A renewal file for motor vehicle insurance lands on a desk with the familiar signs of complexity: two policy schedules, an amended driver profile, and a jump in premium that needs explaining. The changes are not hidden – new items appear, something falls away, and a driver swap introduces new risk. What is less obvious is what those changes mean in practice: how excesses apply, whether tracking conditions still hold, and whether a “routine” update has created a disclosure problem.
This is the kind of work OneSure Direct Brokers says it is trying to structure and speed up with its InsureBot.
Tim Louw, the owner of OneSure, describes InsureBot as an AI-assisted insurance and brokerage workflow platform that helps clients and its brokerage staff understand, compare and process South African short-term insurance information more efficiently.
InsureBot began as a client-facing assistant but expanded into internal brokerage work. It now supports internal workflows, including renewal and comparison work, product-information validation, and operational administration support.
There is also a public-facing dimension. InsureBot is a tool intended to simplify insurance concepts and answer common questions through OneSure’s channels. However, the public-facing version is only one interface into a broader internal platform.
“The public InsureBot app was published in 2024, but the underlying OpenAI integration and brokerage use case had been in development and operationalised inside our brokerage environment for a considerable period before that,” Louw told Moonstone,
Louw frames OneSure Direct Brokers as a medium-sized, full-service brokerage with a technology focus.
The challenge, he says, is operational complexity created by the diversity of product providers and the way insurance information is presented.
“Insurers, underwriters, UMAs, and administrators, each offer different products, wording, benefits, excesses, underwriting rules, exclusions, pricing methods, and document formats,” he said. “For a broker, this becomes even more challenging when there is staff turnover or when new administrators and consultants need to learn multiple insurer products quickly.”
InsureBot started as internal infrastructure.
“The first stage was our InsureBot CRM, which began as a quote and workflow platform,” he said. OneSure integrated with insurer systems where possible and used the platform to compare data, quote faster, and assess product and pricing fit. “The AI layer was added later,” he said, initially using earlier language models before current systems became as powerful as they are today.
InsureBot is not intended to operate as a free-running generative chatbot.
“It is a hybrid model,” Louw said. “The output is dynamically generated by AI, but it is grounded in insurer/product information, policy wording, structured quote data, CRM data, and curated information that we maintain as a brokerage.”
He said the AI layer is used to interpret, summarise, and compare information drawn from available source material. It assists with reading and comparison but does not replace source documents, policy wording, quote schedules, or the authorised brokerage process.
Integration, and the reality of multiple insurers
InsureBot’s integration is mixed and depends on what product providers make available, Louw said.
“We have API integrations with a handful of insurers, underwriters, UMAs, and administrators,” he said, “depending on what each product provider can make available.”
But not all insurers provide the level of API capability needed to support a fully automated end-to-end process. “In those cases, certain information still needs to be captured or checked manually in the CRM,” he said, calling it “part of the reality of operating across multiple insurers in the South African market.”
The purpose of the CRM and AI layer, he said, is to standardise and interpret information across different insurer formats even where data is incomplete or arrives in different forms. He would therefore describe InsureBot primarily as an information provider that simplifies and packages data, a comparison facilitator and workflow-support layer within the brokerage.
He drew a firm line around what the platform does not do: “Final quotation, acceptance, underwriting, disclosure, advice, and policy confirmation remain subject to the insurer and the authorised intermediary process guidelines.”
Premium alone is not enough
Louw said comparisons in short-term insurance are rarely a pure pricing exercise.
“When InsureBot compares products, quotes or policy documents, the comparison is not only about premium,” he said. “Premium is important, but in insurance it can be misleading if looked at in isolation.”
He said InsureBot’s comparisons can incorporate the variables that often determine how cover responds in practice: sums insured and limits, excess structures, exclusions and special conditions, wording differences, rewards and value-added benefits, tracking or other security requirements, and changes in the risk profile (including driver changes).
Premium movement forms part of the picture, he said, but is assessed alongside what changed in the schedule and what may require human review before advice is given.
In practical terms, he said, the platform is most useful at renewal or when a client considers moving to another insurer. In those circumstances, it can compare existing policy information with a renewal, replacement quote, or alternative insurer quote and highlight differences in cover, excesses, exclusions, sums insured, vehicle or driver changes, security requirements, and important conditions.
“This is one of the most useful internal applications because small wording or schedule differences can have a major impact on cover,” he said.
Keeping information current, and keeping the process governed
Louw is clear about a constraint that sits behind most “AI assistant” claims: the system is only as reliable as the information it has been given, and insurance products change.
Because OneSure operates as a broker, he said it receives constant insurer communications – wording updates, product updates, benefit changes, and underwriting changes – as part of its normal business. Those updates are incorporated into OneSure’s internal database and workflows.
“AI is only as useful as the information it is given,” he said. Products change regularly, so the underlying data and wording need to be maintained continuously. He described this as an ongoing operational process rather than a once-off build.
He also emphasised human oversight. Where the system generates a review or summary inside the CRM, he said, it is reviewed by an administrator or consultant before being used in a client interaction. The intention, he said, is to reduce human error and improve consistency, not to create “an uncontrolled decision-maker layer”.
Compressing document work into a structured review
Louw says OneSure has seen operational improvements, particularly in daily policy and quote comparisons.
In a recent internal example, he said InsureBot compared an existing policy schedule with an amended policy schedule and produced a structured summary in under two minutes, highlighting changes and flagging items for review. He contrasted this with manual comparison work: an experienced administrator reading two policy documents, identifying differences, and preparing a written summary “could easily take 30 minutes to an hour”, depending on complexity. The AI-assisted comparison, he said, reduces that time and gives the administrator a structured review template to validate.
He also supplied technical and cost detail for that example: the system processed about 140 908 tokens and produced the comparison in under two minutes. Based on public API pricing, he estimated the model-processing cost at a few rand (about R2.50), depending on the input/output split. For Louw, the larger value is not the API cost but reduced administration time, improved consistency, and reduced risk of missing an important policy change.
He said this matters beyond efficiency. From a compliance and record-of-advice perspective, he said, a structured comparison is valuable because it helps to show what changes are being made, what was considered, and what critical information should be brought to the client’s attention.
Louw is explicit that human review remains essential and that there are circumstances where the system is not suitable.
InsureBot is not suitable where source information is incomplete, outdated or ambiguous, or where complex judgement calls are required without human validation. Policy wording, schedules, disclosures, and insurer confirmation remain critical, he said.
AI can assist in reading and comparing information, but “it must operate within a governed brokerage process with continued human accountability”.
Client response: a ‘first layer of understanding’
On how clients respond, Louw said the strongest effect comes when AI is used to simplify insurance rather than produce long explanations.
“Most clients do not want to read a long policy wording document before asking a basic question,” he said. “They want to understand what is covered, what is excluded, what the excess means, whether a benefit applies to them, and whether one product differs materially from another.”
“Our public-facing InsureBot helps with that first layer of understanding,” he said. Internally, he said, the larger value has been helping staff deliver faster, clearer, and more consistent technical responses.
AI as an interface layer
In Louw’s view, AI is becoming “an interface layer between the client, broker, insurer, underwriting systems, product information, compliance, and servicing workflows.
For brokerages specifically, he believes well-governed assistants will increasingly support comparisons, policy interpretation, renewal reviews, document analysis, servicing workflows, and record-of-advice preparation.
That brings us back to the renewal file at the beginning of this article. The challenge is not simply noticing that something changed. It is interpreting what the change means in practice – and ensuring that what needs to be disclosed, confirmed, and explained is not missed. OneSure’s view is that an AI-assisted workflow layer can reduce time spent on mechanical comparison work, while leaving accountability with trained staff and the insurer processes that ultimately determine cover.
