipience’s own customer base maps directly onto how differently insurers approach the build-versus-buy decision.

• Large carriers — the ones who got large in part by believing they could build genuine competitive advantage in-house — often want to own their data solution outright, and some have done it successfully. But Percipience’s best customers, by Bruce’s account, are companies that failed two or three times first: their first tier-one client had already sunk tens of millions of dollars into failed internal attempts before Percipience got them into production in nine months
• Smaller and mid-tier carriers simply don’t have the IT resources for complex integration work, no matter how elegant the eventual data model or analytics layer might be — they need a packaged capability that delivers value with minimal custom development
• Percipience’s product strategy is built to serve both: a core platform that works out of the box, plus a client extension framework that lets larger customers build genuine competitive differentiation on top without starting from scratch

Bruce’s historical parallel: policy administration systems went through the same arc in the 1980s and ’90s, when most carriers wrote their own, and by the 2000s the overwhelming majority had shifted to licensed vendor platforms. He believes data solutions are at the very early stage of that same curve — and that ten to fifteen years from now, very few carriers will still be building data infrastructure from scratch. His advice to clients making this decision now: invest in vendor solutions today whose future enhancement is funded across an entire portfolio of customers, not just your own budget.

Sorting Signal From Noise on the ITC Expo Floor

We talked about the experience of walking the exhibitor floor at ITC Vegas — rows of vendors, each claiming unique data or a unique platform, some of which won’t exist in a couple of years. Bruce’s framing for sorting them:

• Data providers — companies selling raw or enriched data itself
• Enrichment and services companies — adding context or structure to existing data
• Data platforms, where Percipience sits — not providing data, but providing the mechanism to assimilate and derive value from all the data sources a carrier already has or is buying

His expectation is that the breadth and depth of available data sources in insurance will keep expanding, not slow down — which makes the platform layer, in his view, the more durable investment relative to any single data source.

The Real Opportunity: Process Re-Engineering, Not More Data

Asked directly where he sees the biggest unrealized opportunity, Bruce circled back to the same theme from earlier: most carriers still use data primarily as a retrospective performance-assessment tool rather than embedding it operationally into daily decision-making.

His explanation for why change has historically moved so slowly in insurance is structural: incremental improvements from one legacy system generation to the next were often too small to justify the investment required to leap to genuinely new technology. He believes the sheer expansion of data opportunity today is finally changing that math — even if many organizations haven’t yet recognized or acted on it. He also noted that insurers are no longer just competing with each other on customer experience; they’re competing with banks and every other digital-native business customers interact with daily, which raises the bar regardless of how the industry has traditionally paced itself.

Advice: Understand the Business Before You Support It

Asked what he’d tell his younger self entering the industry, Bruce’s answer was about sequencing his own career differently: he wishes he’d spent more time early on understanding what actually drives business decisions and processes, rather than approaching insurance primarily from the technology-support side first. His view is that most people on the technology side of insurance are technologists first who then apply that lens to the business — and that there’s more value in learning the business first and bringing technology to it second, a sequencing he says would have let him collaborate more effectively with business leadership earlier in his career.

Key Takeaways

• More than two-thirds of US P&C carriers run three or more policy systems simultaneously — a unified data view is the exception, not the norm
• The hard part of integration isn’t mapping data structures; it’s reconciling the different operational context each core system embeds in how it stores the same underlying event
• Machine learning models used in underwriting now require documented bias testing, not just documented rules — a model can discriminate on a protected characteristic it was never explicitly given
• Carriers that failed at building their own data infrastructure two or three times make better platform customers than carriers who haven’t tried yet — they understand the real difficulty
• Smaller carriers need packaged, low-custom-development platforms; larger carriers want an extensible asset they can build genuine competitive advantage on top of
• Data solutions are following the same build-to-buy trajectory policy administration systems followed in the 1980s-2000s, just twenty years behind
• The biggest unrealized value in insurance data isn’t acquiring more of it — it’s operationalizing it into underwriting, claims, and product decisions in real time, rather than reviewing it only after the fact