Field notes

Your dashboard is not the interface anymore


Dashboards are useful and I'm not about to stand here and be rude about them—I've spent a good chunk of my life inside them and a few of the ones I keep open have honestly earned their place. But a dashboard isn't how anyone actually thinks. Nobody wakes up wondering what the current blended ROAS by channel is, segmented by campaign objective, against the trailing seven-day window. What they wonder, standing at the kettle with the laptop half-open, is whether the thing that moved overnight is the kind of thing they need to worry about before lunch or the kind they can leave alone. That's the gap I keep coming back to, between how the numbers happen to be stored and how a person actually asks about them. The interface is quietly sliding toward the second one. Not away from the data, just away from making someone translate every ordinary question into dashboard gymnastics before they're allowed an answer.

The better interface is the question

"What was our best campaign last week, and why?" "Did the sale actually work, or did we just pull demand forward from next month?" "Which products are getting traffic but not converting?" Those are the real questions, and right now answering any one of them means someone opens five tabs and rebuilds the same view by hand, which is exactly the kind of friction that quietly stops people asking in the first place.

The thing I actually wanted was for the answer to come straight from the accounts you already own—Shopify, Meta, Google Ads, GA4, Klaviyo, Search Console—not from a pasted export or a screenshot but from live numbers, with enough of the source showing that you can go and check it yourself. And that last bit isn't decoration. If the system can't show you where the answer came from then it isn't analysis, it's a confident-sounding paragraph, and I've used these tools long enough to take the confident-sounding paragraph with a fairly large grain of salt.

It doesn't replace dashboards—it changes their job

Dashboards are still the right tool for watching things: trends over time, the alert that fires when something breaks, the weekly scorecard the board expects in the same shape every Monday. Keep all of that. But the moment the question is fresh, or messy, or oddly specific, a dashboard turns into a maze and you're the one doing the hard part—choosing the filters, remembering which definition of "revenue" you settled on last quarter, joining the context together in your head, deciding whether the number even matters. A conversational layer can take that first pass for you, so it gathers the relevant numbers and explains what moved and points at the likely drivers, and the part I actually care about—it tells you where the evidence is thin instead of papering over it. Then you decide whether to drill in or just get on with your day.

We already run a version of this every morning, which is where I worked most of this out. The 06:00 briefing pulls the overnight numbers across the accounts and lands a plain-language read before anyone's opened a single tab—what changed, what's probably behind it, what's worth a look. It's a scheduled push rather than something you can interrogate live, so it's the precursor and not the finished thing, but it's the same idea: the answer comes to you, sourced, before you've had to go and assemble it. I worked out the hard way which numbers actually mattered at 6am and which ones just looked dramatic, which is also why I call it battle-tested, because it literally has been.

The plumbing is the real work

The bit you see is embarrassingly simple—it's a box, you type a question and an answer comes back, and that's the whole of it. The bit that decides whether it's any good is all underneath, and none of it photographs well: secure access to each account, the metric definitions written down so they don't quietly drift, permissions, logging, the source links, sensible defaults, and a clear set of rules about what the thing is and isn't allowed to do. For most brands the first useful version should be read-only, and I'd argue fairly hard for that. Let it answer, let it explain, and don't let it start moving budgets around just because somebody typed a sentence with conviction. You can add controlled actions later, once the answers have earned it. First you make the answers trustworthy.

What this unlocks for a team

  • the founder asks a question and gets an answer, instead of waiting two days for a report
  • the marketer sanity-checks performance without rebuilding the same spreadsheet from scratch
  • the agency shows its working—source-backed reasoning—instead of handing over a polished guess
  • everyone shares the same definitions, so "revenue" means the same thing on Tuesday as it did on Monday

And it lowers the cost of curiosity, which matters more than it sounds. When a question is annoying to answer people quietly stop asking it, and you never find out about the thing you didn't look at. When the answer is cheap they keep asking—and a business that keeps asking just tends to catch things sooner.

If this is your problem

This is the Conversational Data Layer—a copilot wired into your own accounts, scoped to your data and how your team actually asks questions. It's a bespoke build from $5,000, because the work is in the plumbing rather than the chat box. If you'd rather map it out before committing, the $2,500 roadmap sprint is the low-commitment way in: we scope what's possible against your actual setup and you decide from there. The morning briefing and the ongoing guardrails sit in the $3,000/mo Intelligence Retainer. Full pricing is at /pages/ai-implementation.

None of this is really about dashboards, or about AI. It's your existing data, finally answering when you ask it a question.