Liberty Mutual launched a quoting app inside ChatGPT in late May. A user opens ChatGPT, connects the Liberty Mutual app, answers a short series of questions in plain conversation, and receives a real, personalised auto insurance price. They are then directed to the Liberty Mutual website to finalise and buy. The whole thing replaces a form that used to take considerably longer.
This matters because of where customers are going. One in three US adults has already used ChatGPT for financial advice. OpenAI only opened the door to insurance apps in February 2026, and the pipeline filled up almost immediately. Spanish insurer Tuio was first, Insurify followed within days, and Liberty Mutual, French broker APRIL, and several others launched within months. The pattern is the same across all of them: meet the customer where they already are, rather than waiting for them to come to a website.
Here is the part that most coverage misses. The ChatGPT app is the visible result of about ten years of back-end preparation. Liberty Mutual has moved 85% of its technology workloads to the cloud and cut its global data centre footprint from 13 facilities down to one primary site. Its core operating systems still run on a large mainframe, and the company is five years into a plan to move those onto Guidewire, the standard platform most large property and casualty insurers use. In the meantime, they built a connection layer between the mainframe and the cloud so the new AI features can access the old data without waiting for the full migration.
The company is also deliberately not locking itself to a single AI provider. It works with AWS as its main cloud, but can shift workloads to Google and Microsoft Azure. On the AI side, it built a technical bridge that lets it work with any model provider, not just OpenAI. That flexibility matters because the AI market is moving fast and costs are rising. Liberty Mutual has a dedicated team that monitors how much AI processing is consumed across the entire company and can cap or increase usage by team or project.
The broader industry picture is not pretty for companies that tried to skip the groundwork. A BCG analysis found that 70% of the problems insurers encounter when scaling AI come down to people, processes, and organisational issues, not the technology itself. Deloitte found that poor data quality and legacy infrastructure were among the top reasons AI projects in insurance fail to reach production. McKinsey research suggests that digital leaders in insurance grow revenue roughly five times faster than peers, yet many are still running on systems that accumulate decades of undocumented rules and siloed data.
For operators in any industry, the takeaway is practical. The companies now getting usable results from AI are not the ones that rushed to deploy chatbots or dashboards. They are the ones that spent years cleaning up their data, migrating off old systems, and building flexible technology that is not hard-wired to one vendor. That work is boring and expensive and largely invisible to customers. But without it, any AI feature sits on a weak foundation: it costs more to run, produces worse answers, and cannot scale.
ChatGPT is becoming a sales and service channel the way comparison websites did in the 2000s. Any company whose products can be quoted, compared, or explained in a conversation will face the same pressure Liberty Mutual is responding to now. The question for most operators is not whether this applies to them. It is how far behind their data and systems are, and how long it will take to close that gap.