GlassBox by Secew reads your own Amazon data and shows its work.
GlassBox is a web application for Amazon selling partners who run their own advertising. It connects to the seller's own Amazon accounts, turns their advertising, catalogue and inventory data into concrete proposed changes, and shows the arithmetic behind every proposal. Nothing is changed in the seller's Amazon account until the seller approves it.
Secew is an independent software company registered in Kyiv, Ukraine, in July 2024. GlassBox is not affiliated with, endorsed by, or sponsored by Amazon.
The problem it solves
An Amazon seller running Sponsored Products has the raw numbers — search term reports, advertised-product reports, Brand Analytics reports, FBA inventory — spread across a dozen exports that do not join up. The decisions that matter are the ones those exports imply but never state: which search terms are provably losing money, which bids are below the auction, which listing is about to run out of stock while its campaign keeps spending, and which query is converting for the category but not for this seller.
GlassBox joins those sources for one seller at a time, computes the decision explicitly, and puts it in an approval queue with the evidence attached. The seller approves or rejects. Approved actions are exported as a Sponsored Products bulk operations file, or executed through the Amazon Advertising API when the seller has connected it.
What the application does
Brand Analytics funnel
Reads the seller's Search Query Performance report and lays out the whole funnel for each query — search volume, impressions, clicks, cart adds, purchases — next to the seller's own share at each step. Where the seller's purchase share is far above their impression share, the query is under-served; where a single funnel step collapses, that step is named as the bottleneck.
Negative keyword engine
Seven layers, from semantic relevance through a Bayesian test of whether a term's conversion rate is below break-even, n-gram aggregation, cannibalisation and harvest hygiene, a deterministic guardrail layer, and a full audit log. Each candidate carries the numbers that produced it.
Bid ceilings from margin
Computes a maximum cost per click from the seller's own break-even ACoS and the revenue-per-click measured over 7-, 14- and 30-day windows, then flags bids above the ceiling and targets whose bids have fallen so far below the auction that impressions stopped.
Stock-aware advertising
Pan-EU stock by country and listing-level quantities are joined to campaigns, so a recommendation to scale spend is blocked when the product will go out of stock before replenishment arrives. Restock quantities are computed with an explicit service level.
Contribution margin, and where each part came from
Referral and fulfilment fee estimates, returns, and the seller's own cost of goods produce a per-product contribution margin and a break-even ACoS — the number every advertising decision above is measured against. Each component carries its source and its date, and a fee taken from an estimate is labelled as an estimate rather than presented as a settled figure. Reconciliation against Amazon settlement reports is not implemented yet; when it is, this page will say so.
Experiment lab
A change can be registered as an experiment with a written hypothesis, a measurement window justified in text, and an after-action review that compares what happened against what was predicted. Where a sibling variant is available it is used as a control.
Full feature list, organised by the Amazon data role each one needs →
How a decision travels through the system
- Ingest. The seller's reports arrive either by file upload or by the application requesting them from Amazon on the seller's behalf, once the seller has authorised it.
- Compute. Deterministic code produces the candidate change and the numbers behind it. Every candidate records which method produced its deciding number.
- Validate. A separate deterministic guardrail layer re-checks the candidate against the seller's brand terms, whitelist, recent conversions, stock position, unit margin, and hard caps on how much a budget or bid may move.
- Approve. The candidate appears in a queue with its reason and its evidence. A human accepts, rejects, or whitelists it. Nothing reaches Amazon before this step.
- Execute. Approved actions become a Sponsored Products bulk operations file the seller uploads themselves, or — if the seller connected the Amazon Advertising API — are written by the application, after a dry-run diff.
- Verify. When the next report arrives, the application measures whether the executed action produced the effect it predicted, and says so on the card.
Data handling in one paragraph
GlassBox works only with data belonging to the selling partner who authorised it, and only for that selling partner. Each customer's data lives in its own separate database file. Credentials are encrypted at rest. The application requests no personally identifiable information about Amazon customers, does not call the Orders API, and stores no buyer names, addresses or contact details. Amazon information is never pooled across customers, never sold, never published, and never used to produce insights about anyone other than the seller who supplied it. The full list of sources, purposes and processors is on the Data & AI page; the full policy is in the Privacy Policy.
Who it is for
- Amazon selling partners running Sponsored Products on their own catalogue, typically with more than a few hundred search terms a month — below that, the statistics have nothing to work with and the application will say so rather than invent a recommendation.
- Sellers operating in more than one European marketplace who need the same rules applied consistently across all of them.
- Brand-registered sellers who have Search Query Performance available and want it joined to their own advertising numbers instead of read in isolation.
It is not a marketplace research tool and not a product-hunting tool. It analyses the data of the seller who authorised it, for that seller, and it does not sell, publish or pool what it learns.
Contact
GlassBox is operated by Secew, an independent software company in Kyiv,
Ukraine, registered there in July 2024.
Registered legal name: LLC Secew
Registered address: Strazheska Akademika St, 5, apt. 15, 03126 Kyiv, Ukraine
Company registration number: 45498326, registered 18 July 2024.
Email: support@secew.com — support, sales, and all data-protection requests. We answer within two business days. There is no login and no form to fill in before you can reach a person.