Liberty Mutual’s AI strategy, as described by Global CIO Monica Caldas, is designed to change how work gets done—not simply to automate tasks or reduce headcount. The company is pairing a secure internal generative-AI platform with employee training, responsible-AI oversight and workflow redesign. Its clearest example: AI handles about 80% of an internal help-desk process, freeing staff to work on more complex problems. The interview does not disclose whether that figure means tickets, labor hours or another measure, or quantify the resulting savings.
What Liberty Mutual means by “unlock human potential”
In a February 25, 2026 interview with CIO, Caldas describes AI as a way to reduce repetitive work and make more employee time available for judgment, problem-solving and other higher-value tasks. That framing is more specific than a general promise to make work faster: it implies redesigning the mix of work people do.
The internal help-desk example illustrates the idea. AI handles about 80% of the process, according to Caldas, while employees can focus on harder issues. The interview does not define the denominator behind “80%,” nor report before-and-after figures for resolution time, error rates, cost, employee experience or customer outcomes. It is evidence of a workflow change, not a complete measure of its effect.
Liberty Mutual’s AI effort builds on earlier analytics work
The company was using predictive AI before the recent generative-AI wave. Caldas cites underwriting tools and claims forecasting as earlier applications. That history matters: Liberty Mutual’s generative-AI program is presented as an extension of a longer effort involving data and decision integrity, rather than a start from scratch after public chatbots became prominent in late 2022.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11#1 Best Overall
Why the company started with an internal platform
LibertyGPT is Liberty Mutual’s internal generative-AI platform, introduced to give employees a controlled place to learn and experiment. In an insurer, that boundary matters because work can involve sensitive customer, employee and business information. A company-approved environment can support access controls, data-handling rules and oversight that unsanctioned public tools may not provide.
Calling a platform secure does not make it risk-free. Internal systems can still produce inaccurate outputs, expose information through poor configuration or use, reflect bias, or lead employees to make flawed decisions. The interview does not describe LibertyGPT’s technical architecture, model providers, logging, retention rules or specific security controls, so it should not be assumed to be a particular kind of retrieval system or application suite.
A separate company-analysis page says LibertyGPT uses OpenAI models and was available to roughly 45,000 employees; those details are not stated in the primary interview and should be treated as secondary claims, not established architecture or reach figures: AI Trace’s Liberty Mutual profile.
Training and experimentation are part of the operating model
Caldas compares baseline generative-AI training to getting a driver’s license: employees need preparation before wider experimentation. The interview also describes a generative-AI hub, the AI@Liberty employee community, an executive program called Executech, a responsible-AI steering committee and a change-champion network. Together, these elements pair guidance from the center with peer learning and local support.
The Digital Progression Framework gives employees a structured way to learn about AI, experiment and advance ideas responsibly. Its formal stages, approval gates and scoring rules are not published in the interview. The framework is best understood from the available description as an enabling structure, not as a documented checklist with known thresholds.
For a similar program to work in practice, training needs to extend beyond tool demonstrations. Employees need to know which information they may enter, how to break a task into useful steps, how to verify outputs, when human judgment is required and where to escalate a problem. Liberty Mutual’s interview does not report curriculum length, completion rates or assessment results.
How Liberty Mutual aims to scale beyond pilots
Caldas says the company prioritizes business outcomes and strategic alignment, defines input and output criteria, seeks sponsorship and applies governance. It favors a platform-oriented approach over a proliferation of disconnected tools and says it does not want to run numerous competing pilots just to generate headlines.
She reports about 50 AI use cases in production and scaling, and more than 30 AI capabilities identified for potential build-or-buy decisions. These are executive-reported figures. The interview does not define where production ends and scaling begins, name the use cases or capabilities, state how many pilots did not advance, or give a conversion rate or time-to-scale measure.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The implied sequence is useful for other organizations: connect an idea to a business goal, choose a workflow, define inputs and outputs, assign a sponsor, apply governance, measure it in operation, and decide whether to make the capability repeatable. “Build or buy” is a decision about solving a real need, not a reason to adopt a fashionable tool. The interview does not name vendors, procurement criteria, evaluation scores or spending.
The technical foundation beneath the AI program
Caldas describes modernization across three connected areas. The details explain why selecting a model is only one part of enterprise AI:
- Core systems: Move toward modular systems that interoperate and can evolve continuously.
- Data: Redesign pipelines, ingestion and enterprise flows so data can move securely and reliably, with quality and governance controls.
- Identity, access and governance: Strengthen controls that determine who or what can access information and how AI use is overseen.
For regulated work, a tool cannot be scaled responsibly on top of unreliable data, brittle integrations, weak authorization or unclear accountability. The interview describes these as modernization priorities, but does not publish a technical blueprint or deployment schedule.
What the insurance context changes
Insurance work can involve sensitive personal information and consequential processes, so the risk depends on what the AI is allowed to do. An internal summarization assistant is different from a system whose output could influence a claim, price, eligibility decision or customer treatment. The interview emphasizes data integrity, decision integrity, security, compliance and human judgment; it does not establish that Liberty Mutual’s generative-AI systems autonomously make underwriting or claims decisions.
Organizations applying this model should match oversight to impact. They should ask whether data is authorized for use, whether outputs can be audited, whether errors are reversible, who reviews consequential recommendations and who is accountable for the final decision. These are practical evaluation questions, not confirmed Liberty Mutual framework requirements.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the reported results do—and do not—show
The 80% help-desk figure is the most concrete operational example in the interview, but it is not enough by itself to establish financial return or improved service. A useful scorecard for an AI-enabled workflow would include:
- Time to resolve work, first-contact resolution and escalation rate.
- Accuracy, rework and the amount of human review still required.
- Employee adoption, repeat use and confidence after training.
- Customer experience where the workflow affects customers.
- Compliance, privacy and security incidents.
- Cost per transaction, including model and infrastructure costs.
- Share of use cases reaching production and benefits actually realized.
These are recommended measures, not results reported by Liberty Mutual. The interview does not quantify financial returns, adoption rates, model error rates, customer impact or governance outcomes.
A practical sequence for other regulated organizations
- Set data and security boundaries. Define permitted information, access rules, oversight and escalation before inviting broad use.
- Give employees a sanctioned place to learn. Pair the platform with guidance on appropriate use and output verification.
- Train leaders and frontline teams. Build practical AI literacy, peer support and local change capacity rather than relying on a launch announcement.
- Choose a workflow with a real owner. Tie selection to a business outcome, establish inputs and outputs, and identify who remains accountable.
- Redesign the work and human review. Specify which steps AI handles, what people check, and how exceptions are routed.
- Measure the workflow, not the demo. Track quality, time, adoption, risk and cost against a meaningful baseline.
- Scale common capabilities carefully. Reuse platforms where that reduces duplication, while allowing specialized needs to justify distinct tools.
- Monitor after deployment. Reassess performance, access, costs and risk as systems and workflows change.
This sequence captures the central lesson of Caldas’s account: unlocking human potential requires more than putting a chatbot in employees’ hands. It depends on secure access, prepared users, modern data and systems, clear accountability, and deliberate decisions about where the capacity created by AI goes.
Quick Recap
Sources
- CIO: interview with Monica Caldas, published February 25, 2026.
- The Heller Report: interview coverage.
- Liberty Mutual Technology Hub.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
