Asaf Bord leads generative AI strategy and research at Northwestern Mutual and co-founded Multinear, an open-source platform for building and evaluating reliable AI applications. He develops enterprise AI systems that turn institutional data into useful answers while remaining accountable to governance requirements, business outcomes, and the people who must trust them.
Bord studied at the Technion–Israel Institute of Technology, with interests including computer vision and medical imaging, before moving into software and product leadership. At Earnix, where he was a senior product manager from 2014 to 2016, he worked on enterprise software for insurance and banking. He subsequently focused on structured data, catalog quality, and shopping experiences at eBay, held product leadership responsibilities at WeWork, and joined Northwestern Mutual. His work there spans business-intelligence copilots, semantic data infrastructure, evaluation, and AI agents for understanding legacy software.
With co-founder Dima Kuchin, Bord built Multinear around practical pass-or-fail evaluations, experiment comparisons, regression detection, and measurable business results. Together, they used the platform for the Strawberry Test, examining how language models handle letter-counting problems and what their performance reveals about reasoning, token consumption, speed, and cost.
Building enterprise AI that earns trust
- Certified reports before autonomous SQL. Bord’s enterprise GenBI architecture starts with trusted dashboards, verified reports, and established business definitions. A metadata agent interprets each question, retrieval finds certified material, and SQL generation enters only when existing assets cannot answer it. Governance and orchestration surround the workflow.
- Practitioner-led evaluation and adoption. Business-intelligence specialists help construct realistic evaluations, identify mistakes, and test systems against messy organizational data while sensitive client information remains outside experimental environments. These specialists become the first users; business managers follow as accuracy and trust improve. Bord connects this staged GenBI rollout to guardrails, semantic context, and measurable returns.
- Evaluation as production infrastructure. Bord’s work on automated RAG testing combines generated question-answer tests, human-refined scoring criteria, retrieval assessment, and repeated regression checks. Multinear applies related discipline to prompts, models, data, and application behavior.
- Incremental enterprise AI delivery. Bord breaks ambitious programs into independently valuable stages, including better metadata, searchable data ownership, certified-report retrieval, and interactive analytics. Each stage gives leadership concrete results and an opportunity to redirect investment or assess outside products against benchmarks grounded in its own data.
