“Best sellers” collections that ignore stock will feature heroes you cannot ship. Merchandising is not a design toy. It is paid and organic demand standing in a queue.
Score, then human
Automatic rank without stock health is how you put a three-day-cover SKU first. Automatic rank without approval culture is how a bad model day wrecks the homepage. The useful pattern is: score → preview → approve → measure impact → revert if needed.
Weights (what the engine actually uses)
Conversion rate, revenue, stock health, margin, seasonal velocity ratio, campaign profit ROAS boost, velocity, Search Console signal, new-arrival freshness. Those are the ingredients. Your store’s outcome still depends on data quality (COGS filled, GSC connected, sales volume).
Bridge to paid
When a ranked set should also become a Meta product set or a bot-made collection for ads, that is a pipeline step, not a free-for-all. See product sets and set rot.
When auto-rank is wrong
New drops need temporary pins. Clearance should not win on raw conversion if margin is trash. Gift season changes velocity ratios. That is why pin/bury and 24-hour style revert windows matter more than a perfect weight table.
If sales volume is thin, scores are noisy. Honest systems show that; they do not invent confidence.
Questions people actually ask
Does Ralph silently rewrite every collection every night with no approval?
Write paths are flag-gated. Preview ranking can be read-only. Approvals and a revert window exist so a bad auto-rank is undoable.
What goes into the score?
Weights include conversion, revenue, stock health, margin, seasonal velocity, campaign profit ROAS, velocity, Search Console signal, and new-arrival freshness.
Can I still pin or bury products?
Yes. Operator controls include pin, bury, manual reorder, approve, and impact readback.
Does merchandising create Meta product sets?
The product set pipeline can create a Shopify collection and hand off toward Meta set creation / Google tagging when those paths are enabled. That is a pipeline, not a guarantee that every store has writers on.
How to merchandise a Shopify collection by stock and sales
This is doable by hand on any Shopify plan. The goal is a ranking that changes when your warehouse changes, and a written rule for the exceptions.
- Understand what Shopify's built-in sorts will and will not do. A collection can sort by Best selling, Newest, Price, Manually and a few others. None of them consider stock. Best selling is a lifetime-ish popularity measure, so it will happily put a hero with two units left at the top of your highest-traffic page. Manual sort does what you told it in March and never updates. Both problems have the same cause: no inventory input.
- Stop sold-out products showing first, natively. The cheapest win takes two minutes. Make the collection Automated and add the condition
Inventory stock is greater than 0, which drops sold-out products out of the collection entirely. If you would rather keep them visible for SEO and sizing context, leave them in and handle position in the next steps instead. Pick one, deliberately. - Pull the two inputs you need. Export units sold per product for the last 14 to 28 days from Analytics → Reports, and available stock per product from Products → Inventory. You now have sell-through and cover. That is enough to rank honestly, before any weighting cleverness.
- Score it in a sheet, and keep the score boring. A workable first score is sell-through rate multiplied by a stock-health factor: 1.0 when cover is comfortable, 0.5 when it is getting thin, 0 when it is below your threshold. Multiply by margin if your COGS is filled in. Resist adding eight weighted factors on day one, because you will not be able to explain why anything moved.
- Apply the order, then pin the exceptions by hand. Sort the collection manually to match your score. Then override deliberately: new drops usually need a temporary pin above where the maths puts them, because they have no sales history yet. Clearance usually needs burying if the margin is bad, however well it converts.
- Write down every pin with an expiry. Pins are the part that rots. A new-drop pin that nobody removed is why a collection still leads with something from two seasons ago. One line per pin: product, why, and the date it comes off.
- Re-run it on a schedule and check the top six. Weekly is enough for most stores, twice weekly during peak. The quick sanity check is the top six positions: can you ship all of them, today, in the sizes people actually buy? If not, the ranking is wrong regardless of what the conversion column says.
- Keep the paid product sets in step. If a collection also feeds a Meta product set, changing one and not the other creates exactly the mismatch described in product set rot. Make it one job on the calendar, not two that drift.
The manual version costs about an hour a week per collection. That is fine for five collections and unworkable at fifty, which is the point at which scoring it automatically starts to pay for itself.
If the catalogue is the problem: Ralph maps every Shopify variant to its Meta and Google catalogue item, scans for missing attributes, and drafts the fixes and the feed changes for approval. See Ralph · How Ralph reads the catalogue · Docs