Threecolts AI Operator / Best practices Updated Jun 2026

AI Operator · Beta guide

Write it like a work order.

AI Operator does its best work when you hand it a clear, complete request — what you want and every filter that matters — and let it return a decision or a clean export. These habits are drawn from how our most effective sellers actually drive it.

The one rule

Put everything in one message — the marketplace, the dates, the ASINs, and every filter — instead of drip-feeding it across replies.

What this is really for

Run your store like you've got an ops team.

The sellers who get the most from Operator stop asking it questions and start handing it the work — the slow, careful jobs you'd give an analyst or a VA. Checking matches, auditing titles, sweeping for dead stock, watching every listing. Done in seconds, on your say-so. The leap from heavy user to power user isn't sharper prompts; it's running your operation through it.

How power users do it

ChainStack simple steps into one request — each step is reliable; the precision comes from the order.

SaveWhen a prompt lands, keep it. Swap the date, retailer, or category and it's a system you reuse.

Run in parallelOne chat per workstream — sourcing, repricing, alerts — several running at once.

Write prompts that land

specific in · useful out

Say it all in one message

Each request goes to a specialist that starts fresh, with no memory of earlier turns. The more you state up front, the less gets lost.

Run it again but for last quarter

FeedbackWhiz profit by product, US, Jan 1 – Mar 31 2026

ROI %BSRSeller count Price rangeCategoryMin profit Gated / ungated

Always name your marketplace

With nothing specified, Operator defaults to US — a common cause of empty or surprising results. Name the marketplace whenever you mean another.

Check the same ASIN on the other one

Look up ASIN {insert ASIN} on Amazon UK

New 03

Narrow step by step

Build a complex selection as a chain of simple filters rather than one tangled condition. Each narrowing step is reliable; the precision comes from stacking them in order.

Min price $10 → out of stock → no sales in 90 days → table

One workflow per chat

Keep sourcing, repricing, and label jobs in separate chats — mixing them muddies results. To adjust, restate the full request rather than saying "do it again."

Run jobs that finish

scoped · repeatable

New 05

Keep big jobs scoped

A tightly scoped request finishes cleanly; a sprawling one that sweeps the whole account often won't. Run sourcing one source page at a time and pull data per batch, then combine the results yourself.

Scan my entire supplier catalogue in one go

Scan this one category page for 30%+ ROI, 3+ sellers

Ask for a file — then open it

For anything past a handful of rows, ask for an export and open it in Excel or Sheets. Glance at it first: confirm the rows you expected are actually there before you build on it.

Export that batch as a CSV

Reuse what works — and ask when you're stuck

When a prompt gives strong results, save it; swap the retailer, category, or date and run it again — good prompts become repeatable systems for your team. And if you're not sure where to start, just ask Operator what it can do. Try: What can AI Operator do for me today?

III

Verify before you trust

leads, not verdicts

Check your account data first

Operator's output is only as good as your inputs. Before anything analytical, make sure the data behind it is in place — gaps produce confident-but-wrong answers.

COGS are entered
Inventory sync has completed
The right marketplace is selected

New 09

Read velocity from sales rank

For "what's selling" and "what's dead," use BSR / sales rank as your velocity signal. It's a strong directional proxy — treat sharp rank moves as the signal, not exact unit counts.

Rank my in-stock ASINs by sales rank, worst first

New 10

Let the matching agent check your finds

No more eyeballing every lead. After a search, ask Operator to verify the matches — the matching agent reviews each result against the Amazon listing and flags the ones that don't line up (wrong size, pack, or product), across Walmart, Target, and most retail sites. One caveat it doesn't cover: live pricing and promos can still lag, so confirm the current price before you commit.

Run a Quick Picks search, then verify which results are real matches

New 11

Scope your alerts tightly

When you stand up listing alerts, narrow them as much as the options allow and prune the product list so monitors stay focused on what's live. Tight scope at setup is what keeps the inbox useful later.

When you hit a wall

push, then route it

Tell a glitch from a limit

A blank reply, a timeout, or a job that stalls is usually transient — rephrase, scope it smaller, and run it again. But if Operator tells you a capability isn't supported, take it at its word: that's a real limit, not a fluke to retry ten times.

Scope down, then retry

Most failures on a big ask clear when you break the job into smaller passes. Halve the date range, run one source or batch at a time, and combine the results yourself. A request that finishes beats a sweeping one that hangs.

Confirm it, then tell us

Not sure a limit is real? Ask the same thing a different way — if it holds, it's a limit, so note it and move on rather than fighting it. Then bring it to your session. Repeat failures and genuine gaps are exactly the signal that shapes what we build next — the fastest way to get the capability you want is to flag it, not to work around it forever.

Copy-ready prompts

paste · swap · run

Sourcing

Search Quick Picks for Toys & Games on Amazon US with at least 30% ROI and 3+ sellers.

Narrow step by stepNew

Show me my SmartRepricer products on Amazon US: minimum price $10, then out of stock, then no sales in the last 90 days. Return the result as a table.

Find dead stockNew

List my in-stock ASINs on Amazon US ranked by sales rank, worst first, so I can spot dead stock to liquidate.

Batch

Create an InventoryLab batch called "June OA Batch" and add these ASINs with $8 COGS.

Labels

Generate 30-up SKU labels for batch "Prep Batch 1" in PDF format.

Reprice by promptNew

Raise the minimum price on {insert MSKU} to $14.99 and drop the max to $24.99.

Reverse searchNew

Take these ASINs I'm ungated in and run a reverse search to find where to source them — then verify which results are real matches.

Product research

Look up ASIN {insert ASIN} on Amazon US and estimate ROI if I buy at $15.99 and sell at $35.00.

Try something bold

big jobs · real time back

These are the slow, careful jobs you'd normally hand to an assistant — or keep putting off. Operator does each in one pass. Pick one and run it today.

Fix every long title at once

The old wayAuditing titles one by one — the weekend job you keep pushing to "later."

With OperatorIt finds every in-stock title over the limit, drafts a compliant rewrite for each, and dedupes — in a single pass.

Try itFind my in-stock titles over 75 characters and draft a compliant rewrite under 75 for each.

Surface your dead stock in one sweep

The old wayScrolling reports and guessing which SKUs have quietly stopped moving.

With OperatorChain filters — min price, out of stock, no recent sales — then rank by sales rank to see exactly what to liquidate.

Try itFrom my in-stock products: min price $10, out of stock, no sales in 90 days — rank by sales rank, worst first.

Put your whole catalog on watch

The old wayFinding out about a hijack, Buy-Box loss, or suppression days after it cost you.

With OperatorStand up live alerts across your active listings, so problems come to you instead of you hunting for them.

Try itSet up alerts for Buy Box loss and listing suppression across my active products.

Each of these is a real workflow sellers run end to end. Start with one — the first time it hands back an hour of work in seconds is the moment it clicks.

Running this hard?

Let's build your playbook together.

If you're pushing Operator across several workflows a week, a short working session with our team turns your routine into saved, repeatable prompts — and puts your wishlist straight in front of the people building it.

Book a working session 30 min · for active beta sellers

Clear in. Useful out.

One complete message, one job per chat, the right marketplace, scoped jobs, and verified data. Get those right and Operator does the rest.