Predicting Wrong Inventory Records

Using AI to Predicting Wrong Inventory Records in Retail

Inventory Record Accuracy is arguably THE most important metric for retail operations in the new omni-channel retail context. Our original research identified a sales increase when inventory records were trued up of between 4 and 11% (click here)

Our new research, with just over a third complete, is showing highly encouraging results.

Transforming Retail through AI

Retail has long grappled with the challenge of inventory inaccuracies. The academics are starting to reveal the power of statistical methods and an approach that could revolutionise how retailers manage inventory, shifting from reactive to predictive strategies.

Impressive Predictive Accuracy:

Based on initial results from three retailers their methods have has shown some remarkable results. In one retailer, they were able to predict 81% of perfect error-free predictions, with a 93% prediction where the absolute error was lower than one.

The Bigger Picture:

Improved inventory accuracy isn’t just about correcting numbers; it’s about enhancing the entire retail experience. Accurate records lead to better stock management, efficient cycle counting, and increased sales. These improvements directly translate into business growth.

What next?

The research will be completed by December 2024, however for a recap of the meeting see the video with the full transcript below. If retailers, CPG’s or academics would like to see the full recording of the meeting, email Colin Peacock at colin@ecrloss.com

Inventory Record Accuracy Full transcript summary (edited for clarity / brevity)

Colin Peacock: John, thank you so much for joining the call and the other day we had a long-awaited call with the academics Aris, Yacine and Christoph, and they were sharing their early results from their research that looked at predicting wrong inventory records.

In the meeting, they shared that they had results from three of the eight retailers where, using the retailers data, they were looking to predict wrong inventory records i.e. can you tell whether the inventory is wrong?

We invited you to join the meeting as you have over 40 years of experience in inventory management in retailing, from your time working for retailers such as Petsmart, Osco Drug and Ulta Beauty. You have also worked for Blue Yonder and RGIS, I think you have all angles covered. We also know each other from the Consortium of Excellence in Operations or COER, where we are both members and every year attend their summit, at either Harvard or Wharton.

In short, you’ve seen this inventory counting, inventory record accuracy problem from all sides, quite amazing.

This is why we value so highly your perspective on what we saw presented last week. And what we’re going to be doing going forward in terms of this group.

I know you wrote down some notes, so what were the highlights from your time with us last week?

John Bloomfield: Well, I was very excited by the work that Aris, Christophe, and Yaseen had done.

I’ve been looking at inventory accuracy most of my career.

The issue is always what’s on the shelf for the customer. It doesn’t matter what your book says.

I can’t tell you how many times I would meet with stores, and you look at the shelf and then determine what the book says, and it’s very different. Now It might be in the back room, or it might be gone.

So, the ability to actually start predicting what is going to go wrong with the inventory is huge.

And then it’s only changed as buy online, pick up in store has become popular.

Recently I ordered an item for somebody for a Christmas present. It said it was there, they’re gonna pick it, and no.

Guess what? About an hour or two later, it came back and says your order is cancelled, they didn’t have the inventory.

So it’s still a huge issue, and it’s costing sales. So, you know, I believe inventory accuracy is a huge opportunity. And so I was very excited about the work that they did, where they can actually start predicting inventory inaccuracy because what you can do from that, you can start fixing it.

The accuracy they got was huge. So first of all, they were smart in terms of how they separated things and said look, we determined 15% aren’t predictable and that’s OK.

You know, if I can get 85%, 80% on anything, that’s huge, that’s actionable.

And then within that 85%, I mean, they did two grocery stores, but also an apparel store.

And the issue with apparel is that, in my opinion, it’s much harder because you’ve also got colour, flavour, style.

Oftentimes you can make inventory mistakes and accounting doesn’t care because it’s the same dollar.

But if I have an extra-large black sweatshirt on the books and physically I have a small brown sweatshirt, that’s a real problem for the customer.

So the numbers that they were getting were huge. They were getting 90 plus percent accuracy in terms of predicting, and you know when you went plus or minus one, it was even better

So, I think it’s huge what they’ve been able to do. And then in terms of next steps, you can start doing things like what’s the value of the stocktake, what about cycle counting, and you start getting an ROI.

In my past lives in retail, basically, stock take was considered an unnecessary evil, an expense.

So generally, you know, the thought is, how do I get this as cheaply as possible and just be done with it?

And now you’re looking at potentially saying I can actually give you the ROI, those extra cycle counts will get you this much more in stock.

And there’s plenty of work that’s already been done that says improvements in stock, equals improvement in sales.

So, I think the follow-on from the research they’re doing will be huge in terms of kind of moving towards those types of things.

And then the second part is, once you do that using the AI, and they talked about it a little bit already, but you can start saying what’s the root cause, you know, is it multiple sales locations?

Is it too much stock in the backroom?

Is it the physical attributes of the product?

It’s small and there’s 15 different colours.

Whatever it might be, but now you start seeing that you can even work with manufacturers saying I need you to give us better ways to identify items.

I did some work in cosmetics and I remember, I think it was Revlon maybe, that they started putting the number on the base of the lipstick that faced out to the consumer and the stock takers.

So you could easily see that, hey, this is not the right item, it’s in the wrong place.

So, I just see that there’s a huge amount of opportunity for improvement for retailers. And to me, that was very exciting.

Colin Peacock: Yes, it’s almost like we’re making inventory record accuracy and inventory counting sexy.

But it’s clear that they’ve got a massive computer. It’s probably the size of Wales. But I think it takes a huge computer to have done all this, and that’s what they keep on impressing upon me

The project’s going to be done with the full report in December 2024.

But in the meantime, I think people really got a grasp of the direction of travel and you captured it very well.

So, really big appreciation for your time sharing your thoughts.

And join us again. We’ve got a meeting on January the 9th.

A lot of times we talk about inventory accuracy, and we define it, certainly in the academic world defines it as very binary. It’s either right or it’s wrong. But in the world of practice, you know, we’ve recognised that ± 10% is often used as a zone.

And what we’re trying to do is get through some new research. To get some precision on that.

When is it right to use binary and when is it right to use zones and what determines that? There’s some common sense in there, but we want to try to figure it out

There’s a meeting on January the 9th at 10:00 AM Eastern which is 3:00 PM UK time where the academics will present their research proposal and the retailers will say where they want to do it or not. So, we’ll see you then, John.

John Bloomfield: Yep.

Colin Peacock: Until then, all the very best.

John Bloomfield: Great. And look forward to it.