Where the loss actually is
527,500
Shoppers who reached the product page and never reached a cart. Sessions grew 9.5% and orders 4.5% — every point of the gap sits in one step.
Read the step, not the totalE-commerce growth agency
Most agencies manage a piece of ecommerce. We run the whole operation.
Marketplace · DTC · Media · Creative · Data
operated by one team, against the same numbers.
The Welli Warehouse
Most agencies show you a report. This is the model underneath it.
Every channel lands in one warehouse at day and SKU grain — real fees from settlement data, cost of goods, freight, advertising and returns. What follows is the actual instrument, not a picture of one.
Revenue by channel
Ranked by contribution. Bar darkness is rank, not category — the darkest bar is the largest channel.
View as table
Illustrative composite account · trailing 30 days against the same days last year.
Net margin, 12 months
After referral and fulfilment fees, advertising, returns and cost of goods.
Monthly net margin. Hover for the month.
The dashboard on this page runs on an illustrative composite, not a client account. The structure, the fee lines and the detector set are real.
Decomposition
A decline is never one number. It is four.
Traffic, detail page views, add-to-cart and orders each have their own story. Aggregate conversion hides which one moved.
Full-funnel decomposition
Volume at each stage, with the step conversion and what was lost between stages.
View as table
Illustrative composite · trailing 30 days.
Gross to net
Sales and profit are separate problems.
A brand can add thirty percent in sales on Amazon and make less money doing it. The waterfall is the only view that shows both at once.
Of gross revenue kept
14.8%
Advertising is the third-largest deduction on this account — larger than returns and fulfilment combined, and the only one anybody can change this week.
See every fee lineGross revenue to net margin
Every deduction between what the customer paid and what the business kept.
View as table
Illustrative composite · trailing 30 days, settlement-derived fees.
From finding to fix
We don't just find the problem. We operate the fix.
A finding that lands with a vendor who cannot act on it is a report. The same team that runs the detection holds the account, the inventory plan and the pricing — so the fix happens the same week.
Between the alert and the fix
No handoff
Every action is logged against the finding that prompted it, with who authorised it and what it was expected to do — so the loop from insight to action to result is auditable rather than asserted.
How the loop is recordedThree disciplines, one week
What the detection found, what we did about it, and what changed. Each of these is an operating decision, not a recommendation in a deck.
DisciplineWhat fired What we didWhat changed
Inventory Stuck inbound — 3,480 of 3,894 units still in working status nine days after the plan was created. Chased the shipment, split the plan, released the 3PL stock sitting behind it. 24,050 units moved on ten SKUs carrying $10,015 of 90-day sales.
Pricing Buy Box flicker on 15 SKUs against a third-party seller. Re-argued the floor against contribution margin rather than against whoever discounted last. Buy Box back to 83.5% without going under the margin floor.
Forecasting Days of cover under fourteen across the replenishable set. Weighted velocity at 15, 30, 90 and 365 days against FBA and 3PL stock on hand. A ship-this-week list in every weekly check-in — not a dashboard to interpret.
Illustrative composite account · one operating week.
Detection
Twenty-eight things checked every morning.
Deterministic detectors run against the warehouse every day and are triaged by a person before they reach a client. The language model writes the findings up and connects them — it never decides what counts as a finding.
Today's run
3 firing
four more on watch
Versioned and reproducible: the same data produces the same findings, every time. When a detector fires on something we can't explain, it becomes a question for you — and your answer permanently retires that false positive.
Put your account through itDetector grid · today's run
State is shown by colour, an icon and a word — never colour alone.
○ Quiet ▲ Watch ● Firing
Illustrative composite · daily run. Counts are SKUs affected.
What we found, and what changed
Work, stated as numbers.
A US textured-hair care brand +36.1% Amazon sales year on year, across an account we've held since 2021 — four channels on one P&L. Read the work A Caribbean resort 1.63% vs 3.45% Booking conversion on mobile against desktop — a gap worth more than any campaign. Read the work A US confectionery brand Ship this week A days-of-cover call in every weekly check-in — an Amazon department for a brand too small to staff one. Read the work
Where we work
Three surfaces, one model underneath.
Analytics
Your channels modelled into one warehouse at day × SKU grain, with real fees from settlement data — not Amazon's estimates.
MarketplacesAmazon Seller and Vendor Central, Walmart Marketplace and Retail Link, TikTok Shop. Listings, inventory, advertising and brand protection.
Direct to consumerShopify storefront, retention and lifecycle, shipping economics, email and SMS on whatever platform you already run.