Why fragmented intelligence is commercial underwriting's defining problem

There's a pattern we've observed across the commercial P&C insurers we work with. Each one has invested, repeatedly, in better technology. Better submission portals. More sophisticated rating engines. More capable workflow platforms. The investments were genuine. The returns were partial.

The tools improved. The combined ratios didn't follow. Underwriting teams got more software and more submissions. They got busier, not faster.

We think we know why. And it has nothing to do with the quality of the tools.

The problem is the architecture

Data, logic, judgment, and learning are split across systems that were never designed to share a common picture of a risk.

When a new submission lands today, a pricing actuary's model sits in one system. An underwriter's workflow sits in another. Portfolio signals that would change the decision sit somewhere else entirely. Each of those systems might be good at what it does. But they don't share context when it actually matters. Pricing strategy changes in the rating engine. Triage logic stays the same. An underwriter makes a judgment call. It doesn't feed back into future pricing calibration. A claim pattern emerges. Someone has to manually surface it and figure out what it means.

This is what fragmented intelligence costs. Not the sum of individual inefficiencies. The compound effect of systems that never talk to each other at the moment a risk needs to be understood.

A new challenge for commercial insurance, a 20-year-old problem in tech

We're not alone in this observation. Andreessen Horowitz recently described the same pattern playing out across enterprise software. Their analogy: the friend graph in social media was once the entire moat. All the value lived in the connections. Then the news feed arrived. The graph didn't disappear. It became infrastructure: one input into a reasoning layer that did the actual work of surfacing what mattered.

Their argument is that the CRM is now following the same path. Salesforce keeps its database. But the layer where sales reps actually work, where they get context, make decisions, and take action, is migrating upward to what a16z calls a system of intelligence. The switching costs that used to come from "all our data is in Salesforce" start to come from "all our reasoning, our workflows, our accumulated institutional context live in the intelligence layer." Salesforce becomes infrastructure. The gravity moves to the layer that sits above it.

In GTM software, this is still playing out. In commercial insurance, the structural conditions for exactly the same shift are already in place. The catalyst is arriving now.

Why now?

The best analogy for what's coming is mobile banking. When it arrived, the institutions that tried to shrink the desktop interface down to a phone screen struggled. The transformation only happened when they redesigned from first principles. The question wasn't "how do we put online banking on a phone?" It was "what does mobile-native banking actually need to be?"

The same question is now in front of commercial P&C. AI agents are starting to handle work that genuinely shouldn't require human attention: pulling submission data, running quality checks, flagging anomalies. Why is this risk showing a 13-and-a-half-month BI waiting period when we've never written one? Where did the 2011 claims history go? That kind of observation used to take an actuarial analyst two to three days on each major risk. It will be table stakes.

But agents can't operate on fragmented systems. When data, logic, and portfolio context sit across disconnected tools, an agent either makes recommendations without the full picture or escalates to a human who has to reconstruct the context that should have been shared. The bottleneck isn't the model. It's the missing intelligence the model needs to work from.

The carriers that get to agent-led underwriting first won't be the ones with the best models. They'll be the ones that solved the underlying architecture problem first.

What is the intelligence layer in commercial insurance technology?

It's not a new tool. It's an architectural change.

An underwriting intelligence layer is a single, living understanding of risk, shared across pricing, underwriting, and portfolio decisions. When pricing strategy changes, triage logic updates automatically because both run from the same models. When a submission is extracted incorrectly, the underwriter can see exactly where it came from and correct it in seconds. When a pattern emerges in the portfolio, it feeds back into how the next risk gets priced.

This is what makes the switching cost migrate. Away from the spreadsheet, the PAS, the rating engine, and toward the accumulated judgment, calibration, and context encoded into how decisions get made. Every decision that runs through the system makes the next one better.

What this creates for commercial P&C leaders

The sharpest observation in the a16z piece is about institutional memory: captured in the right system, it becomes something a company can actually ship. In commercial insurance, the stakes of that observation are particularly high. The pricing actuary who retires carries fifteen years of market-cycle intuition that no spreadsheet fully captures. The senior underwriter who moves carries his risk assessment patterns with him. An intelligence layer that has been ingesting the full context of every decision, not just outcomes but the judgment behind them, changes that. The learning stays.

Most large carriers know their underwriting technology stack is a problem. Submissions are piling up. Deployment cycles are too long. Portfolio visibility arrives too late. The question isn't whether to change. It's whether to lead the change or follow it.

Carriers that build the intelligence layer first will build a compounding advantage: faster responses, better risk selection, tighter feedback loops between decisions and outcomes. Carriers that wait will still have their PAS, their models, their workflow tools. They'll just find themselves answering to an intelligence layer someone else designed.

That principle runs through every part of the industry right now, including distribution. How conversational commerce and LLM-driven discovery reshape broker and aggregator relationships is the same question applied to a different layer. Companies that design the agentic frameworks for their market will fare better than the ones that get fitted to frameworks built by someone else.

What is the future of underwriting?

None of this changes the core of the discipline.

Actuaries didn't become actuaries to debug spreadsheet models. Underwriters didn't build careers to re-key data into three systems and wait two days for pricing input. No one stood outside the Institute of Actuaries hoping to spend their career drawing borders around cells.

The intelligence layer removes that overhead. What remains is the work that actually requires judgment: assessing a risk that doesn't fit the model, steering a portfolio through a market shift, building the broker relationship that earns the next renewal.

The best version of agent-led underwriting isn't less human. It's more human in the ways that matter. Underwriters spending time on risk assessment, not data entry. Actuaries working on pricing strategy, not spreadsheet maintenance. CUOs with real-time portfolio visibility to act on, not delayed reports to wait for.

That's what building on the right foundation makes possible. Not replacing the judgment. Making room for it.