Gil Nechushtai
Chief Product Officer at Five Sigma
Gil is an experienced insurance technology leader with over two decades of success in crafting InsurTech solutions. His expertise spans developing AI platforms and streamlining operational efficiency.
6 MINUTE READ
Across industries, leaders have bet big on GenAI: $30–40 billion by recent estimates. Yet, according to an MIT’s Project NANDA study, only about five out of every hundred projects – a mere 5% – created real business improvement. Most projects never made it past a pilot, or went live with no measurable lift in KPIs (Key Performance Indicators). That’s 95% of GenAI projects that end up in failure!
In insurance claims, I’ve seen where GenAI initiatives break. I’ve implemented AI in enterprise operations for years before Five Sigma. I’ve seen pilots stall after the demo, and I’ve seen AI move KPIs. Below, I’ll map five failure patterns highlighted by the MIT study. I’ll also share – for the benefit of insurance leaders out there – how to avoid such failures, and how we engineered Five Sigma’s Clive™, the industry’s first Multi-Agent AI Claims Expert, so that our AI implementation projects end in success. We’re at the Top 5% of AI projects for a reason.
First, a brief overview of Clive and how he operates at a high level. Clive is purpose-built to transform how insurers handle every step of the claims process. Instead of a monolithic system, Clive delivers a portfolio of specialized AI agents — each one focused on a specific claims function. Each agent is trained on vast claims knowledge and expertise specific for his task, so he can automate it to the highest point of accuracy.
Turning AI Adoption into Claims Results
Companies roll out horizontal tools (like ChatGPT, Copilot) that individuals love, but they rarely affect operations in its entirety. Adoption is high, yet structural change is minimal across industries. In enterprises, generic tools boost personal productivity but don’t change operations for the better — like shorter cycle times, lower leakage and LAE, fewer handoffs, and better customer satisfaction at scale.
At Five Sigma, Clive isn’t just a chatbot who answers questions (although he does that as well); he operates multifunctional agents, each executing every step across the entire claim lifecycle, according to the claim’s state and the insurer’s rules. That means triage, coverage opening, damage assessment, next-step planning, fraud detection, compliance and document processing are automated and accurate.
The fact that AI solutions are purpose-built for a specific domain, and not generic, makes a significant difference in accuracy and efficiency.
Workflow-Integrated GenAI for Claims
The MIT study shows a steep cliff: many teams evaluate AI, some reach a pilot, and only about 5% of custom AI tools make it into production. Enterprises lead in pilot volume yet lag in pilot-to-scale conversion. Mid-market teams implement faster because they align solutions with real workflows sooner.
Clive was developed by our insurance specialists, designed specifically for insurance claims workflows. He operates on top of existing claims management systems (CMS) or within our cloud SaaS CMS. He uses secure APIs to integrate with the insurer’s third-party systems, including policy admin, payments, fraud services, and FNOL.
This means that no organizational redesign is needed to “make the AI fit.” Teams use Clive on day one because he aligns with how they already work.
Asking users to change their ways of work requires adaptation, which could be hard. However, if you plug AI into existing practices and provide value – it’s easier for people to understand it and use it almost seamlessly.
The GenAI Learning Gap in Claims
Learning is the missing link, the study says. Many GenAI deployments don’t retain feedback, adapt to context, or improve over time. Users enjoy chat interfaces for quick tasks, but for high-stakes work they choose humans 90% of the time because current AI tools lack context, memory, adaptability, and persistent improvement.
Clive is different. He is pre-trained on claims, infused with adjuster logic, years of industry data, best practice workflows, and compliance requirements. At the same time, Clive maintains context for each claim and adapts as new information arrives – adjusting triage, rechecking coverage, recalculating exposures, and replanning next steps. Clive remembers everything about the claim, better than any adjuster.
Software that retains both domain knowledge and exact context consistently outperforms generic tools. It saves time and money, reduces work, and drives both user and customer satisfaction.
Claims-Focused AI that Delivers Results
Claims are a regulated, data-dense workflow spanning policies, jurisdictions, vendors, and payments. Impact comes from AI that understands the logic of these structures and executes within them. Generic do-it-all AI tools have amazing appeal but cannot generate optimal results for any specific industry or use case. The MIT study finds impact when tools are built for a specific process and wired into existing workflows.
Clive is built specifically for handling insurance claims, nothing more. He is an expert on claims and orchestrates purpose-built agents, each trained for a specific claims handling job.
A triage agent scores severity and routes to the adequate agent or adjuster; an FNOL agent collects and structures first-notice details and required documents to setup a claim in the system; a damage-assessment agent reviews evidence to recommend inspections, repairs, or reserves, etc. Clive coordinates them end-to-end so the right claim specialist agent handles each claim stage. Every action is logged and explainable inside the claim file.
Build vs. Buy for Claims AI: Why Partnerships Win
The study highlights an implementation advantage: partnering with a specialized AI company beats building in-house. Partnerships are about twice as successful as internal builds and tend to deliver integrated, learning tools faster.
Clive is live in production, and proven to provide results. He integrates with any claims system, and starts doing real work and delivering results on day one. Five Sigma provides end-to-end support for every module integration, including calibration, accuracy testing, and maintenance, so organizations get all the benefits with none of the upkeep hassle.
The result: higher productivity, lower LAE, and no implementation risk, making Clive a cost-effective choice with immediate, measurable ROI.
A Practical Checklist for Your Next AI Project
Moving forward with your AI initiatives, here are five questions (inspired by the MIT study’s patterns and our experience) to predict AI implementation success:
- Will it go into production? Is there an integration plan signed off by IT, Security, and Claims Operations?
- Will it improve a process that matters? Is there a defined KPI (cycle time, FNOL-to-contact, time to first payment, leakage) and a baseline to benchmark?
- Will it improve over time? Where does feedback come from, how is it stored, and how does it change behavior and outcomes going forward?
- Will people actually use it? Is it embedded in the current workflows, and is the user experience simple?
- Who owns outcomes? Is there a named executive in the organization who is responsible for the measurable results, far beyond rechecking as a “pilot sponsor.”
Why Clive AI Works
At Five Sigma, we built Clive to say “yes” to all five:
- Production-ready: Runs with existing stacks or on Five Sigma’s platform, and live in production at many claims operations of Five Sigma customers.
- Improves any claims stage you need: Through multi-agent orchestration Clive coordinates function-specific expert agents to improve important workflows, such as triage, FNOL intake, damage assessment, communications, compliance.
- Continuous Improvement: Feedback loops from outcomes, remembering claim context, adapting upon developments, learning from experience.
- Adoption-friendly: Works inside the adjuster’s daily flow and existing systems.
- Claims leaders own results: Clive delivers results on the measurements that matter to claim leaders – productivity, LAE reduction, leakage prevention, and time to settlement.
Check out Clive’s measured outcomes!
