Why Insurance AI Projects Stall at the CRM, and How to Avoid It

Most insurance AI pilots start with the easy part. A voice bot answers the phone, understands "I want to add my son to my car policy," and replies politely. Then it has to do something: check the policy, work out the premium, take the payment and update the record. That's where many projects stop, because the answer lives in a CRM or policy administration system the AI can't reach.

Insurance AI projects stall at the CRM when the contact centre platform, the AI and the policy or broker systems aren't integrated for the specific task. They succeed when integration is scoped per call type first: which system holds the data, which one must be updated, and how identity and payment fit in.

TL;DR

• Conversation is the easy half of insurance AI. Completing the task means reading from and writing to policy, CRM and payment systems.
• A mid-term adjustment touches three systems: identity verification, payment processing and the policy record.
• Scope integration per call type, starting with the calls that need the fewest systems.
• Summaries and wrap-up are a low-risk first step, because they write to the CRM without changing a policy.

Why do insurance AI pilots work in demos and stall in production?

Demos run on sample data and a single system. Production runs on the insurer's real estate: a broker platform, a policy administration system, a CRM, a payment provider, and a contact centre platform, each with its own data model and access rules.

DigitalWell's Insurance Brokers & Providers page puts it bluntly: brokers' back-end systems are notoriously difficult to integrate with. A routine mid-term adjustment, such as changing a car, adding a driver or updating an address, has to verify the caller, take a payment and update the policy record. If any one of those three links is missing, the AI can talk about the change but can't make it.

The five stall points below cover the usual gaps.

Stall point 1: nobody mapped the task to the systems

Projects often start with "automate renewals" rather than "for a renewal query, read the premium from system A and the cover from system B." Without that map, integration scope keeps growing mid-project.

Stall point 2: read access was agreed, write access wasn't

Answering "what's my premium?" needs read access. Changing an address or adding a driver needs write access, which brings security review, audit and testing. Projects that only planned for reading hit a wall at the first real transaction.

Stall point 3: identity sits in a different system

The AI can't release policy details, or change them, until the caller is verified. If verification lives in a separate system from the policy record, the AI needs both connections and a rule for linking them.

Stall point 4: payment wasn't designed in

Mid-term adjustments and renewals often end in a card payment. If payment handling isn't part of the AI flow, the call gets transferred to an agent at the last step, and the automation saves little. Card data also has to stay out of recordings under PCI DSS.

Stall point 5: the contact centre and the CRM don't share context

When the AI hands over to an agent, the agent needs the conversation so far. If the contact centre platform isn't linked to the CRM, the agent starts again and the customer repeats themselves.

How do the stall points map to fixes?

The table pairs each stall point with the symptom and the fix.

Which call types should an insurer automate first?

Start with the calls that touch the fewest systems, then move up. This order keeps early wins simple and builds the integrations the harder calls need.

1. Call summaries and wrap-up. The AI writes a summary to the CRM after a human-handled call. It changes no policy, so risk is low.
2. Read-only queries. Premium, cover, renewal date and claims status. These need read access and verification.
3. Document requests. Policy schedules and certificates, issued on the call.
4. Simple changes. Address or contact detail updates, with write access and audit.
5. Mid-term adjustments with payment. Changing a car or adding a driver, with verification, payment and policy update in one call.

Summaries come first for a reason. Karen Hickey, Enterprise Account Director at DigitalWell, describes the question she asks contact centre teams:

"Is there something that we could do to reduce the amount of time an agent is having to do work after the call?" Karen Hickey, Enterprise Account Director, DigitalWell

The AI can pull out the details that matter for an insurance call, such as policy number or car registration, and log them against the customer, as a by-product of the call.

What does a working integration look like?

A working integration has three layers, each connected to the next.

The voice and network layer carries the call and, with DigitalWell's AI Voice Network, transcribes and summarises it and runs AI Voice Agents. Because it sits in the network, it works with the insurer's existing phone system.

The contact centre platform routes calls and digital contacts, gives agents their desktop and reports on performance. DigitalWell implements Genesys Cloud CX, Amazon Connect, Zoom Contact Center and its own DigitalWell 360.

The business systems hold the data: the broker or policy platform, the CRM and the payment provider. Integration here is where projects are won or lost.

An Post Insurance is a working example of the third layer. It chose DigitalWell partly because DigitalWell had proven integration capability between Applied Connect, the core quoting platform used by Irish brokers, and Genesys Cloud CX. The deployment took An Post Insurance from one customer channel to four and saved €220,000 in technology total cost of ownership over three years. Campion Insurance shows the payment side: DigitalWell built a desktop application that pauses and resumes Genesys call recording through the API whenever a payment screen opens.

DigitalWell 360 adds native CRM integrations with platforms including Salesforce, HubSpot, Dynamics and Zendesk, as listed on the Smart Workplace page.

What should an insurer ask before starting an AI project?

Ask these before the first sprint, and the stall points become scoping decisions instead of surprises.

• Which call types are in scope, in which order?
• For each, which systems are read and which are written?
• Who approves write access, and what audit trail does it need?
• Where does caller verification happen, and how is it linked to the policy?
• How are card payments taken, and how is card data kept out of recordings?
• How does the conversation reach the agent and the CRM when a person takes over?
• Who has already integrated with our broker or policy platform, and can we speak to them?

Frequently Asked Questions

Why do AI projects in insurance take longer than expected?

Often because integration with policy, CRM and payment systems wasn't scoped per call type at the start. The conversation works quickly; completing the task takes the integration work.

Do we need to replace our CRM to use AI?

No. The AI and contact centre platform connect to existing systems through APIs. The work is in the connections, and in agreeing read and write access.

Can AI complete a mid-term adjustment on its own?

Yes, when it's connected to identity verification, payment processing and the policy record. DigitalWell's AI Voice Agents are built to complete all three in one call, with an audit trail of what changed.

What's the lowest-risk place to start?

Call summaries and wrap-up. The AI writes to the CRM after a human-handled call, without changing any policy, and cuts after-call work for agents.

Does the contact centre platform matter for AI integration?

Yes. The platform needs to pass context to agents and the CRM when a call is handed over. Genesys Cloud CX, Amazon Connect, Zoom Contact Center and DigitalWell 360 all support this, with different integration approaches.

Key Takeaways

• Insurance AI stalls where conversation meets the policy, CRM and payment systems.
• Map every call type to the systems it reads and writes before building anything.
• Plan write access, identity and payment from the start, because transactional calls need all three.
• Start with summaries and read-only queries, then move to changes and mid-term adjustments.
• Proven integrations, such as Applied Connect with Genesys Cloud CX at An Post Insurance, shorten the path.

Stuck between an AI pilot and production? Talk to DigitalWell about mapping your call types to the systems they need.