Why a “Business Intelligence as Code” Release Is Quietly Relevant to Every Seller Who’s Drowning in Dashboards
If you run a cross-border operation, you don’t have a data problem. You have a dashboard problem. Your Shopify store spits out one version of revenue, Amazon Seller Central another, and your TikTok Shop ads manager a third that disagrees with both. Meanwhile, your finance person is exporting CSVs into a spreadsheet that has become a sacred, unmaintainable artifact. We pay for tools like Helium 10 and Klaviyo that give us more charts, not fewer. The bottleneck was never data collection—it’s the last mile of turning that data into a decision you can defend. That’s why the open-sourcing of Evidence Core matters more than its “developer tool” label suggests. It points to a future where the reporting layer of your business is version-controlled, reviewable, and—crucially—buildable by the same AI agents you’re already using to write product descriptions. This isn’t about becoming a data engineer. It’s about making your analytics as manageable as your product listings.
The Problem: Your Reporting Stack Is a Black Box You Can’t Audit
Let’s be honest about how most e-commerce reporting actually works. You log into a BI tool like Looker or Tableau, click through a dozen filters, and hope the person who built the dashboard five quarters ago defined “net revenue” the same way you do today. You can’t see the logic. You can’t trace a number back to the SQL query that produced it. And if you ask for a change, you’re queued behind a backlog that stretches into next quarter.
For a cross-border seller, this is existential. Your margin depends on reconciling Amazon fees, currency fluctuations, and logistics costs that change weekly. When a dashboard says your Walmart Marketplace channel is profitable, you need to know *why*—not just trust that it is. The evidence is in the query, but you can’t see it.
The launch of Evidence takes a different approach. It treats the entire BI platform as code. The metrics, the dashboards, the reports—all defined in plain text files using SQL and markdown. This is not a new idea in the broader software world, but it’s a radical one for the analytics space, where most tools are WYSIWYG drag-and-drop affairs that generate unreadable, unmaintainable configuration blobs behind the scenes.
The commercial product, Evidence Studio, was built on this open-source foundation. By now open-sourcing the whole framework as Evidence Core, the company is betting that the future of analytics isn’t a proprietary UI—it’s a set of files that live in a repo, that you can diff, review, and roll back. For a seller who has been burned by a “temporary” fix in a dashboard that turned out to be permanent, this is the difference between a black box and an open ledger.
Why Amazon sellers should care more than Shopify ones
If you’re a Shopify-first brand, you’re probably living in a world of app-store dashboards that are good enough. Triple Whale or Polar gives you a decent picture of ad spend and revenue. It’s not perfect, but it’s fast and it’s visual.
Amazon sellers have it worse. You’re dealing with a data dump that arrives daily with dozens of columns, a fee structure that changes with FBA updates, and a P&L that has to account for storage fees, referral fees, and the occasional “other transaction” that eats your margin. The generic dashboards don’t cut it. You need custom logic, and you need to be able to explain that logic to a partner or a lender. With Evidence Core, you can write a metric that says “Net Margin = (Revenue - COGS - FBA Fees - Shipping) / Revenue” and have that definition be the single source of truth, not a hidden formula in a spreadsheet. That level of transparency is worth more on Amazon than on any other channel.
How It Differs From the Incumbents: The Agent-Friendly CLI
The most interesting part of this launch isn’t the open-sourcing itself—it’s the bet that coding agents will be the primary builders of analytics. The maker, Adam McAskill, explicitly frames the new CLI as a way for agents to be “productive” with the tool. He even suggests you can tell Claude Code to “Take a look at docs.evidence.dev and set up a new project for me.”
This is a direct challenge to the incumbent tools. Metabase is open-source and great for ad-hoc queries, but it’s not built for the kind of structured, version-controlled metric definitions that a team needs to maintain. Superset is powerful but has a steep learning curve and a UI that feels like it was designed by engineers, for engineers. Evidence’s bet is that the future isn’t a better drag-and-drop interface—it’s a text file that an AI agent can edit, and a human can review.
For a cross-border operator, this is a massive unlock. You don’t need to hire a full-time data analyst to build a custom dashboard. You need a clear spec, a few hours of an agent’s time, and a review process. The agent can write the SQL, define the metrics, and structure the report. You, the domain expert, review the logic. The launch materials and reviews point to this exact use case—one user mentions using it with ClickHouse for “AI internal apps” that need data “FAST.”
Where the math breaks: The “review” problem
The catch is what the maker himself admitted in the comments. When asked where coding agents perform best, he noted that “things with analytical weight like metrics definitions and query logic end up being the parts you want to review more carefully.” This is a polite way of saying the AI will generate a beautiful report with a subtle logic error that makes your “Total Revenue” actually mean “Total Revenue Minus Refunds.”
That’s a real risk. If you hand a coding agent your data warehouse and say “build me a profitability dashboard,” you will get a dashboard. Whether it’s correct is another story. The tool gives you the mechanism to review—the files are right there in a repo—but it doesn’t give you the discipline. You still need to be the one who checks the math. The good news is that this is a solvable problem. The bad news is that it requires a level of rigor that most sellers don’t currently apply to their BI tools.
What Cross-Border Sellers Can Borrow From This (Even If You Never Write SQL)
You might be thinking, “I’m not a developer, and I don’t want to be.” Fair enough. But the philosophy behind Evidence Core is transferable to how you run your entire operation. Here are three concrete lessons you can steal this week, no coding required.
First, treat your reporting logic as a product. In the Evidence world, a metric is a file. You can version it, you can review it, and you can roll it back. If you’re still running your P&L in a single Excel file that gets emailed around, you are one accidental deletion away from a crisis. Move your core metrics into a system where they are defined once and used everywhere. If that system is Google Sheets with named ranges and query functions, so be it. The point is to have a single source of truth that you can audit.
Second, use AI agents for the grunt work, but review the logic. You’ve probably already used ChatGPT to write a product description or a customer service response. The next step is using it to build your reporting. Describe your business logic in plain English—”I want a daily report that shows ad spend from Meta, revenue from Shopify, and the calculated ROAS”—and let an agent generate the initial version. But do not deploy it until you’ve manually checked the numbers against a known-good period. The tool enables this workflow, but the responsibility is yours.
Third, demand transparency from your vendors. When you sign up for a new SaaS tool, ask to see the definitions of their key metrics. If they can’t show you a clear, reviewable definition of what “Active Customer” means, that’s a red flag. The open-source movement in BI is a response to the opacity of legacy tools. You should apply the same standard to every tool in your stack. If a vendor can’t explain their math, they’re hiding something.
Where My Judgment Says It Falls Short
No tool is a silver bullet, and Evidence Core has some gaps that matter for the e-commerce use case. The first is that it’s still a developer-centric tool. The open-source release is aimed at people who are comfortable with a command line and a code editor. If you’re a solo seller or a small team without a technical co-founder, the learning curve is real. You could hire a freelancer to set it up, but that adds a dependency.
The second gap is the lack of a managed hosting story for the open-source version. Evidence Studio is the commercial product, but the core framework is now free. That means you’re responsible for the infrastructure. For a cross-border business that’s already juggling 3PLs and customs brokers, adding “self-host our BI platform” to the to-do list is a tough sell. The product is powerful, but it requires operational maturity that many sellers simply don’t have yet.
Finally, the integration story is not plug-and-play. You’ll need a data warehouse or a direct connection to your databases. If you’re currently exporting CSVs from Shopify Admin and Amazon Seller Central and stitching them together in Excel, Evidence Core is a step too far. You’d first need to invest in a proper data pipeline—something like Fivetran or Airbyte—to get your data into a warehouse like Snowflake or BigQuery. That’s a significant infrastructure investment for a business that just wants to know if their latest ad campaign is profitable.
The “good enough” trap
There’s also a strategic risk. Tools like this make it easier to build custom dashboards. That’s a double-edged sword. You can end up spending more time building and maintaining dashboards than actually making decisions. The best operators I know have a “kill your darlings” mentality—they delete reports that don’t drive a decision. Evidence Core makes it cheap to create, but it doesn’t make it cheap to maintain. You still have to update the logic when your business model changes, and that’s a recurring cost.
What I’d Watch / Test Next
This week, I’m not going to tell you to go rewrite your entire BI stack. That’s a weekend project, not a Tuesday one. But here’s what I’d do to test the waters without diving in headfirst.
First, set up a sandbox project. If you have a technical person on your team—or if you’re willing to spend an hour with the docs—spin up a small Evidence Core project that connects to a sample dataset. Don’t try to build your entire P&L. Just build one report that answers a single question, like “What was our blended ACOS across Amazon and TikTok Shop last month?” Get a feel for how the SQL and markdown work together and how easy it is to review the logic.
Second, run a side-by-side comparison. Take one of your current dashboards—say, a weekly sales report from your existing BI tool—and rebuild it in Evidence Core. Compare the time it takes to make a change. In your old tool, a change might require clicking through a UI. In Evidence, it’s a text edit. See which one feels more maintainable to you, not to a developer. That will tell you if this philosophy is worth adopting.
Third, watch the agent angle closely. The most exciting part of this launch is the potential for coding agents to build and maintain analytics. In the next few weeks, try prompting a tool like Claude or GitHub Copilot to generate a metric definition for you. See how close it gets. The gap between “close” and “correct” is your job. If that gap is small, you have a new superpower. If it’s wide, you know you still need a human in the loop.
The takeaway is simple: the era of the black-box dashboard is ending. Whether you adopt Evidence Core or not, the standard it sets—version-controlled, reviewable, agent-friendly analytics—is the direction the industry is heading. Start demanding that level of transparency from your tools, and start building the muscle to review the math yourself. Your margins will thank you.
