How to shift AI from productivity enhancement to agentic risk trading

The AI maturity curve, introduced in our last article, ends on a stage that is easy to agree with and easy to underestimate: digitization is the foundation, orchestration is the transformation. That single distinction is the difference between a carrier that plateaus at stage three of the curve and one that reaches stage four; software that reads your submissions versus software that runs them. It is worth slowing down on, because almost every stalled or underdelivering AI programme in commercial insurance stalls at exactly this line.

The existing paradigm and the new horizon

Digitization is the act of turning unstructured communications -the broker email, the PDF schedule, the loss run, the third-party report - into structured, decision-ready data. It answers one question: what does this submission say? Sophisticated digitization will also enrich with external data, use agentic reasoning to calculate and infer, and it will even summarize and recommend next steps. However, the current approach taken to submission completion and AI technologies means that workflows cannot be progressed automatically. This is where orchestration comes in.

Orchestration is the self-executing flow of risk across the steps and workflows that follow - running appetite checks, invoking pricing andrating engines, chasing missing information, routing to the right person -without a human driving each step. It answers a different question: what happens next, and who or what makes it happen? Digitization is a foundational capability upon which orchestration must be built, but it has a ceiling when it comes to truly automating workflows.

When most vendors and most carriers talk about automation, what the really refer to is digitization. Technically this isn’t incorrect, digitization automates the production of data, but with today’s AI capabilities we can think bigger. We can automate the production of movement within workflows.

Digitization: necessary, foundational, and not the destination

If your organization has invested in digitization, that investment was right. Digitization is the infrastructure layer of any modern underwriting operation:ingest from any channel, extract the fields that matter, enrich with external data. It maps to stage one of the maturity curve, is a critical foundation to the following stages, and it is genuinely hard to do well accurately and at scale.

But digitization has a ceiling, and the ceiling is structural rather than a question of quality. Done on its own, it is static and per-communication. Each email, each attachment, each submission is digitized in isolation - read once, structured, and handed on. There is no memory of the other communications that concern the same risk, no linkage across the stages of that risk's lifecycle, no bridge to the data sitting in a parallel workflow. Critically, there is also no imperative or mechanism to progress a workflow past the completion of a submission. A carrier can digitize every document that crosses its desk at 98%accuracy and still be making decisions one disconnected fragment at a time.

That is the plateau. It is why “we've already digitized” is both true and not enough.

The real dimensions of risk digitization

The reason digitization plateaus becomes clear once you consider the dimensions risk moves along. Digitization is not one thing a carrier either has or doesn’t have. The dimensions of risk create layers. Each layer wraps more context around the same risk, and the more layers in place, the closer the data gets to being genuinely decision-ready, and when all of the layers are considered carriers progress from transactional digitization to agentic risk trading.

The layers of risk context

Layer one: data provided by the broker. The submission itself: the email, the schedule, the loss run. Classifying these files, then reading and structuring this data is the minimum criteria for digitization.

Layer two: data from external sources. The submission enriched with firmographics, property attributes, hazard and peril data, financial and sanctions signals: everything the broker didn't tell you. The is where most of the market stops.

Layer three: linked communications about the same risk. The follow-up email three weeks later, the revised schedule, the surveyor's report, recognized as belonging to a risk the carrier has already seen and folded into one picture, rather than digitized again as an unrelated document.

Layer four: data from the same risk or client within the insurer. The earlier submission, the bind, last year's endorsement, the open claim: the carrier's own history with this risk, wherever in the business it sits.

Layer five: data from similar risks within the insurer. The book itself: how comparable risks in the same class, geography and size band have been priced, written and how they have performed.

Each layer added progresses carriers from transactional digitization, where the risk is read accurately, but in isolation, toward digitization that produces decision-ready risks, where communications get linked and the submission arrives contextualized with what the broker said, what external sources say, what the carrier already knows about this risk, and what it knows about every risk like it.

Decision-ready risk is the destination of digitization, and reaching it is genuinely hard and genuinely valuable. But it is worth being precise about what it does and doesn't deliver. It makes a risk knowable; but doesn’t make it move. Getting from a decision-ready risk to a quote still requires something to act - to run the appetite check, invoke the rating engine, request the site visit and pull its results back into the workflow. That is orchestration, and the state it unlocks is agentic risk trading. Most of the market is clustered in the transactional corner and calls it digitization. Agentic risk trading is the North Star.

What are the consequences of incomplete digitization?

A fair question follows from the layers: if only layers one and two are in place, is orchestration off the table? Not necessarily - but what it can achieve is bounded by the context beneath it. Orchestration executes on what it knows, so a thin picture of the risk means appetite checks on partial data, referrals triggered by missing fields rather than judgement, and low straight-through processing as every gap pulls the underwriter back into the mechanics. What is at stake is not permission to progress but the autonomy the flow can be trusted with. This is why digitization and orchestration are built in parallel rather than in sequence: start orchestrating the workflow you have, and each layer of context you add raises the ceiling on how much of it can run itself.

Orchestration progresses risk

Orchestration is what carries a carrier from the transactional end of the spectrum toward the agentic end. It does three things digitization alone cannot.

It links every communication about the same risk, however far apart in time, into a single holistic policy view - detecting that a follow-up email three weeks later belongs to a submission it has already seen, and updating the picture rather than starting over.

It acts across parallel workflows - so data extracted and enriched in one place is available everywhere it is relevant, and the carrier's entire data arsenal is democratized rather than compartmentalized.

And it executes the steps between data and decision - invoking a rating engine, running an appetite check, ordering the next piece of missing information, requesting a site visit and digitizing the results - so risk flows toward a quote on its own, with the underwriter engaged for judgment rather than mechanics.

This is the shift from software that describes a risk to software that acts on it. Digitization tells you what the submission says. Orchestration authors your view of the risk and carries it forward.

Multi-participant risk flows across the entire organization

This is where agentic risk trading earns its precision. A productivity tool assists one person with one task; an orchestrated flow is populated by multiple agentic personas, each holding a mandate and accountable for a step - intake linking communications, exposure resolving schedules, appetite testing against guidelines, referral assembling the case for a human. Around them sit multiple participants - broker, underwriter, surveyor, claims handler, reinsurer - who each need a different view of the same risk at a different moment, so the risk itself becomes the shared object: a revised schedule reprices the aggregate, re-tests appetite, and reaches the underwriter as a changed recommendation rather than a new email. Productivity enhancement makes each participant faster at their step; agentic risk trading removes the steps as discrete human events, leaving judgement, negotiation and appetite.

What orchestration unlocks that digitization can't

Step back from the mechanics and there are only three outcomes a carrier is really buying: getting to a quote faster, growing premium without growing headcount at the same rate, and improving risk selection. These outcomes are not solely intake problems. A submission that is read perfectly on day one still sits waiting if the appetite check, the referral, the rating call, the missing-information chase and the site visit each need a person to pick them up. Time to quote is the sum of every wait between those steps, improving risk selection demands more time, and headcount is what carriers add to absorb that time. That is why the objective can only be reached by orchestrating the entire underwriting workflow rather than the front door of it - the compounding gains come from the handoffs, not the extraction.

The distinction is not academic; it shows up in the numbers, and specifically in numbers that extraction alone cannot produce. A Tier-1 global carrier deployed this model globally and manual triage time fell by 80% thanks to a rise in digitization accuracy from 70%  to 98%. The model was linking and routing risk, not just reading it, meaning context across communications resolves the ambiguity a single document leaves behind. Intake straight-through processing climbed from 10% to 95% - a figure that is impossible without orchestration, because straight-through means the flow executes itself. Better extraction nudges the first number. Only orchestration moves the last one.

Digitization is the foundation of orchestration

The takeaway is an uncomfortable one for a market that has spent several years and considerable budget on digitization. Digitization is complex and necessary, both in of itself and as the enabler of agentic workflows, but in isolation it is not enough. There are further steps to be taken toward decoupling premium growth from headcount and becoming the fastest to quote. There are opportunities to further differentiate, and those opportunities live in the orchestration layer.

Which raises the question the rest of this series turns on. If the value has moved into the model that orchestrates the flow, then the interface a carrier has spent so long thinking about is no longer what drives the design. The screen still has a role. It simply stops being the thing the workflow is built around.That is the case for going headless - and it is where we go next.

The handoff between digitization and orchestration

So where is the handoff - and can one platform do both? It can, and it should. Most of the market treats them as two purchases for architectural reasons: digitization vendors were built to produce data, workflow vendors were built around a screen that assumes a human is driving, and the seam between them is where value leaks. A headless architecture collapses that seam - when the model rather than the interface is the system of action for the risk, digitization and orchestration become two behaviours of the same engine, able to loop back to extract, enrich or link at any point. Very few solutions on the market are built for both sides of that line, which is precisely why carriers end up owning the integration risk themselves. The rest of this series will explore in greater depth how headless architecture allows carriers to perfect digitization, unlock orchestration and achieve agentic risk trading.

‍