Connor Folley, co-founder and CEO of Alkemi, previously built and sold Downstream to Jungle Scout. His new company is built around a blunt idea: most companies don’t have a data volume problem, they have a data shape problem.
Everyone assumes the reason AI isn’t delivering more value inside their company is a data quantity issue. Not enough data, not clean enough data, not enough history. Connor Folley, co-founder and CEO of Alkemi, told me that’s almost never actually the problem. Most companies are sitting on plenty of data. The problem is it’s scattered across Snowflake, Salesforce, HubSpot, spreadsheets, and a dozen other tools, with no shared context about what any of it actually means.
That distinction matters more than it sounds like it should. Connor built and sold his last company, Downstream, an AI-driven e-commerce platform, to Jungle Scout in 2021. Post-acquisition, he ran a data-as-a-service business selling structured data to enterprises making high-stakes decisions. That’s where he watched LLMs arrive and realized the gap wasn’t technology. It was that real business decisions happen in Slack, docs, and spreadsheets, and the data needed to make those decisions usually lives somewhere else entirely.
Why Isn’t My Company’s Data Already Usable by AI?
Connor’s framing here is one of the clearer breakdowns I’ve heard. It’s not one problem, it’s three stacked on top of each other.
First, fragmentation. Your data lives across a dozen systems that don’t talk to each other cleanly.
Second, missing context. AI doesn’t just need rows and columns, it needs to know what the metrics mean, how they’re calculated, and how to reconcile the fact that marketing and sales probably define “qualified lead” differently.
Third, and this is the one that surprised me, the access bottleneck. Most data is technically accessible but operationally locked behind analysts and dashboards.
“Most BI solutions, it’s rare to see a BI solution gain more than 20% penetration across the enterprise. This is a problem that enterprises have solved for decades. Yet we still find that most business users do not have access to the appropriate data at the point of decision.”
— Connor Folley, Co-Founder & CEO, Alkemi
That stat alone reframes the whole AI-readiness conversation. You can have a perfectly clean data warehouse and still fail here, because the failure is about access, not cleanliness.
What Happens When You Skip the Context Step?
We got into this because I’ve watched people hand AI a dataset and just let it decide what matters. That’s backwards. Connor’s point, and mine, is that without a clearly defined goal, AI fills the gap with assumptions, and those assumptions aren’t grounded in what you actually care about.
“Without that context, models just guess, and guessing is not what we want to base our decisions on.”
— Connor Folley, Co-Founder & CEO, Alkemi
He also flagged something worth repeating to anyone rushing to wire up autonomous agents: prove the output is valuable to a human first, before you hand the task fully to an agent. If your process doesn’t work with a human in the loop, it’s not ready to run without one.
How Do You Give Teams AI Access Without Losing Control of Your Data?
This is where a lot of leadership teams freeze up, because they assume there’s a hard tradeoff between security and access. Connor doesn’t think there is. The mistake, he said, is trying to solve permissioning top-down, from the CIO out, across the entire organization at once.
Instead, Alkemi builds bottoms-up, team by team. Marketing gets access to HubSpot, Meta, and Google Analytics data. Sales gets what sales needs. Nobody’s handing an AI system blanket access to the whole company on day one, because any team lead already knows exactly what their team should and shouldn’t touch.
“Most organizations try to give employees direct database access, which is dangerous. Instead, what you want is more controlled exposure.”
— Connor Folley, Co-Founder & CEO, Alkemi
There’s also a data custodianship angle that doesn’t get talked about enough. A lot of companies assume that for AI to reason well over their data, they have to hand that data over to whichever foundation model provider they’re using. Alkemi built around avoiding exactly that, using a retrieval-based framework that connects to your existing warehouse, Snowflake, BigQuery, Databricks, wherever, without the underlying models ever taking custody of it.
Are AI Coding Tools Creating a New Security Problem?
This was a tangent worth keeping in the article, because it’s a real risk most teams aren’t thinking about yet. With vibe coding tools making it trivially easy to spin up internal or external-facing apps, Connor pointed out that those apps often ship with real security gaps, because the tools optimize for the fastest path to something working, not the most secure path.
If you’re letting teams build and deploy tools this way without review, you may be creating exposure faster than your security team can track it.
Will AI Replace Data Analysts?
No, and Connor’s reasoning here is worth sitting with if you’re an analyst wondering where this leaves you.
“AI actually makes the analyst more important, not less, by allowing them to move to higher order, more strategic work. Today analysts spend far too much time serving as human middleware to the business user who can’t write a SQL query.”
— Connor Folley, Co-Founder & CEO, Alkemi
The shift he described is from analysts building one-off reports to analysts architecting the underlying data layer the whole company relies on, defining metrics and business logic once instead of answering the same ad hoc question fifty times.
“Their role shifts from being the person who builds the report to being the person who architects that data layer that the entire organization relies on.”
— Connor Folley, Co-Founder & CEO, Alkemi
That’s not a demotion. That’s the analyst becoming infrastructure.
The Bigger Picture
What stuck with me most from this conversation is that “AI readiness” gets treated like a data cleanliness checklist, when it’s really an organizational design problem. Fragmentation, missing context, and locked-down access aren’t technical footnotes, they’re the actual blockers standing between most companies and AI that produces trustworthy answers. Fix the shape of your data and who can reach it, and the AI part gets a lot less mysterious.
Connect with Connor Folley on LinkedIn or learn more about what Alkemi is building at alkemi.ai.
If your own organization is sitting on identity data that’s fragmented, hard to trust, or hard to act on, talk to a BDEX expert about turning it into something your teams and your AI can actually rely on.
This article was adapted from an episode of Deconstructing Data, BDEX’s weekly podcast on data-driven marketing. Tune in live every Thursday at 4:15 PM Eastern on LinkedIn.
About BDEX: For companies that need clean identity data to power their products, BDEX offers unmatched quality and execution. Visit bdex.com and click “Talk to an Expert” to get started.
Watch the full episode: Preparing Enterprise Data for the AI Era
