How Expert works

Your own sourcing process, encoded and tireless. Here is the whole loop.

What Expert actually is

Expert is automated reverse sourcing. The manual version is something every experienced FBA seller has done: find a small seller with few feedback reviews, open their storefront, run every listing through SellerAmp, keep the ones that clear your margin and eligibility bar, then look at who else sells those winners and repeat. It works because new sellers who are already getting sales have usually found real arbitrage gaps, and their storefronts point straight at more of them.

The problem with doing it by hand is stamina. You lose focus after 40 products. You forget which storefronts you already checked. You keep grinding through a dead store because you spent ten minutes on it and want it to pay off. Expert runs the same loop, but it never gets bored, never rechecks the same ASIN, and cuts losses on cold stores by a rule instead of a mood.

Every verdict comes from your own criteria: your minimum profit, your ROI floor, your BSR cap, your exclusions. Expert does not have opinions about products. It has your opinions, applied thousands of times without drift.

The loop, step by step

A run starts from seeds. You can seed it three ways, and mix them freely:

• ASINs. Products you already know are good, or suspect might be. These are the strongest seeds because winners lead directly to the sellers carrying them.
• Seller IDs. A small storefront you found by hand. Expert scans it directly.
• Keywords or pasted Amazon URLs. A search term like "kitchen gadgets" harvests up to 60 result ASINs (default) as fresh depth-0 products.

From there the loop runs on three queues: searches, products, storefronts. Searches always go first, then products, then storefronts. Within products and storefronts, the best-scored task is picked next by default, so the most promising work jumps the line while ties stay in arrival order.

Each product task opens the ASIN in a SellerAmp lookup inside a small minimized worker window, pulls the numbers, and scores the product against your criteria. Pass or fail, the verdict is recorded and the ASIN goes into a persistent seen-history (last 5,000 kept) so later runs never waste time re-checking it. A passing, non-excluded product becomes a finding: saved with its convergence count, source seller, root seed, depth, and the reasons it passed, and pushed to your Google Sheet if connected (up to 3 attempts with backoff; a quiet no-op if Google is not linked).

Then comes the part that makes it reverse sourcing. When a product passes (or is a seed, or qualifies under the expansion mode you picked), Expert reads the offer list and evaluates every seller on it. Sellers with 10 or fewer reviews (default maxSellerReviews) and a real storefront get queued for a storefront scan at one level deeper. Each storefront scan harvests up to 100 listings (default, before adaptive scaling), and every new ASIN enters the product queue carrying its source seller. Winners lead to sellers, sellers lead to more products, products lead to more winners. That is the whole engine.

Depth is capped at 5 rounds by default, so a run cannot wander infinitely far from your seeds.

The three expansion modes

Expansion is the decision to follow a product's sellers outward. All three modes only ever expand through products confirmed UNGATED for you; a GATED product is a dead end no matter what. The modes differ in how picky they are beyond that.

• Discovery is the widest net. Any ungated product can expand, even ones that failed your numbers. Only two blocks ever apply: blacklisted brands, and off-brand products when you are running a brand focus. Use it when you want maximum map coverage from thin seeds.
• Balanced (the default) also blocks expansion through products that trip your structural exclusions: category mismatch, variations, private label, IP complaints, hazmat, Amazon on the listing, oversized, overweight, multipack. But it deliberately ignores your soft numeric criteria (price, BSR, sales, spread) for the expansion decision. A product that misses your profit target by $0.80 is still a good clue about where its sellers shop, so Balanced follows it while never saving it as a finding.
• Quality adds a real-sales test on top of Balanced: estimated sales per month must be known, and at least half your minimum SPM. With the default minimum of 100, a product needs 50+ SPM to open new doors. Slowest growth, cleanest neighborhoods.

In every mode, what gets saved is identical: only products that fully pass your criteria. The modes only change which products get to recruit new storefronts.

Adaptive pruning: cold stores get dropped

Expert keeps a per-seller yield ledger: products scanned from that store, winners found, and how often the store's items clear your margin target. Two things happen with it.

First, cold stores sink. Once a seller has 8+ scanned products, a low winner rate pushes all of its remaining queued products down the priority order. They are not deleted, just deprioritized, so warmer leads run first.

Second, cold stores get cut. If a store hits 12 scanned products with zero winners (default abandonColdStoreAfter: 12), Expert abandons it: its remaining queued products are pruned, and it is never expanded through again. The pruned ASINs are freed, so if a warmer store also carries one of them, it can still be rediscovered from there. This is the disciplined version of "this store is a waste of time, move on," applied at exactly 12 instead of whenever frustration kicks in.

Pruning also shapes harvest size up front. Storefronts owned by sellers with 2 or fewer reviews get their full listing cap; 3 to 5 reviews get 75%; more reviews get 50%. The newest sellers, the most likely to be sitting on fresh gaps, get the deepest scans.

Convergence: when small sellers agree

Here is a signal manual sourcing almost never catches: the same ASIN showing up in three different low-review storefronts. One small seller carrying a product is a data point. Three carrying it independently means three separate people did their homework and reached the same conclusion. That product is very likely a live gap, and its offer list is a directory of exactly the sellers you want to scan next.

Expert tracks this in a convergence map: every ASIN it encounters, and the set of sellers seen carrying it, built from storefront harvests and offer lists as the crawl runs. Convergence is the dominant term in task scoring, so a product spotted in multiple storefronts jumps to the front of the queue. Scores are computed live at dequeue time, which means a product already sitting in the queue gets promoted the moment a second storefront is seen carrying it. Convergence also gates expansion: by default a non-seed product needs at least 1 seller in its set before its own sellers are followed.

Why it moves at the speed it does

Expert is deliberately paced like a careful human, because that is what keeps a run alive for hours instead of minutes. The defaults: at most 3 concurrent tabs, a 6-second gap between task launches with random jitter of up to 35% either way, and a 60-second timeout per tab. The jitter is a feature. Fixed intervals look robotic; a 4-to-8 second wobble looks like a person browsing.

When Amazon or SellerAmp pushes back, Expert backs off instead of pushing through. A CAPTCHA, robot check, or rate-limit error triggers an immediate cooldown; so do 5 consecutive errors of any kind. A cooldown pauses everything for 5 minutes (default), then resumes with the failed task retried first. If cooldowns keep repeating, notifications escalate, but the crawl never kills itself: only your Stop or genuine completion ends a run.

Login trouble gets the same measured treatment. Repeated Seller Central login prompts are double-checked with an independent probe before Expert believes them, so a false alarm becomes an ordinary cooldown instead of a dead stop. A confirmed logout pauses the run, and it re-verifies on a timer and resumes on its own once you are signed back in. A heartbeat every 30 seconds plus persisted state means the run survives Chrome suspending the extension's background worker mid-crawl; in-flight tasks are simply requeued at the front.

The takeaway

Expert is not a black box and it is not a scraper hoovering up random listings. It is one specific, proven sourcing method, the low-review-seller reverse-sourcing loop, encoded step by step and pointed at your own criteria. Every number it respects is yours: your minimum profit, your BSR ceiling, your exclusions, your gating status. Every product it keeps is one you would have kept yourself. The only difference is that it checks product number 400 with exactly the same rigor as product number 1, drops dead-end stores on schedule, and notices when three strangers quietly agree on the same product. It is your process, tireless.