Definity's planned AI assistant for insurance brokers illustrates a wider business opportunity: bringing reliable organisational knowledge into everyday workflows. The potential gains depend on document quality, clear boundaries and evidence that faster answers improve the work as a whole
A customer asks a question that sounds straightforward. Somewhere in the business, the answer exists: in a product document, a servicing guide or a clarification issued after the main manual was written. Finding it, confirming that it applies and explaining it accurately can take longer than the conversation itself. This is an illustrative scene, but it captures the organisational problem behind a new insurance announcement. On 12 November, Definity and Guidewire announced that Definity would enable ProNavigator within its Vyne broker platform as Definity Assist, covering personal auto, property and Individually Rated Commercial Auto. Brokers are intended to ask questions in ordinary language without leaving their existing workflow.
The companies expect quicker inquiry resolution and easier onboarding for new broker staff. Those are announced expectations, not independently demonstrated outcomes. The release does not supply a measured improvement in accuracy, handling time or financial returns, nor a detailed rollout timetable. Its importance lies in the business design it proposes: make knowledge available where a professional is already working. That proposition extends beyond insurance. A small advisory practice, a service company or a distributor could face a similar gap between having information and being able to use it reliably. The interesting question is how AI might narrow that gap without turning a convenient answer into an unchecked instruction.
The opportunity is inside the workflow
An additional search tool does not automatically simplify a working day. If employees must leave the current task, reconstruct its context and decide which result applies, some of the original burden remains. An embedded assistant could reduce those transitions. The organisational opportunity is therefore more specific than giving everyone access to a conversational interface. It is to connect a question to the task that prompted it and to the approved information needed to resolve it. In insurance, an answer may depend on the relevant product and version; in another business, it might depend on a service agreement or an equipment model. The business value would come from resolving that context correctly.
There is a corresponding reason to begin with a bounded problem. A hypothetical brokerage could initially examine recurring servicing questions whose answers are already documented, rather than ask an assistant to resolve every complex coverage issue. The distinction is practical: a narrow scope makes it easier to define what a correct answer should contain and when a person must take over. It also allows managers to examine whether the assistant removes work or merely changes its location. If staff spend less time searching but more time repairing ambiguous answers, the interface may look efficient while the underlying process improves little. The whole task needs to be observed.
Retrieval changes access, not the authority of a document
Guidewire’s product documentation describes ProNavigator as using Anthropic’s Claude models, retrieval from a client-specific document store, role-based access controls and paragraph-level citations. These are the supplier’s descriptions of the product, rather than an independent audit of Definity’s planned configuration. They help explain the mechanism being offered: retrieve relevant organisational material and use a language model to formulate an answer. The approach has a research lineage. The 2020 paper by Patrick Lewis and colleagues combined a language generator with retrieved external passages, studying knowledge-intensive tasks. Its results do not establish that a particular insurance assistant answers correctly in a live business setting.
For a manager, the distinction between access and authority is crucial. A model may find a passage that resembles the question while missing a condition elsewhere in the material. A citation can help a reviewer inspect the evidence, but its presence alone cannot show that the answer follows from it or that the source is current. An illustrative product question might require the latest servicing instruction rather than a closely matching older guide. The relevant test would examine the relationship between question, source and answer together. Buying retrieval capability does not relieve a business of deciding which documents should govern its work. It makes that decision more consequential.
Document ownership becomes part of the operating model
Before connecting an assistant to internal material, a business could discover that it has several competing versions of its own instructions. That possibility should be treated as a preparation question, not as a claim about Definity’s documentation. The useful response would be to establish ownership: who approves each source, how changes become effective and how superseded material is withdrawn. A document collection is not necessarily a coherent knowledge base. Informal notes may help experienced colleagues understand an exception while being unsuitable as a general rule. The work of separating approved guidance from background explanation requires people who understand the service, not only people who can configure software.
Maintenance then becomes a continuing operational responsibility. If a policy or process changes, the business needs a way to check that the assistant’s accessible material reflects the change and that relevant test questions still receive appropriate answers. Permissions need similar attention: making a document easy to retrieve must not make it available to someone who should not see it. The supplier says ProNavigator maps access to business dimensions and integrates with single sign-on. Whether the resulting arrangement suits a specific organisation must be tested during implementation. A useful acceptance exercise would include attempts to retrieve material outside a test user’s role, alongside ordinary questions that the user should be able to answer.
Measure the resolved inquiry, including the work that follows
The first evaluation question should concern the business process rather than the assistant’s popularity. A pilot could compare representative inquiries handled with and without assistance, recording correctness, necessary escalation and the time spent reaching a usable resolution. That would be an evaluation proposal, not a report of results from this announcement. Cases should include questions the system cannot safely answer, because recognising insufficient evidence is part of useful performance. Expert reviewers could examine whether the response identifies the relevant source and its limits. They would also need to distinguish a technically plausible explanation from one appropriate to the customer’s actual situation. Faster production of the wrong explanation has little operational value.
Costs need to follow the same boundary. A licence fee is only one part of the commitment; preparing documents, integrating access, reviewing outputs, training staff and maintaining tests all consume resources. The business case would be stronger if it connected those costs to a defined improvement, such as reducing repeated internal referrals while preserving answer quality. No numerical return can be inferred from Definity’s announcement. Even a measured reduction in search time would require interpretation: does it free capacity for additional work, reduce delays or improve the attention given to customers? A smaller company should make those questions explicit before treating theoretical time savings as money available to spend elsewhere.
A small firm can adapt the discipline without copying the platform
The announcement concerns an insurer’s integration with an established broker platform. It does not establish that the same product, pricing or implementation would suit a small practice. A more transferable idea is to organise a limited body of approved knowledge around a recurring task. An illustrative independent service business might begin with internal questions about standard procedures, keeping customer-specific decisions with staff. It could assess a permission-controlled tool against a small set of reviewed documents before attempting a broader connection. This is a possible adaptation that would require validation, not a recommendation that every firm should adopt a particular application or upload its records to a general-purpose chatbot.
Scale changes the economics of ownership as well as the technology. A large organisation may distribute content review and implementation across specialist teams; a small firm may rely on the same person to manage the service and its documentation. A modest scope can protect that person’s time. It may also reveal that improving the documents and their organisation delivers much of the desired benefit without AI. That is a legitimate outcome of an adoption assessment. The aim is a better-running business, and the appropriate tool should follow the problem. An assistant becomes attractive when its contribution to finding and explaining relevant material exceeds the additional work needed to keep it dependable.
Questions can improve the organisation’s knowledge
One feature of the Definity announcement deserves particular attention: the company intends to use insights from broker questions to improve content, training and policy documentation. This is a proposed feedback process, not evidence that those improvements have occurred. Its business logic is nevertheless clear. Repeated questions could reveal an instruction that is difficult to find, a distinction that needs a better explanation or a process that creates unnecessary uncertainty. Analysis would need care, because frequent questions do not automatically identify the most serious problem, and inquiry records could contain information requiring restricted access.
Used thoughtfully, that feedback could change the relationship between a knowledge system and the organisation maintaining it. The assistant would be more than a faster route through a static manual: it could help staff identify what the manual fails to explain. Human owners would still need to investigate the pattern and decide what to change. The strongest business interpretation of the news is therefore conditional. Definity has announced a way to place AI assistance closer to brokers’ work; the operational gains remain to be demonstrated. For other organisations, the useful lesson is to connect convenience with disciplined knowledge ownership and evaluation, so that an easier answer also becomes a better basis for action.
Image credit: Property, keys and financial paperwork by Jakub Żerdzicki, used under the Unsplash License. Illustrative photograph unrelated to Definity, Guidewire or their software. Resized for this site; WordPress may crop display versions.
