The most expensive line item in a cross-border seller’s P&L is the one nobody has audited. You can tell me your blended CPC, your Amazon referral fee, your freight cost per CBM, and your Shopify app bill to the dollar. But ask what your team actually spends on AI tools each month — which seats are active, which projects that spend maps to, which vendor’s API quietly repriced itself — and most operators go quiet. That blind spot is exactly what DepthData — a new product and company by Ali Uyanik — is trying to fix. It isn’t an ecommerce tool. It’s a finance and governance product. And it’s worth your attention because the discipline it preaches — label every number by how it was verified — is the discipline your marketplace P&L has been missing.
What DepthData actually does, and why the “system of record” framing matters
DepthData’s pitch is refreshingly direct: “Companies now pay for four or five AI tools (ChatGPT, Claude, Copilot, and more) but can’t answer the basics: what are we spending, who’s using it, and which seats sit idle?” The product “connects every AI tool into one audit ready view of spend and adoption.” That sounds like a line from every overfunded SaaS deck. But the maker adds a harder standard: “we never show a number we can’t actually verify from the tool’s own API. Every figure is labeled by how we know it, and we never read prompts, only metadata like usage and seats” — you can read his explanation in the launch thread.
In the enterprise world, this problem already has markets. Expense platforms like Ramp and Brex capture the transaction but not seat-level usage. SaaS-management platforms like Zylo and BetterCloud track licenses and renewals but aren’t built for AI API spend, where cost is metered per token, per project, per teammate. Cloud-cost platforms like CloudZero are built for engineers, not for a brand manager expensing a ChatGPT Plus seat. DepthData’s wedge is narrower: AI-tool spend specifically, with “exactly what each vendor’s API can and can’t expose” made visible. That transparency is what separates it from a dashboard that just looks pretty.
The mechanics matter more than the demo. In the comments, Uyanik explains that Anthropic, OpenAI, and Cursor have real cost APIs, so DepthData pulls actual dollar spend straight from them. Where a vendor doesn’t expose cost, it pulls usage data and combines that with the seat prices you enter from your contract. Managers assign people to departments in an employee table, and spend rolls up from there. For projects, the tool can pull project spend directly when the API supports it — OpenAI’s API platform and Vercel if you set up tagging — and where it doesn’t, it allocates each person’s real spend across projects based on usage, labeled ALLOCATED rather than measured. That last distinction is the most interesting part of the whole launch.
Why Amazon sellers should care more than Shopify ones
DTC operators running on Shopify have a weird advantage: much of their AI spend is already inside the Shopify app ecosystem, billed through the platform, visible in one admin. Amazon sellers, by contrast, live in a decentralized hell. Amazon Seller Central is your system of record for sales, but not for the AI tools you use to generate listings, translate keywords, write ad copy, summarize reviews, and automate customer service. Those tools are bought direct: Helium 10 for product research, ChatGPT and Claude for copy, repricers with AI bid logic, review-analysis tools, email automation — each on its own billing cycle, each with seats that may or may not be active.
The maker describes the pain perfectly: “The answer usually means logging into five different admin consoles and cobbling together a spreadsheet, and even then you’re mostly guessing.” That is the story of every multi-account Amazon operation I’ve ever consulted. The finance person sees the total on the credit card, the operations lead sees who has a login, and the owner sees neither. DepthData’s team-and-department rollup is the exact model a multi-marketplace seller needs: treat each marketplace or brand as a department, each listing group as a project, and require every AI purchase to be labeled by who uses it and what it’s for. You can argue about the product’s maturity all day, but the architecture is right.
The verification-label idea is a masterclass for ecommerce reporting
The core habit worth stealing is the label. Uyanik says it plainly: “real data where the APIs give it, your input where they don’t, and everything labeled so finance can actually trust the numbers.” In cross-border ecommerce, we don’t label enough. We drown in confident numbers that don’t share their sources.
Your Amazon settlement report says one thing, your accounting ledger says another, and the reconciliation is a monthly ritual of hope. A PPC campaign’s ROAS is measured by Amazon attribution, but a blended ROAS across TikTok Shop and your Shopify store is at best allocated — because no platform sees the other platform’s clicks. A landed cost per unit is allocated, not measured, because freight and tariffs get split across SKUs using rough rules. We don’t tag these numbers with their confidence level, so finance treats a rough split like a hard fact. Then the inevitable happens: the owner asks where the number came from, nobody can answer, and the meeting turns into a game of blame.
DepthData’s labels — measured, allocated, manual — are a habit worth copying. Before you argue over a number, ask: did we read this from the source system, or did we derive it from a model? That single question would kill half the finance meetings in cross-border ecommerce.
Where the math breaks
Allocation is honest, but it’s still guesswork. The maker openly says that Claude tells you project usage but not project cost, so DepthData splits each person’s real spend across projects based on how much they used each one. That is reasonable. It is also the same trap as ecommerce attribution: usage volume is a proxy for value. A ten-minute conversation that unblocks a listing optimization is not equal to a ten-minute conversation that drafts five hundred product titles. If you govern spending with allocation rules, you are managing a budget, not measuring ROI. That is fine as long as you label it. The danger is that “allocated” sounds slightly more trustworthy than “estimated.” It isn’t.
There’s also a garbage-in problem. If a vendor’s API doesn’t expose cost, DepthData relies on seat prices you type in from your contract. If your contract has been renegotiated and nobody updated the panel, the output is precise-looking fiction. The label tells you the number is manual, but labels only help people who read them.
What cross-border sellers can borrow this week
You don’t need to be the buyer for DepthData to steal its architecture. Here is the pattern I’d implement in any ecommerce operation, regardless of what tool you use:
- Inventory every AI tool. Ask the team what they actually use: ChatGPT, Claude, Microsoft Copilot, Cursor, image generators, translation tools, review-analysis tools, anything with an API. Pull the last three months from admin consoles and card statements.
- Define verification labels. Force every line item to carry one: MEASURED if the vendor’s billing API returns it, MANUAL if you typed it from an invoice, ALLOCATED if you split it across projects or SKUs using a rule.
- Enter your real contract prices. DepthData’s contract pricing panel is the right idea: use what you actually negotiated per seat, not the vendor’s marketing page price.
- Map tools to departments and projects. Treat each marketplace or brand as a department, and each listing group or product line as a project. Then run the exact report DepthData is built around: what are we spending, who’s using it, which seats sit idle?
- Apply the same discipline to non-AI costs. Your Klaviyo dashboard measures campaign revenue; it doesn’t measure how much AI time went into writing those flows. Tag that AI time as manual or allocated, and suddenly your marketing margin tells a truer story.
The point isn’t to adopt a particular SaaS tool. It’s to stop pretending that a number is true just because it appears on a dashboard.
Where my judgment says DepthData falls short
DepthData is exactly the kind of tool that should exist, and exactly the kind of product I’d be cautious about making the center of my finance stack on day one. The launch page does not disclose pricing or a full integration list. The maker says it’s early and he’s building it mostly solo. The built-with page lists Claude by Anthropic, which is good dogfooding but also a reminder that DepthData itself is dependent on vendor APIs — including the cost-tracking endpoints it’s meant to watch.
Granularity is another open question. One commenter asks whether you can trace spend down to a specific feature or session, “or is it more of a monthly aggregate view right now?” The source material doesn’t answer that. If you sell a product that consumes Anthropic’s API per customer request, you need per-feature cost, not a monthly rollup. DepthData’s project allocation may be too coarse for that use case.
It also doesn’t, from the launch material, close the loop with procurement or payments. A true system of record should match the vendor invoice, the card charge, and the usage ledger in one place. DepthData gives you the usage ledger and a contract pricing panel. Reconciliation is still on you.
The bigger strategic gap: by refusing to read prompts, DepthData stays privacy-safe but value-blind. You can see how much a tool costs and how much people use it, but not what it produced. For ecommerce, that’s where the ROI question lives. A listing writer that costs $50 a month and produces 200 product titles is a bargain. A $50 seat that autocompletes one email a week is waste. Seat utilization is a hygiene metric, not a value metric. I’d rather have a tool that looks at usage patterns and flags “this seat is idle” than a tool that claims to know “this seat is worth it.” DepthData is honest about that limit, which is rare. But honesty doesn’t solve the ROI gap.
What I’d watch / test next
I’d test DepthData this week the same way I’d test a new repricer: small, reversible, and aimed at a specific question. Pull your last three AI vendor invoices, enter your real seat prices, map your teams and brands, and ask the live demo to show you which seats are idle and which projects are burning money. In parallel, build the same verification-label spreadsheet internally. The tool may not be ready to replace your finance stack, but the habit is ready now.
I’d also watch whether DepthData adds feature-level cost tracing, invoice reconciliation, and SSO. Those three additions would take it from a useful dashboard to an actual system of record. Until then, treat it as a very good mirror for a problem every cross-border seller has: you’re spending on AI in four or five places, and nobody can answer the basics. That’s not an AI problem. It’s an audit problem. And it is not going to fix itself.
