Cotera founder Tom Firth explains why the dashboard was never the point, and what actually has to happen before AI can act on your business data.
Dashboards have been the default output of enterprise data for forty years. Someone builds the chart, someone else logs in every morning, stares at it, and makes a call. Tom Firth, founder of Cotera, thinks that whole loop is about to become optional. Not because dashboards are useless, but because they were never actually the goal. They were a proxy for the goal.
I had Tom on Deconstructing Data to dig into why AI agents are starting to replace the dashboard-then-decide workflow, what’s still standing in the way at most companies, and why the answer to “is our data ready for AI” is almost always no, at least not yet.
Why Are AI Agents Replacing Dashboards Instead of Just Analyzing Them?
Here’s the thing. A dashboard was always downstream of a decision someone wanted to make more consistently. Tom traced it back to the first big data warehouses at companies like Walmart in the 80s and 90s.
“What was the point of data in the first place? The point of data was basically for companies to try and make more repeatable higher quality decisions and what was the point of making decisions? The point of decisions was basically to do stuff.”
— Tom Firth, Founder, Cotera
That’s a big deal when you actually sit with it. The dashboard was the middleman. A human looked at the chart so they could go do something. Now AI agents can skip the middle step entirely for a huge category of decisions, especially the tactical, high-volume ones.
“It’s not that dashboards as a whole are obsolete. It’s that most dashboards were always a proxy for us trying to do action more efficiently, and now you can just get straight to the action. So why go through the dashboard half the time?”
— Tom Firth, Founder, Cotera
Tom’s example was support tickets. A dashboard used to surface a list of 20 tickets past some resolution threshold, and a human worked through them one by one. That’s exactly the kind of tactical, rules-based triage an AI agent can now just do. The strategic calls, the “where should I take my business” questions, still belong to a person. But a lot of what lived on a dashboard was never actually that strategic.
What Does It Actually Take to Break Down Data Silos?
Every company I talk to has some version of the same problem: customer data scattered across Salesforce, a data warehouse, a support tool, a product analytics platform, and nobody’s sure which record is the source of truth. Tom made a point I don’t think gets said enough. This problem predates AI by a decade, and AI doesn’t magically fix it.
“I wish I could offer a sort of magic bullet there, but that is actually a lot of work.”
— Tom Firth, Founder, Cotera
There are two paths. You can do the traditional, boring work of consolidating everything into one place with proper data modeling. Or you can give an AI agent live access to each individual system and let it piece together a customer record on the fly by querying Salesforce, the warehouse, and the product database in real time. Tom said the second approach works better than you’d expect, but it’s slower, more expensive in tokens, and less reliable than just having your data actually joined together with a clean identity key.
That second point matters more than people give it credit for. Without a consistent way to match a customer across systems, even a capable model ends up guessing.
“If you haven’t got a single key to query on, then it’s surprising what AIs can do to figure it out. Claude will go and run three or four experimental queries and come up with ideas… but it’s very slow, it’s very wasteful, it spends a lot of tokens doing that, and it’s just less reliable.”
— Tom Firth, Founder, Cotera
And when teams do try to tackle this, the failure mode is almost always scope. Tom’s advice: don’t boil the ocean.
“The failure mode is generally to kind of try and do this like gigantic, ‘let’s do all the data all in one go,’ and then 18 months later you’re still fighting and there’s no value. Pick a very specific use case, cut the scope down to just the data you need for that one question, and deliver it end to end.”
— Tom Firth, Founder, Cotera
What Does It Take to Prepare Your Data for AI?
This is where Tom’s take got genuinely interesting to me. Getting data ready for a dashboard and getting it ready for an AI agent are not the same project. The type of data matters. AI is comparatively bad at raw numbers, that’s still SQL’s job, but it’s suddenly very good at everything that used to be a black box: Slack messages, emails, survey responses, PDFs sitting in a cloud bucket that nobody’s touched in years.
Tom told a story from his time as a junior auditor, tying source documents back to numbers people claimed. Some companies kept an organized, alphabetized room of receipts and invoices. Others handed him a cardboard box of loose paper.
“You have to treat it surprisingly. It’s not a human, I don’t like anthropomorphizing it per se, but you have to think about it in a fairly similar way. If you’ve given it the cardboard box, it’s not going to find anything. But if you give it the well-indexed booklet, then it’s going to get right there.”
— Tom Firth, Founder, Cotera
That’s the real work. Not cleaning numbers, but extracting just enough structure, tying a PDF to a customer ID, stripping the junk out of an email header, so an AI system can actually locate the right answer instead of guessing at one.
“Most businesses are quite a bit behind in the sort of operational sense of using AI versus where they are in the coding sense. It’s just so much work to organize the data and get it ready. The models are good, but they’re not that good unless you’ve given them the right context.”
— Tom Firth, Founder, Cotera
The Bigger Picture
I keep coming back to Tom’s point about coding versus business operations. AI writes good software right now because the context, the whole codebase, is already sitting there organized. Business data almost never has that luxury. It’s scattered, unstructured, and undocumented, and no model is going to fix that for you overnight.
That’s actually the opportunity. Companies that put in the unglamorous work of organizing and connecting their data, especially around a consistent identity key, are the ones who are going to get real leverage out of AI agents in 2026. Everyone else is going to keep getting confident, well-written, wrong answers.
Learn more about Tom Firth and Cotera at cotera.co or connect with him on LinkedIn.
At BDEX, clean, resolvable identity data is exactly what makes AI-driven decisioning possible in the first place. If your data isn’t organized around a reliable customer key, no agent is going to save you. Visit bdex.com and click “Talk to an Expert” to see how we can help.
Watch the full episode: How to Prepare Business Data for Automation and Smarter Decisions
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.
