Why Does the Fastest-Growing BI Company Care About AI?
Every BI company can ship a chat box. Jamie Davidson, co-founder of Omni, is focused on the layer that makes AI analytics worth trusting in the first place: the semantic context defining what "revenue," "pipeline," "user," and a Salesforce date field all mean. In this conversation with Bryan Bischof, Jamie gets specific about semantic layers as AI infrastructure, deep research on changing metrics, evals for model regressions, and BI moving beyond dashboards into other interfaces.
In our conversation, we discuss:
- Why AI for BI starts with friction removal across technical, semi-technical, and non-technical workflows
- Why "how many users do we have?" becomes four different questions when business context is missing
- How deep research becomes a while loop for interrogating a KPI until the pattern shows up
- In the alpha zone, how Omni turns curated topics into a JSON semantic query spec the LLM can generate
- How chat corrections become proposed semantic model edits, like choosing AE lead over BDR lead for pipeline questions
- Why roughly 500 Salesforce date fields become the real analytics disambiguation problem
- Why golden query sets and eval tooling matter when model upgrades shift answer quality
- How Model Context Protocol and APIs become the integration surface for chat experiences that need structured data
Three takeaways from this conversation:
1. AI gives analysts a while loop. Jamie describes deep research as a way to keep interrogating a KPI: what changed, by country, by product line, by salesperson, by channel, until the shape of the move shows up. The human keeps the business judgment; the agent supplies the repetitions.
2. Jamie is a semantic-layer maxi because the model is smart and blind. “Users” can mean signups, active product usage, purchasers, or something tied to a timeframe. The durable AI work is encoding that business context — metrics, labels, roles, drill fields, dates — so natural language has something trustworthy to stand on.
3. The winning BI agent probably won’t be just a chat box. Jamie has a hypothesis that data work still needs product shape: field pickers, date ranges, cross-filtering, drilldowns, and structured query specs behind the scenes. Chat can be the front door, but the real product is an agent that knows when to generate UI, call APIs, and propose semantic model updates humans can approve.
