The 'Say no early' protocol: Lloyd's wisdom as the blueprint for A2A

By Alexis Renaudin, Head of Data at Artificial

During my time at Artificial, I have spent a lot of time thinking about friction.

Specifically, the "hidden friction" in commercial insurance. It's not the big, obvious stuff. It's the death-by-a-thousand-cuts: the re-keying data from a PDF, the follow-up email to chase missing files, the clarifications about data points, the mismatched formats.

The grinding friction of 'ping-pong'

Individually, these are just minor annoyances. But collectively, they create this slow, grinding hinderance, which I like to call the "ping-pong" interactions : long and inefficient streams of back and forth clarifications. A broker lobs a submission. The underwriter lobs back, "Missing data on X." The broker replies, "Here is X." The underwriter lobs, "Ok, but what about Y?"

Each lob can take hours, even days. A simple placement gets stretched from a one-day decision into a two-week ordeal. It's inefficient, it's costly, and it drains the human expertise from the system, replacing it with admin.

Borrowing from Lloyds’ wisdom: The 'say no early' Protocol

I recently stumbled across a 2019 video of Nigel Kemble-Clarkson, a long-serving Lloyd’s broker and respected commentator on the London market, where he was laying out the ‘unwritten’ rules for underwriters. His number one rules reads:

"If you do not intend to write the risk, say so straight away without wasting the broker's time by asking a load of daft questions in order to justify your refusal. A decent broker will always accept ‘p**s off’ but is likely to get very p****d off with a prolonged negative result."

This is actually the "ping-pong" problem, perfectly articulated. He's describing a human-run protocol to stop it dead, essentially preventing a denial-of-service attack on a broker's time.

He also has a matching rule for brokers:

"Never tell porkies. Trust is the essence of face to face dealing, which we're all determined to preserve."

What struck me is that these two rules are the absolute, fundamental core of an efficient market:

1. Don't waste people's time on discovery if the answer is "no"

2. The data exchanged must be 100% trustworthy and straight to the point

The challenge isn't that the market doesn't know the solution. The challenge is that we're trying to execute this high-throughput, high-trust protocol using tools (email, phone calls, spreadsheets) that are inherently inefficient, low-trust, and terrible at it.

What if we codified the wisdom?

This got me thinking. What if we could build a system that had all this logic baked in?

What if an underwriter could deploy a "digital representation of themselves" - an Underwriter Agent with access to all the right context and data - programmed with their explicit "say no early" appetite rules?

And, and this is the key bit, what if the broker could do the exact same thing? Ie a Broker Agent that holds the "no porkies" client’s data, ready to share and present the risk with the relevant agent(s).

This Agent-to-Agent (A2A) model isn't just a theory. It's a concept that's gaining real technical legs. It's an idea I've been entertaining for a while, so it was fascinating to see Google recently release their own open-source A2A protocol, a powerful validation that this is the right direction.

What's particularly exciting is that while their initial work focuses on agents collaborating within the same environment, I see A2A as a secure bridge between different organisations, like a broker and an underwriter.

Let the agents handle the ‘ping-pong’

This is the core of the concept. Instead of a human broker starting the "ping-pong" with a human underwriter, the Broker Agent talks directly to the Underwriter Agent, over a secure A2A protocol. The agents, digital representations of an individual broker, underwriter, or team, have that entire initial conversation autonomously.

This automated conversation solves a deeper, more stubborn problem than just 'missing data.' It solves the format and convention problem. Every broker structures their risk data differently. Every underwriter has their own required fields and naming conventions. Today, a human has to manually translate this. But agents can reason. They can interpret 'Insured Limit' vs. 'Policy Limit' vs. 'Sum insured,' and if they encounter ambiguity, they can clarify it via dialogue, all autonomously. That lengthy human 'ping-pong' and pondering over data disparity vanishes.

At the same time, the Underwriter Agent runs its "say no early" rules. It can’t ask "daft questions" to be polite; it just checks its appetite rules. The Broker Agent provides the data. It can't "tell porkies"; it just serves the facts from its source.

That "prolonged negative result"? That two-week email chain? It's compressed into a two-second, auditable chat log. The only risk here is one agent "wasting" another agent's time. Which is precisely the point, as we’re now looking at a "transaction" that costs fractions of a penny, not thousands of pounds in human hours.

Beyond the hype: A grounded agentic insurance future

We've been trying to solve a protocol problem with better inboxes. It's time we built a better protocol, one that drives real structural change and makes commercial insurance future-proof by leveraging inevitable AI progress.

There’s a deafening amount of noise and buzzwords in the market right now about AI and agents. A lot of vague and mostly overhyped promises about "replacing people and automating workflows" with some magical agentic black boxes.

I think that's profoundly the wrong way to look at it.

The real, structural future of agentic AI, especially in a trust-based market like insurance, isn't about opaque point solutions. It's about something far more practical, secure, and controlled. It’s about your agent, owned and deployed by your company, running securely on your proprietary rules and your intellectual property.

This vision is about secure, transparent, and auditable agents talking to each other. They are designed to handle the mundane discovery, the data chasing, and the "ping-pong" over mismatched data, freeing up human brokers and underwriters to focus on what they alone can do: the complex negotiations, the nuanced judgments of underwriting, and the relationships that actually run this market.

This is our philosophy at AgLabs. It's not a toy; it's a structural and futuristic shift in transaction throughput.

In the rest of this series, our team will explore the technology, the security, and the practical application of this new agentic frontier.