During a lunchtime session at the 2025 Impact conference in Boston last month, five leaders from organizations that won Impact Awards shared how they are using artificial intelligence and other new technologies to produce tangible business outcomes. Here are the initiatives they launched; the technologies they used; and the approaches that make them successful. And at the end, you’ll find a list of common threads among this year’s cohort of award winners.
Wayfair
About the initiative: The catalog science team at online retailer Wayfair, represented by Jeff Arena, used generative AI to improve the accuracy and completeness of product data across its massive online catalog. The project began by focusing on key attributes like color and material — the details most likely to influence shoppers — and expanded from there. In its first phase, the effort improved conversion rates by two percent, translating into more than $20 million in additional revenue.
Underling technologies: The team used Google’s Gemini models to generate predictions for product attributes and compare them against supplier-provided data, later incorporating OpenAI models to generate multimodal attribute definitions used for prompting. Human-in-the-loop evaluation ensured accuracy and refinement, with ongoing work to shift more evaluation to AI judge models for scalability.
What made it successful:
- They began with high-impact attributes like color and material — areas where better data immediately improved customer experience and sales.
- A rigorous, human-in-the-loop process ensured quality early on before scaling automation across the catalog.
- Cross-team collaboration between product, merchandising, and data science refined definitions and boosted model precision.
- Wayfair balanced performance and cost, selecting Gemini for scalability while experimenting with OpenAI for specific use cases.
- Arena emphasized that keeping the ROI clear — and showing measurable gains — built momentum for further expansion.
Nationwide Insurance
About the initiative: Terrell Bretz, an innovation product manager at Nationwide, led development of Resolution Assist — a platform to streamline how the Columbus, Ohio-based insurance firm’s complaints department processes customer feedback. The tool summarizes complaint text, recommends reason codes from a set of roughly 450, and drafts empathetic responses for human review. It has helped a small team of analysts handle thousands of complaints more efficiently, as well as reducing complaint categorization errors.
Underling technologies: Built in‑house and integrated with Nationwide’s existing Emplifi system, the application uses natural language processing to summarize and categorize complaints, sentiment analysis to gauge tone, and generative AI to draft suggested responses. The design keeps a human in the loop so associates retain discretion.
What made it successful:
- The team invested heavily in early discovery, spending weeks interviewing complaint analysts to identify pain points and workflow gaps.
- Starting with a simple prototype for five users enabled rapid iteration and built trust through visible improvements.
- Integrating directly into the existing Emplifi platform minimized disruption and encouraged adoption.
- Maintaining a human‑in‑the‑loop approach reduced fear of job loss and reinforced that AI was a tool, not a replacement.
- Regular communication and training fostered confidence — and ultimately a 70% reduction in categorization errors.
American Arbitration Association
About the initiative: The American Arbitration Association is a nearly century-old nonprofit that provides arbitration and mediation services to help businesses and individuals resolve disputes outside of court. Vice President of Innovation Linda Beyea described a project that uses AI‑powered semantic search to help case managers quickly identify the right arbitrators for disputes. By analyzing more than 5,000 resumes and interpreting language contextually, the tool significantly reduced the time required to prepare arbitrator lists. The idea originated from an employee suggestion and became a model for innovation within the century‑old nonprofit.
Underling technologies: The AAA’s tech team, working in a Microsoft environment, used OpenAI models and semantic chunking to create an intelligent search tool. It links related concepts automatically (e.g., “football stadium” and “sports facility”) and includes practical filters for geography and rates. After iterative testing with case managers, it was rolled out broadly.
What made it successful:
- The project began as an employee‑generated idea, supported by AAA’s structured innovation pipeline.
- Semantic AI allowed case managers to search more naturally, saving 40–55% of the time previously required for list creation.
- Iterative pilots with initially 10, and then 40, case managers ensured the tool fit real‑world needs before scaling nationwide.
- Strong executive sponsorship and a culture requiring innovation training hours gave employees confidence to adopt new tools.
- Measuring and communicating time savings helped demonstrate clear productivity gains to both staff and leadership.
EllisDon
About the initiative: EllisDon is a global construction and engineering services company based in Toronto, Canada, known for integrating technology and innovation into large-scale building and infrastructure projects. Chief Information Officer Brandon Milner outlined two AI‑driven projects recognized with the Impact Award — a digital twin program for the city of Burnaby, British Columbia, and an AI‑based site‑selection tool to accelerate affordable housing development. Both initiatives aimed to replace slow, manual processes with data‑driven analysis and automation, showcasing how AI can modernize decision‑making in construction.
Underling technologies: EllisDon used Unity as the base modeling platform for the digital twin, paired with Microsoft Azure AI for predictive analytics and automation. For the affordable housing work, the team built an AI visualization platform that integrates semantic input, public data APIs, and financial modeling to produce feasibility studies and site recommendations in minutes.
What made it successful:
- Starting with a small proof‑of‑concept in Burnaby demonstrated the potential for digital twins at a city scale.
- Milner’s team built credibility by delivering quick, practical wins — like automating invoice processing, which saved tens of thousands of hours — before proposing larger AI initiatives.
- Stakeholder engagement was critical; the team spent months educating non‑technical decision‑makers about AI’s benefits.
- EllisDon fostered adoption through bottom‑up innovation, empowering employees to lead pilot projects and share results upward.
- By focusing on time and cost savings, the projects turned skeptics into supporters across a traditionally conservative industry.
PACCAR
About the initiative: Rahul Deshpande discussed how the 100‑year‑old truck manufacturer developed a machine‑learning model for predictive maintenance in heavy‑duty trucks. The system detects early signs of engine issues such as injector fouling, helping reduce costly breakdowns and improve fleet uptime — a major factor in trucking profitability.
Underling technologies: The project used machine‑learning models trained on extensive sensor data from truck engines, gathered via accelerated test environments. The on‑truck model processes signals locally (e.g., turbo speed, rail line pressure, valve timing) to avoid latency and connectivity issues, and AI techniques were used to select the most predictive signals.
What made it successful:
- The team chose a clear, high‑value problem — predicting injector fouling — that directly affected customers’ uptime and cost.
- Accelerated testing using doped fuel allowed them to gather meaningful data without waiting hundreds of thousands of road miles.
- Close collaboration between AI scientists and mechanical engineers ensured that the model outputs aligned with physical reality.
- Executive support gave the small Silicon Valley‑based team access to resources and visibility within the company.
- Deshpande emphasized that innovation was “a team sport” — success depended on the right mix of data, domain knowledge, and cross‑functional trust.
Common Threads Among the Winners
Across all five initiatives, several common themes emerged that explain why these projects succeeded.
First, each organization emphasized deep collaboration and cross‑functional teamwork. Whether it was Wayfair’s catalog scientists partnering with merchandisers, Nationwide co‑designing tools with complaint analysts, or PACCAR engineers working side-by-side with data scientists, success depended on combining technical skill with on‑the‑ground experience.
Second, every team embraced iterative experimentation and human‑in‑the‑loop design. Rather than launching grand programs, they started small, ran pilots, and used user feedback to refine and expand. That approach built both trust and measurable impact.
Finally, strong executive sponsorship and supportive cultures made experimentation possible. Leadership empowered teams to test new ideas, linked them to clear ROI, and celebrated tangible business results — creating an environment where innovation could thrive at scale.
You must be logged in to post a comment.