← AgileMesh
Research / Communications

The Insurance Communication Gap

Multiple policy admin systems create a bottleneck that disconnects insurers from their customers. The problem has a name.

11 March 2026

A large insurer running 40+ brands will typically operate multiple policy administration systems -- one per brand, per line, per acquisition. Each system holds its own customer records, claims data, and underwriting history. None of them talk to each other. The result is a communication bottleneck at the centre of the business.

Lemonade Insurance processes 96% of first notices of loss without a human touching them. Their AI settled a claim in 2 seconds -- reviewing the policy, running 18 anti-fraud algorithms, and paying the customer. They can do this because they have one system, one data model, one view of the customer. The bottleneck does not exist.

For traditional insurers, the bottleneck means the customer, the claims handler, the underwriter, and the board all live in different information universes. Data flows one way -- upward, into reports that nobody reads until the regulator arrives.

We looked at the structure. This is what we found.

Two Models of Insurance Communication

There are fundamentally two architectures for how information moves through an insurance organisation. Most companies use the first. The ones winning use the second.

TRADITIONAL Sales Claims Underwriting Management Reports (monthly, one-way) Board (quarterly review) Customer: no feedback loop CONNECTED Knowledge Graph Sales Claims Customer Under- writing Board
Left: Information flows upward, never back. Right: Everyone reads and writes to the same graph.

In the traditional model, a claims handler in Portsmouth has no idea what the underwriter in London priced the risk at, what the customer said on Trustpilot last week, or what the FOS upheld about a similar claim last month. Each silo generates its own reports. Nobody connects them.

In the connected model, every interaction -- a quote, a claim, a complaint, a renewal, a phone call -- becomes a node in a knowledge graph. The relationships between these nodes are the intelligence. The shape of the data is the answer.

The Numbers

What AI-first communication looks like

96%
Lemonade claims opened without human intervention
2s
Fastest claim settled -- policy review, fraud check, payment
12pt
Loss ratio improvement in one year (79% to 67%)
3M
Policyholders, growing 23% year-on-year
6.1x
Shareholder return: AI leaders vs laggards over 5 years (McKinsey)
$1.24B
In-force premium, growing 31% year-on-year

These are not projections. These are audited results from a company that started in 2015 with no actuarial history, no broker network, and no legacy systems. What they had was a communication architecture where every piece of data feeds back into every decision.

The 2.3x multiplier: J.D. Power's 2025 Claims Digital Experience Study found that customer satisfaction more than doubles -- from 337 to 777 -- when communication during a claim is easy. Not when claims are paid faster. Not when payouts are larger. When communication is easy. The channel is the product.

What the communication gap costs

42%
One-star reviews at a major UK insurer despite 4.3/5 headline score
52%
FOS uphold rate for pet insurance complaints -- highest of any line
21%
Employees who would recommend their own company (Glassdoor)
The paradox: A 4.3-star Trustpilot score masks extreme polarisation. 48% rate 5 stars; 42% rate 1 star. There is almost no middle ground. Customers either had a good experience or a terrible one. The difference is usually whether their information reached the right person at the right time.

Why the Gap Exists

Traditional insurers are not stupid. They are structurally constrained. The root cause is specific and identifiable:

1. Multiple policy admin systems. A group with 40+ brands will typically run a separate policy administration system for each brand, line of business, or acquisition. Each system holds its own customer records, its own claims history, its own underwriting data. A customer complaint about one brand never reaches the underwriting team that priced the policy under a different brand. The systems are the bottleneck. Everything downstream -- the one-way reporting, the disconnected customer experience, the regulatory blind spots -- flows from this.

2. One-way data flow. Because no single system holds the full picture, information can only move upward through management reporting. It never flows laterally between functions, and it never flows back to the point of customer contact. A claims handler cannot see what the customer said on social media. A social media manager (if one exists) cannot see claims data. The result is what we call the Panopticon problem -- a system designed for surveillance that nobody watches.

3. Legacy communication channels. Phone, email, and letter are still the primary customer contact methods. These are inherently one-to-one, synchronous, and unstructured. They generate no graph. Every conversation dies when the call ends.

The Lemonade difference is not AI. It is one policy admin system feeding one data model. Every customer interaction -- every message, claim, photo, question -- feeds into a single connected graph. The AI is just the read head. The graph is the memory. The bottleneck does not exist because there is nothing to bottleneck.

What Connected Communication Changes

Capability Disconnected Connected
Claim notification Phone call, 15-min hold App message, instant acknowledgement
Fraud detection Manual review, days 18 algorithms in 2 seconds
Customer sentiment Quarterly board report Real-time graph signal
Cross-sell signal Batch marketing campaign Event-driven, contextual
Complaint resolution Escalation ladder, weeks Root cause visible instantly in graph
Employee knowledge Tribal, lost on turnover Encoded in graph, permanent
Regulatory evidence Retrospective data gathering Continuous, auditable provenance

The Ontology Approach

An ontology is a formal map of what an organisation knows and how those things relate. In insurance, this means: a Customer holds a Policy, which covers a Risk, which has a Claim, which references a Loss Event, which occurred at a Location, which was assessed by a Handler, who works in a Team, which reports to a Division.

When these relationships live in a knowledge graph rather than in spreadsheets, something emerges that no flat report can deliver: you can ask questions that cross boundaries.

Which handlers have the highest customer satisfaction for pet claims in the South East?

Which broker relationships produce policies with the lowest loss ratios?

Which complaint patterns predict FOS escalation before it happens?

These are not hypothetical questions. They are the questions that connected organisations answer every day and disconnected ones cannot even ask.

The shape of the data is the answer. When you map an insurance operation as a graph -- customers, policies, claims, handlers, brokers, locations, risks, complaints -- the hidden concentrations, the broken feedback loops, and the communication dead-ends become visible as topology. No spreadsheet shows topology. The graph does.

What This Means in Practice

An insurer with 8 million customers and 30 million quotes per day sits on one of the largest behavioural datasets in the UK. Every quote that does not convert is a signal. Every claim that takes longer than average is a signal. Every customer who leaves without complaining is a signal.

The question is not whether these signals exist. It is whether anyone can hear them.

A connected communication architecture -- graph-based, real-time, bidirectional -- turns those signals into action. Not through more meetings, more reports, or more dashboards. Through structure. Through making the hidden connections visible.

7%
UK insurers who have scaled AI. The other 93% are still experimenting.
$50-70B
McKinsey estimate: generative AI revenue opportunity in insurance

The 60-point gap between testing and scaling is not a technology problem. It is a communication problem. The data exists. The connections do not.

The Inverted Panopticon

The traditional insurance model is a panopticon: the insurer watches the customer. Surveillance-based underwriting, claims investigation, fraud detection. The customer submits; the insurer judges. Information flows one way.

Lemonade inverted this. They made themselves transparent to the customer. Open-source insurance policies on GitHub. Transparency Chronicles publishing their failures and internal metrics. A flat-fee model where the insurer has no financial incentive to deny claims. The Giveback program donating unclaimed premiums to the customer's chosen charity -- over $12 million since 2017.

The result: customers who feel watched become adversarial. Customers who feel trusted become honest. Dan Ariely, Lemonade's Chief Behavioral Officer, designed this deliberately. When fraud feels like stealing from charity rather than from a faceless corporation, dishonesty drops. When the insurer's incentives are visibly aligned with the customer's, trust becomes structural rather than aspirational.

The panopticon works. It just works better when the customer holds the watchtower.

Further Reading

We explored the Panopticon concept -- why one-way data architectures kill organisational performance -- in a separate analysis:

The Panopticon: Communications for Business →

And our analysis of pension fund AI concentration -- the same structural blindness applied to investment:

Your Pension's Hidden AI Bet →

We build the connective tissue

Knowledge graphs. Ontology design. Communication architecture that makes hidden connections visible.
If your organisation has the data but not the structure, we should talk.

[email protected]
Sources and methodology: Lemonade data from SEC filings (10-K, 2025) and investor presentations. UK insurer data from Trustpilot public profiles, FOS annual report 2025, Glassdoor public reviews, FCA register. McKinsey AI leader/laggard TSR analysis from "The state of AI in insurance" (2025). All Trustpilot scores verified February 2026. Company names anonymised where analysis is structural rather than specific. AgileMesh has no financial relationship with any company mentioned.