I used to trust Amazon’s restock recommendations blindly. After all, who’d know better than the behemoth itself how much inventory we needed? Turns out—almost anyone with a decent calculator and an ounce of common sense.

Three months ago, we started experimenting with an AI agent built on Claude Sonnet 3.7 (now upgraded to Sonnet 4.0), tasked with managing our Amazon FBA stock replenishment. It wasn’t about replacing our warehouse team, but rather, empowering them. The AI analyses our actual sales data, compares it against Amazon’s recommended replenishments, then drafts an email to our warehouse team, clearly instructing them what to ship. Simple, effective, and entirely human-friendly.

We haven’t gone out of stock even once since starting this test—and more crucially, we’re no longer sitting on mountains of inventory gathering dust (and fees) in Amazon’s warehouses.

Here’s what I’ve learned—and why it matters for every seller navigating the complexity of Amazon’s ecosystem.

Why the Stakes Are So High for Amazon Sellers Right Now

Cash flow isn’t just king—it’s the whole monarchy. Every pound locked in inventory is a pound that can’t be spent on marketing, innovation, or growth. Over-reliance on Amazon’s recommended stock levels creates a dangerous blind spot, leading sellers to tie up crucial capital unnecessarily.

I didn’t fully appreciate this risk until we ran the numbers ourselves.

My Personal Moment of Clarity

My old mindset was simple: Overstocking felt safer than risking going out of stock. But, as every seasoned seller knows, excess stock isn’t just wasted space—it’s wasted cash and an operational headache.

My turning point was embarrassingly practical: when our warehouse manager casually mentioned that over 30% of the products he was sending to FBA were unnecessary. I remember thinking, “That’s not safety—that’s laziness dressed up as caution.”

We needed clarity, precision, and agility—things Amazon’s default recommendations just didn’t offer.

Data Snapshot: Amazon vs Our AI Agent

When we finally compared Amazon’s recommendations against our AI’s numbers, the results were stark:

Think about that: hundreds of thousands of pounds of frozen capital, released back into our business. It was a revelation.

Principles We Learned the Hard Way

The clarity we gained boils down to four truths:

1. Amazon’s Goals Aren’t Always Your Goals.
Amazon’s primary interest is never running out of your products. But they have less concern about your cash flow or holding costs.

2. Stock Levels Should Be Driven by Sales Velocity, Not Fear.
Data-driven inventory decisions save capital and reduce risk more than stockpiling ever will.

3. Trust, But Verify.
AI doesn’t replace judgement—it sharpens it. Our AI agent doesn’t blindly dictate stock levels—it calculates, recommends, and drafts actions clearly and transparently.

4. Automation Doesn’t Kill Jobs; It Frees Talent.
Our warehouse team spends less time packing surplus stock, allowing them to refocus on high-impact work, like optimising fulfilment processes or handling real customer demands.

Tactics You Can Implement Immediately

If you’re keen to replicate our success, here’s the playbook:

Step 1: Run a Controlled Pilot.
Start small—pick 5–10 SKUs and compare your AI-calculated stock against Amazon’s recommendations.

Step 2: Analyse Sales Velocity.
Establish your baseline sales velocity (units per day, per week, per SKU). Feed this into your AI’s logic—it’s more precise than broad averages.

Step 3: Automate with Precision.
Set your AI agent to send weekly emails to your warehouse clearly stating exactly what to ship. Human-friendly clarity is vital for smooth operations.

Step 4: Monitor and Adjust.
Check performance weekly, tweaking your AI’s parameters based on real-world results and seasonality changes.

Facing Up to the Frictions

I won’t sugarcoat this—implementing an AI agent comes with hurdles:

But here’s the crucial point: the friction of implementation pales in comparison to the perpetual drain of unnecessary inventory.

Real Stories: The Power of Getting It Right

Three weeks into our experiment, one of our top SKUs—a health supplement—suddenly experienced a spike in sales due to unexpected social media attention. Our AI agent reacted swiftly, recognising the increased velocity. It drafted an email with adjusted replenishment instructions, and the warehouse team shipped immediately. No meetings, no guesswork, no panic.

By the time Amazon’s algorithm caught up, our product was already stocked and thriving. If we’d followed Amazon’s original recommendations, we’d have been swimming in unnecessary inventory elsewhere, leaving little capital to flexibly respond to this sudden spike.

What Does Good Look Like?

Imagine this scenario:

That’s exactly what good looks like—and exactly what we’re living today.

Your Actionable Next Step

Here’s my challenge to you: Pick your top 10 SKUs right now. Manually compare the stock Amazon recommends to what your actual sales data suggests.

Then ask yourself, honestly: What would AI say?

Because whatever you do, don’t let blind trust freeze your capital when clarity could release it.

Final Thoughts

We’ve transitioned from Claude Sonnet 3.7 to the even more powerful Sonnet 4.0, and the results continue to amaze us. AI isn’t just a flashy piece of tech—it’s become an indispensable business partner, freeing our team from unnecessary burdens and unlocking capital we didn’t realise we were missing.

For us, going back isn’t an option. The question is, why haven’t you moved forward yet?