Introducing Subcontext

Subcontext is a quote and bind platform that plugs directly into an insurer's underwriting and product rules, with a decision layer on top that lets an AI agent speak to customers through WhatsApp, email, SMS and web chat under the insurer's own brand.

This is not the deterministic, rules-based chatbot technology.

Our technology uses a conversational agent that gathers context, asks the right next question, and progresses the customer from intent to bound cover in one continuous conversation. The agent operates conversationally in natural language, gathering context and progressing the customer journey dynamically rather than simply following scripted flows.

In practice the conversation does the work that an application form, a medical questionnaire and a point of sale used to do separately. The agent interprets what the customer means in context, asks whichever question logically comes next, recognises when something a customer has said carries underwriting implications, and routes the conversation accordingly. From the customer's perspective they are not filling in a form at all, but having a conversation that happens to produce a bound policy at the end of it.

Why we built it

Most life and health protection products are still bought in much the same way they were twenty years ago. A customer who decides on a Sunday evening that they need cover will typically encounter a contact form, a callback queue and a folder of static PDFs, and by Monday morning whatever prompted them to look has usually faded. The cost of distribution stays high in this market largely because human telesales has historically been the only way to bridge the gaps that the technology stack leaves open.

The fast-quote, instant-bind digital experience that works in general insurance has been harder to translate into life and health because the underwriting is more involved, the disclosures are more sensitive, and the product rules denser. And the obvious-sounding response of replacing the underlying systems has, on the occasions it has been attempted, tended to absorb several years of programme time without producing much commercial gain.

Subcontext is built on the view that the insurer's IP is by and large fine. The product definitions, the underwriting rules, the rating tables and the reinsurance arrangement collectively represent a great deal of accumulated knowledge, and they are not really the problem. The problem is the engine layer that has historically sat between those rules and the customer.

What the platform actually does

The platform has two layers. The first is the quote and bind engine, which is the part that actually performs the underwriting. The insurer brings their own product definitions, their underwriting rules and their reinsurance treaty constraints, and the engine takes those in as configuration and runs them natively, pricing the risk, generating the policy documentation and binding cover at the end of the conversation. From an insurer's perspective the engine replaces the patchwork of legacy components that have historically handled quote, decision, pricing and bind across separate systems, with a single coherent engine that handles all of it as one process. This is a quote and bind engine designed for the agentic sales process.

The second is the decision layer, which is where the agent lives. The decision layer holds the conversation, decides what to say next, decides when to query the quote and bind engine for a price or a decision, decides when a customer's situation needs to be escalated to a human, and decides when the customer is ready to be presented with terms. Its job is essentially to turn the underwriting process into a conversation rather than a questionnaire.

Building the quote and bind engine from scratch was a deliberate choice. We did not want to put a thin conversational layer on top of systems that were originally designed around paper applications, postal forms and call centres. Instead we've built an engine that is AI-native from the ground up, designed with the next generation of life and health products in mind. That means it accepts the kinds of inputs that traditional protection systems were never really designed to handle, including app-based customer engagement, third-party data extensions, wearables signals and continuous health information. The data model is built on the expectation that what counts as relevant evidence for an underwriting decision is going to keep expanding over time, and the engine's underwriting hooks are designed to keep up as it does.

Why conversational, not scripted

Conventional chatbots tend to fall over in insurance contexts because an insurance application is not a linear form. The answer to one question often changes which questions matter next. A disclosure made halfway through a conversation can change the set of products on the table; a pre-existing condition can shift the underwriting path; a change of heart about the cover amount can move the conversation into a different affordability discussion altogether.

A rules-based bot handles this kind of variability by branching. Each branch is a rule that someone has had to write, and each rule then becomes something that needs to be maintained. In practice the result is usually either a bot that can only handle the simplest journeys or one that has become so brittle that it cannot easily be changed.

Subcontext takes a different approach. The agent reads the underwriting rules, the product rules and the conversation so far, and decides what to ask next based on what it actually knows at that point. When the customer says something unexpected, the agent understands it in context. When a disclosure opens up a new line of questioning, the agent follows it. When the customer wants to ask a question of their own, whether that is about the product itself, about what a particular term means, or about whether some historical health issue is going to be a problem, the agent can answer the question and then pick up where the application had got to.

From the customer's perspective the experience feels much more like talking to a knowledgeable broker than filling in a form. From the insurer's perspective the result is a higher completion rate and a structured record of the conversation that the underwriting engine can act on directly.

The channels

The agent operates across WhatsApp, email, SMS and web chat, and does so under the insurer's brand throughout. These are not separate products with separate logic; they are different surfaces onto the same underlying conversation. A customer who begins an application on a website and finishes it on WhatsApp does not have to start again, because the context, the disclosures and the partial application all carry across.

Channel choice matters because the channel an insurer makes available determines who they can realistically reach. Web chat reaches the customer who is on the site already. WhatsApp reaches everyone else, and in many markets reaches a customer who would never have come to the website in the first place. Email and SMS handle the longer tail of the journey, including reminders, follow-ups, document delivery and any subsequent renewal conversations.

The brand remains the insurer's throughout the entire interaction. Subcontext itself does not appear in the conversation in any visible way. The customer experiences themselves as talking to the insurer, on a channel they already use, in language that reads as the insurer's voice rather than as a generic assistant.

Compliance and oversight

Selling regulated insurance products through a conversational agent is only really viable if the regulatory trail produced by the agent is at least as strong as the trail produced by human telesales. Subcontext logs every turn of every conversation, together with the reasoning behind each decision the agent took and the underwriting and pricing artefacts that produced the final terms presented to the customer. That log is fully replayable, so a regulator, an internal compliance function or a customer complaints team can reconstruct exactly what the customer was told and why.

The agent also knows when to stop. Conversations that fall outside what the agent is permitted to handle, including indicators of customer vulnerability, disclosures that require human underwriting review, and products outside the agent's remit, are escalated to a human automatically. The agent is not intended to be a substitute for human judgement at the edges of the customer base. It is intended to handle the volume of conversations that should never really have needed human time in the first place, so that the human time available can be spent on the conversations that genuinely do.

Building the future of protection

Protection has historically been a series of disconnected forms held together by people. Subcontext turns that into a continuous decisioning process, one in which the customer can be spoken to, underwritten, priced and bound inside a single conversation, through whichever channel they prefer to use, and in the insurer's own brand. The underwriting rules and the product rules and the treaty all stay where they are. What changes is that the customer can now reach them directly, at the moment that they actually want to.