Delphina’s AI Agents Now Answer 73% of Questions From Business Teams

The startup, founded by former Uber data leaders, has raised $15 million to turn data warehouses into conversational analytics for non-technical employees.

About Delphina

Published

The most expensive piece of enterprise software is the one that sits unused. It’s a quiet, familiar failure in business intelligence, where million-dollar data warehouses are queried by a handful of specialists while the rest of the company relies on guesswork and stale reports. Delphina, a San Francisco startup, is betting that the fix isn’t a better dashboard, but a tireless, junior data scientist that lives in the data itself.

Founded in 2023 by former Uber data and engineering leads Jeremy Hermann and Duncan Gilchrist, Delphina calls its product a ‘Context Factory’ [Perplexity Sonar Pro Brief, retrieved 2026]. It connects to systems like Snowflake and Databricks, uses AI agents to map a company’s metrics and documents, and then lets anyone ask questions in plain English [Ground Level AI, November 2026]. The initial idea was to automate machine-learning workflows, but the team found a more immediate pain point: business teams couldn’t get answers. Now, the company reports that 73% of questions on its platform come from business users, not data scientists, and that CEOs at more than half of its customers are active users [Yahoo Finance, November 2026].

The wedge: from ML ops to CEO queries

Delphina started with a technical wedge. Hermann, who was previously VP of Engineering at feature-store company Tecton, and Gilchrist, a former Uber data scientist, built AI agents to handle the grunt work of predictive modeling: finding data, cleaning it, training models, and productionizing pipelines [Perplexity Sonar Pro Brief, retrieved 2026]. The vision, as one investor put it, was to be ‘GitHub Copilot for data science’ [Yahoo Finance, November 2026].

But in talking to customers, a clearer problem emerged. The bottleneck to using data wasn’t always building a fancy model; it was answering basic, recurring business questions. ‘What drove the dip in North American subscriptions last week?’ or ‘How does this quarter’s marketing spend compare to the same period last year?’ required a data engineer to write SQL, a process that could take days. Delphina pivoted to focus on this ‘context’ layer,the living understanding of what metrics mean and where they live. Its agents now draft that understanding, fill in gaps as employees ask questions, and route complex changes to human experts for approval [Yahoo Finance, November 2026].

Traction with a non-technical audience

The evidence that this is working comes from an unlikely user: the CEO. At newsletter platform Substack, more than 70% of employees actively use Delphina in a given month, including CEO Chris Best [Ground Level AI, retrieved 2026]. The company’s customer roster now includes LATAM Airlines, Gen Digital, Upstart, Udemy, and Coursera [Yahoo Finance, November 2026]. This shift from a tool for data scientists to a platform for business teams is the core of Delphina’s current pitch.

Customer Reported Use Case
Substack Company-wide analytics; >70% employee adoption [Ground Level AI, November 2026]
Gen Digital Partnership for analytics across Norton, Avast brands [dbt Summit, retrieved 2026]
LATAM Airlines, Upstart, Udemy, Coursera, Medely, BaseCamp Franchising Named enterprise customers [Yahoo Finance, November 2026]

To support this expansion, Delphina has raised a total of $15 million across two seed rounds, both co-led by Radical Ventures and Costanoa Ventures, with participation from AI pioneer Fei-Fei Li and others [Yahoo Finance, November 2026]. The most recent $7.5 million round closed in November 2026 [Citybiz, November 2026].

The team betting on context

The founders’ backgrounds at Uber and Tecton provide a specific kind of credibility. They’ve seen the scale and complexity of data operations at a company that runs on real-time predictions. The team they’ve built includes engineers from companies like Gopuff and Wealthfront, focused on building an LLM-powered toolchain for machine learning [LinkedIn, Patrick Schwarz, retrieved 2026]. This technical depth is a deliberate counter to simpler ‘ask your data’ chatbots; Delphina’s value is in maintaining an accurate, governed map of a company’s data reality.

  • Founder pedigree. Jeremy Hermann (CEO) and Duncan Gilchrist (President) built and scaled data products at Uber and, in Hermann’s case, the MLOps platform Tecton [Bloomberg, December 2023].
  • Technical bench. The engineering team includes veterans from Uber and Gopuff, focused on infrastructure for machine-learning systems [LinkedIn, Patrick Schwarz, retrieved 2026].
  • Commercial push. The company is hiring its first Founding Account Executive, signaling a move to formalize its sales motion beyond founder-led deals [Perplexity Sonar Pro Brief, November 2026].

Where the wheels could come off

The risk for Delphina is that it’s navigating a crowded and noisy market. Its differentiation,the persistent, learning ‘Context Factory’,is sophisticated but harder to explain than a simple chatbot. Furthermore, its pricing model (custom seat- or consumption-based) and enterprise features like SOC 2 compliance and VPC deployment suggest it’s going after large, complex organizations [Delphina.ai, retrieved 2026]. Selling into those accounts is a different game than winning over a tech-savvy team at Substack.

The company’s answer likely hinges on governance and trust. By routing important changes to domain experts and building on existing data infrastructure, it aims to be the reliable layer that IT approves and business teams actually use. The high CEO adoption rate it cites is a powerful signal that it can cross the chasm from data team to executive suite [Yahoo Finance, November 2026].

The unit economics of unanswered questions

The real cost of unused data software isn’t the license fee. It’s the collective hours lost to meetings debating which number is correct, or to waiting for a data ticket to be prioritized. If a data engineer spends two hours a day answering basic, repetitive queries for the sales team, that’s roughly 500 hours a year. At a fully loaded cost of, say, $150 an hour, that’s $75,000 in engineering time spent on work a junior analyst could do,if they could access the data. Delphina’s bet is that automating that layer of questioning doesn’t just save money, it changes how a company operates. The incumbent it must beat isn’t another AI startup; it’s the entrenched habit of not asking questions because the answer is too hard to get.

Sources

  1. [Bloomberg, December 2023] AI Startup Delphina Gets Funding to Help Firms Build Data Models | https://www.bloomberg.com/news/articles/2023-12-13/ai-startup-delphina-gets-funding-to-help-firms-build-data-models
  2. [Yahoo Finance, November 2026] Delphina Brings Total Funding to $15M to Take Trusted AI … | https://finance.yahoo.com/technology/ai/articles/delphina-brings-total-funding-15m-150100284.html
  3. [Ground Level AI, November 2026] Chris Best spent years looking for a better way to understand Substack's data. One AI startup finally gave him the answers he needed | https://www.groundlevel-ai.com/p/chris-best-ceo-substack-data-delphina
  4. [Citybiz, November 2026] Delphina Raises $7.5 Million in Funding Led by Radical Ventures and Costanoa, Bringing Total to $15 Million | https://www.citybiz.co/article/933232/delphina-raises-7-5-million-in-funding-led-by-radical-ventures-and-costanoa-bringing-total-to-15-million
  5. [Delphina.ai, retrieved 2026] Delphina.ai, Better, faster decisions with data | https://delphina.ai/
  6. [LinkedIn, Patrick Schwarz, retrieved 2026] Patrick Schwarz profile | https://www.linkedin.com/in/patrick-schwarz
  7. [TechCrunch, April 2020] Tecton.ai emerges from stealth with $20M Series A to build machine learning platform | https://techcrunch.com/2020/04/28/tecton-ai-emerges-from-stealth-with-20m-series-a-to-build-operational-machine-learning-platform/

Read on Startuply.vc