Author: Nick Scheffers
Date of publishing: 1st May 2026
Imagine: your company has implemented a brand-new system that automatically predicts the right quantities for purchasing and allocating stock.
Sounds awesome, right? But there’s one problem: it doesn’t seem to be working properly.
The system says: “Order 500 units.”
But you’re an experienced retail professional, and you know that you can only sell 350 units, at the most.
So, you override the system, and put in 350 units instead.
After all, it’s still learning. It can’t possibly know how to make the right decision…
…Right?
But here’s the problem: when users override the system, find workarounds, and ignore the recommendations, it becomes impossible for automation to do its job.
And then it fails to deliver the benefits you want to see.
Trust is the foundation of automation
Whenever automation fails, I generally see that the problem is rarely the technology itself.
Instead, I see that automation most often fails when users do not trust it.
After all, if you’re relying on a system for critical decisions, you must be able to put your faith in it.
It’s hard to build trust in automation. This just doesn't happen on its own, especially when it never gets the chance to prove itself.
Instead, this must be catalyzed with a precise combination of mechanisms that allow users to feel in control, informed, and aligned with the right priorities.
Mechanism 1: Transparent forecasting
Too often, AI technologies are a mysterious ‘black box’. There’s no way of knowing how they generate predictions, so how can they be reliable?
I’ve seen this myself when using generic LLMs like ChatGPT; these models can generate shocking hallucinations and factual errors (glue pizza, anyone?).
And, I would also say that a retailer shouldn't trust a ‘general model’ AI like that for crucial inventory decisions.
This is why transparent forecasting is a foundation of forming trust with algorithmic technologies.
So, if a system says, “Order 500 units”, you should be able to see where that number comes from, and why.
If a user can see the underlying logic behind the system (including factors like the local weather, MOQ/discounts, re-ordering intervals, seasonal patterns, logistics constraints, redelivery schedules, and replenishment times), then they can understand the basis for the forecast.
They realize that the AI is doing exactly what they normally do, just with much more data and faster.
Mechanism 2: Business rules in your hands
The next mechanism that builds trust in automation is control with business rules.
Again, this comes back a little bit to the ‘black box’ problem. But, instead of seeing how the AI works, this is about gaining some control over that yourself.
This is done with your own team defining precise business rules. These act as guardrails for the algorithm’s logic and reasoning.
Unfortunately, not all AI solutions allow this, but by consciously choosing a platform that lets you set (and fine-tune) business rules, you can ensure that AI decisions are made with logic you determine.
This gives you real ownership over the AI, and this translates into greater trust because you know decisions reflect the nuances of your operations.
Mechanism 3: Explicit exceptions
Naturally, whenever the algorithm creates an unexpected prediction, users will feel the temptation to overrule it.
For this reason, you need to find a way to let the algorithm run (so it can make different/better decisions), but still feel like you can take back the steering wheel if you start to go off-course.
The solution is to define explicit exceptions for when overriding the system is allowed, and to define conditions for when stepping-in is expressly not allowed too.
These clear and precise behavioral rules avoid doubt and make it clear when you should step in and take over.
…And wrap it all up with shared KPIs
The overstock-understock cycle is hard to beat. But retailers must stop seeing this as a ‘stock problem’ and understand that the root cause is always a decision problem.
Automation can help, but only if everyone is pulling in the same direction. And this requires shared metrics.
People working in different retail teams routinely make sub-optimal inventory decisions because they use metrics that tell them it is the right thing to do.
Examples of misaligned departmental metrics:
- Purchasing teams are rewarded for a low cost-per-unit, and this pushes (over)stock downstream.
- Planning teams want to maximize assortment availability and sell-through rates, even when this is uncoupled from real demand at each location.
- And the supply chain is focused on minimizing the costs of inventory, opting for bulk shipments, large MOQs, and low logistics costs (infrequent replenishment).
By contrast, a shared definition of success helps automation to ‘stick’, especially when decisions and KPIs are unified in a single automation platform.
Just making this single shift can massively improve things. For example, when MS Mode adopted shared KPIs, the results were practically immediate.
Want to dive deeper?
Mastering retail has always relied on a combination of data, human experience, and instinct.
But the world is changing, and there’s less room for error.
Automation and advanced algorithms can navigate mind-boggling amounts of data to find the best path forward.
But these technologies can only work if people use them. And that means being able to put trust in the system.
Until the next,
Nick.
Quote: The overstock-understock cycle is hard to beat. But retailers must stop seeing this as a ‘stock problem’ and understand that the root cause is always a decision problem.
