The real bottleneck in cross-border data isn’t dashboards — it’s definitions
Every cross-border operator I know has the same dirty secret: the numbers in your Monday meeting don’t match the numbers in your Tuesday board deck, and nobody can explain why. Not because anyone is lying — because “revenue” means something different in Shopify, Amazon Seller Central, TikTok Shop, and the spreadsheet your ops lead built in 2023. Add returns, chargebacks, FX, and marketplace-specific fee timing, and you get five defensible answers to the same question. Querri’s latest release — the Library and the Librarian, launched November 18th, 2025 — is a direct bet on this problem. That’s why it matters to sellers more than to the average SaaS buyer: our data lives across a dozen systems that each define the business slightly differently, and AI that “picks one” definition is worse than no AI at all.
What the Library actually solves
The founder’s framing is blunt and, I think, correct: pointing a general-purpose AI at real company data with hundreds of tables and five definitions of “active customer” means it picks one — and may pick a different one tomorrow. Querri’s answer isn’t a smarter guesser; it’s a semantic layer the AI stands on. The Library maps the questions your team keeps asking, the KPIs that answer them, the definitions behind those numbers, and the raw data they’re computed from. The Librarian is the agent that builds and maintains that map in plain English.
Concretely, the release includes 125+ connectors plus spreadsheets and files, KPIs defined once and tagged to Revenue, Cost, Efficiency, Quality, or Risk, business rules settled in chat (their example: “sales means paid orders only”) saved as facts, and monitors that watch a KPI by region, store, or rep and tell you which one moved. Every answer shows the rows, the calculation, and the sources it read; every column loaded gets 180+ quality checks. Querri claims it flagged an alternate defensible answer on 73% of responses versus 12% for a leading general-purpose agent on the same data — a vendor benchmark, so treat it as directional, but it’s the right metric to benchmark.
Why Amazon sellers should care more than Shopify ones
A DTC brand on Shopify has one source of truth and a clean Shopify Analytics view. An Amazon FBA seller has settlement reports, Amazon Seller Central reports, ad console data, and a 3PL feed — and Amazon’s fee timing means your “profit” depends on which week you cut. That’s exactly the five-definitions problem. If Querri’s semantic layer can hold your definition of net margin after FBA fees, referral fees, and returns, the chat layer becomes genuinely useful instead of a toy.
How it differs from what you’re already using
The obvious comparison is the warehouse-native agent layers: Snowflake Cortex and Databricks Genie. A commenter raised exactly this, and the maker’s answer is fair: those are agent-building tools inside platforms you already have to staff and fund. Querri sells the full stack — ingestion, data lake, semantic layer, chat, dashboards — on one bill, aimed at teams without the budget for multi-quarter data projects.
The second comparison is the BI incumbents. Looker and Tableau already have semantic modeling; what they don’t have is a natural-language front door that a non-analyst will actually use. The third is the e-commerce analytics specialists — Triple Whale, Helium 10, and the like — which are excellent at their lane but don’t span Amazon, TikTok Shop, and your 3PL in one model. Querri’s pitch is horizontal; the specialists’ pitch is depth. Most sellers will end up wanting both.
Where the math breaks
The sharpest exchange in the launch thread is worth reading in full. A commenter pushes: how do you verify the number is right unless analyses are pre-orchestrated — and pre-orchestration kills agent autonomy? The maker’s response is the most honest thing on the page: pre-orchestration works in the Library (exact SQL for a KPI, precomputed views for common questions), but Querri can fall back to raw data and still run the autonomous agent. That’s a real architectural answer. It’s also an admission that the trust problem isn’t fully solved — it’s made inspectable. For a seller reporting to a board or a lender, “inspectable” is better than “confident,” but it’s not the same as “audited.”
What cross-border sellers should borrow from this
Three things, regardless of whether you buy Querri.
First, write down your definitions. The single highest-ROI data project for a cross-border brand is a one-page document defining revenue, net revenue, contribution margin, active customer, and return rate — per channel. Querri productizes this; you can do a crude version in Notion this week.
Second, treat ambiguity as a feature, not a bug. The 73% alternate-reading claim is the interesting number, not the accuracy claim. Any tool that surfaces “here’s the other defensible answer” is teaching your team where your data model is weak.
Third, the screenshot-to-dashboard feature is a migration wedge. Rebuilding an old dashboard from a sketch, calculations included, is the kind of feature that gets a tool past the “we already have dashboards” objection. If you’re consolidating tools in Q1, that’s the demo to ask for.
The governance question nobody answered cleanly
Asked who owns the truth when definitions evolve, the maker said the Data Owner — the library or workspace administrator — with the Librarian surfacing ambiguous overnight questions and suggested semantic-layer updates. That’s a sensible answer, but it quietly creates a new job: someone has to actually review those suggestions. In a five-person e-commerce team, that someone is probably you.
Where my judgment says it falls short
The pricing isn’t disclosed on the page beyond a HUNTER1000 promo for 1,000 bonus credits, so I can’t tell you whether this is a $200/month tool or a $2,000/month tool — and that changes the recommendation entirely. The compliance stack is real (SOC 2 Type II, ISO 27001:2022, HIPAA support, no training on customer data), which matters if you’re handling customer PII across markets, but it doesn’t tell you anything about answer quality on messy marketplace exports. Scale claims — “into the 100s of millions of rows” — are fine, but an Amazon seller’s problem isn’t row count; it’s column semantics. And the horizontal positioning cuts both ways: a tool built for “ops, customer success, and revenue teams” won’t ship an FBA fee reconciliation template the way a vertical specialist would.
What I’d watch / test next
This week, before any trial: export your last 90 days of Amazon settlement reports and your Shopify orders, and write down your own definition of contribution margin. Then sign up with the promo code and ask Querri the question that broke your last AI attempt — specifically, “what was contribution margin by channel last month, after returns and marketplace fees?” Watch whether it surfaces the alternate reading or hands you a confident number. Ask for the screenshot-to-dashboard demo on a report you actually use. And ask sales, in writing, for the price tier that includes the 125+ connectors and workspace access controls. If the answer comes back reasonable and the margin question survives inspection, this is worth a serious Q1 evaluation. If it doesn’t, you’ve still learned where your definitions are broken — which was the actual problem all along.
