From 100 models to one outcome: Bloomreach's agentic commerce playbook

Key points

Key takeaways from theCUBE interview with Bloomreach CEO and co-founder Raj De Datta, recorded at the BofA Private Tech Trailblazers event in Palo Alto (October 2026):

The settling phase of AI. De Datta argues AI is leaving its "amusement park" era of demos for a settling phase in which builders understand what models can and cannot do — the guard rails, risks and opportunities — and says great companies get built precisely in such phases.

Commoditised models, vertical value. Bloomreach runs around 100 models — proprietary, foundation and open-source, swapped in and out — and sees the model layer as increasingly commoditised, with value migrating to vertical domains such as e-commerce and marketing.

Proprietary data, vertical agents. The company trains models against outcomes on a stack of roughly 7 billion human profiles, feeding outcome data back so the models improve recursively; De Datta predicts the future of applications is vertical-specific agents trained on proprietary models and proprietary data.

AI momentum without hiring. Revenue is accelerating with AI while headcount stays flat, lifting operating margins; almost half of customers already use at least one AI agent or model, and the share using four multiplied 23x in a year.

Agents that act. Beyond consumer agents such as Muse, Bloomreach ships a conversational shopping agent and a marketing agent that builds and optimises campaigns; the value sits in agents taking action, not merely in security and control planes.

Three moats. Harness (governance, security, repeatability, reliability), proprietary vertical AI models trained on proprietary data, and distribution with domain depth are the foundations De Datta says support Bloomreach's agentic position.

Lumi AI and Lumi Connect. The Lumi AI engine pairs 7 billion profiles and billions of products with machine-learning models coupled to LLMs, performing roughly 5-10x better than out-of-the-box LLMs; Lumi Connect opens the intelligence layer to third-party agents, whose calls to the platform are growing 83% month on month.

Outcomes, not software. The TAM has expanded from selling software to solving outcomes; Bloomreach is free cash flow positive and does not need capital but will tap markets for step-function opportunities, keeps M&A core to product strategy (five deals to date) and partners with Databricks, Snowflake, Google and Shopify.

Read more: theCUBE — BofA Private Tech Trailblazers