Brokers Risk Falling Behind on AI Without Operational Excellence, Warns Clear Group CCO

"AI isn't a free ticket": Clear Group leader urges brokers to fix operations first

At a Chartered Insurance Institute conference on Innovation and Impact, Phil Williams, Clear Group's group chief commercial officer and managing director of retail, warned brokers that AI value depends on operational discipline. "AI isn't a free ticket." He urged firms to do the hard yards before expecting meaningful outcomes from new tools. Chartered Insurance Institute

Williams said insurers are outpacing brokers on AI. He found "about three and a half times more global use cases for insurer implementations of AI than for broker implementations." Insurers are already ingesting client presentations and pushing structured insights to underwriting teams. "You don't hear those conversations being led out by brokers in the same way."

Why insurers are ahead

Insurers tend to have standardised workflows, which makes it easier to embed AI into discrete tasks. That structure also supports offshoring and centralised decisioning. As Williams put it: "You can't offshore thousands of back office functions unless you've standardised your workflows."

Brokers, by contrast, have leaned on market platforms like Acturis to meet technology needs. Williams' view: many brokers have effectively outsourced innovation and stopped treating it as an order winner. The result is slower learning loops and fewer production-grade use cases.

The broker bottleneck

• Fragmented processes and local workarounds
• Inconsistent data capture across placement, service, and claims
• Vendor dependency without internal process ownership
• Limited instrumentation to prove value (cycle time, hit rate, leakage)

A practical AI playbook for brokers

• Map and standardise. Document the critical paths: new business, renewals, endorsements, and claims triage. Remove variations that don't add client value.
• Fix the data. Set required fields, define submission and broking pack standards, and clean client/insurer reference data.
• Instrument results. Track quoted-to-bound, time to quote, remarket rates, loss ratio impacts, and rekeying effort saved.
• Target narrow use cases. Start with repeatable, text-heavy work: summarising presentations, extracting submission data, QA checks, MI preparation.
• Embed in the workflow. Trigger AI from the systems staff already use; return outputs into the same systems. No side-channel chaos.
• Add controls. Keep audit logs, version prompts, restrict sensitive data, and run human-in-the-loop reviews for material decisions.
• Develop owners, not tourists. Name process owners, AI champions, and success metrics for each pilot before you start.
• Vendor strategy. If you rely on a platform, push for process and data standards first-then co-develop targeted use cases.

Prompt engineering isn't the strategy

"The idea that we can train our entire workforces to be prompt engineers is a step too far," Williams said. Broad-brush training won't fix messy processes or poor data. Focus on role-based workflows, curated prompts, and small sets of approved assistants tied to defined tasks. Train for outcomes, not for chat tricks.

What smaller brokers can do now

• Use affordable, off-the-shelf tools for email triage, meeting notes to CRM, and document summaries.
• Automate checks: find missing information in submissions, highlight coverage gaps, and compare terms and conditions.
• Keep a human in the loop for client-facing outputs and binding decisions.
• Measure time saved and error reduction; reinvest into client service.

Williams' message: experiment, but stay client-first. "Be a good broker first and foremost, then think how AI might help some of that."

Readiness check

A live poll underscored industry uncertainty. Asked whether broking is set up to develop AI use cases, 68% said "somewhat, but there are challenges," and 18% said "no, the industry is not ready." The gap is clear-and fixable with operational excellence.

Next steps

• Pick one journey (e.g., renewals). Standardise it, clean the data, and run a four-week pilot with clear metrics.
• Scale only when you've proven impact and controls. Don't skip governance.
• If you need structured learning by job function, see curated options at Complete AI Training.