35% Fewer Stockouts: What AI Demand Forecasting Changes in Supply Chain Operations

Get your forecast wrong in one direction

and you've got empty shelves.

Wrong the other way, and you've got a warehouse full of product nobody wants

AI promises to fix that.

Some of that promise is real.

This is part two of our AI strategy series with

Professor Brett Duarte of ASU's W. P. Carey School of Business

Demand forecasting, inventory optimization, and how

those two decisions connect.

Here's what actually works.

Welcome back to the Data Blades podcast, part of the Retail Razor

podcast network, where we cut through the clutter and get to the data-driven

insights that move your business forward.

I'm Ricardo Belmar.

I'm Casey Golden.

This is part two of our three-part series on AI strategy for the supply chain.

Our guest across the series is Professor Brett Duarte,

clinical associate professor in the NASPO Department of Supply Chain Management

at Arizona State University's W. P. Carey School of Business, and one of

the faculty teaching their AI-enabled supply chain strategy executive program

in October from the 5th to the 7th

Last episode, we spent our time on procurement, supplier

risk, contract intelligence, and where agentic AI is starting to

show up in sourcing decisions.

If you haven't heard it, be kind, rewind, go back one episode.

Because today we move downstream into operations.

This is the part of the supply chain where a bad forecast turns into

either empty shelves or a warehouse full of product nobody wants

. Casey Golden: Which is exactly where AI gets interesting and

also where it gets oversold.

So, we're asking Brett to be specific.

We'll dig into demand forecasting and what AI methods

actually improve on, inventory optimization and how forecasting and

inventory decisions connect, and what the operations side of that three-day

executive program at ASU covers.

If you're interested in registering for the AI-Enabled Supply

Chain Strategy Executive Program, October 5th through 7th, we have the

details in the show notes, including a discount code for Data Blades listeners.

That's right.

If you're a supply chain leader in your org, this is an excellent opportunity

to come away with new frameworks for evaluating AI opportunities, get a

clear read on where companies are getting genuine competitive advantage

from AI, and be in a room full of your peers working the same problems.

Seats are going fast, so don't wait.

Register using the link and ID code in the show notes and on screen right here

for $1,500 off the cost to register.

Now, before we jump into today's topic, let me tell you about the Retail Razor

Podcast Network sponsor, RetailClub.

There is still time to join us at the RetailClub AI Festival.

From the founders of Shop Talk and Grocery Shop, RetailClub AI Festival

brings together 2,000 industry leaders from September 22nd to 24th

for three days fully outdoors in Huntington Beach, focused entirely

on how AI is transforming retail.

Hear from 150 leaders shaping how AI is being used across the industry,

explore the technologies changing how retail works, and connect with

senior leaders from across retail, brands, technology, and investment.

Late registration is now available at retailclub.com/retail-razor-podcast.

We'll see you in Huntington Beach!

Now here's part two of our AI strategy for supply chain series

with ASU professor Brett Duarte.

Brett, welcome back to the Data Blades Podcast

Thank you for having me again

Last time we covered the procurement side of AI strategy.

So we're gonna keep going and, continue our series, and this time

turn to supply chain operations.

So tell us, where are you seeing the biggest applications there?

Yeah.

Well, you know, supply chain operations spans, a lot of

different, functional areas.

you know, demand forecasting, inventory management, capacity planning.

There's quality management, production, scheduling, order

management, network design.

There's the whole sales and operations process with IBP.

of course, you can get into, sustainability, cost management, and, and

that's all, in the manufacturing side.

You could also see lots of applications, in service operations management as well.

But of course, Ricardo, there's, there's so much to do.

We've kind of focused ourselves really on, two distinct areas, which I think

is, is really very pertinent and prevalent, and perhaps ones that, that

maybe matter more, more so, though I'm sure other people believe that

everything else is important, as well.

But, demand forecasting.

Okay, demand forecasting is a big area.

We'll talk about that.

And then, of course, inventory management, 'cause there's so much

of money tied up in inventory.

And so those are really gonna be the, the two key areas that we'll focus on,

at least in, in the three-day intensive.

I'm sure as we move along and, and we have more of these offerings, we'll kind

of expand the scope of operations and,

Mm-hmm.

with some of the topics that I mentioned.

Well, I spent a lot of time in the ERP space and supply chain

tech space and, capacity planning typically felt like pretend functionality.

as well as, you know, very focused on supply and demand balancing.

But when it came to forecasting and being able to predict and run

scenarios, that was always a dream.

So has AI helped?

Where are we now?

Yeah.

I mean, wouldn't we all love to have that crystal ball, right,

I know, right?

Absolutely

life, it would make life a lot easier for a, a lot of people.

but yeah, I mean, I think this is, this is an area that I think

AI has helped, quite a bit.

you know, so, when you think of the, the spectrum of different tools that

are out there, know, gone are the days where, where people make any kind of

decision based on, just pure business acumen, experience or judgment, right?

We live in this day and age of, of big data.

and, and using that data is becoming really important.

It is the new oil, if you will, right?

It's such a valuable resource.

And so, when it comes to AI and what we kind of focus on, is really using

various different techniques, right?

Extrapolation, looking at historical data and, and trying and seeing if

those patterns, extend into the future.

various econometrics-based approaches, regression, but also, machine learning.

I think that, that lands up being, such a great space, especially

when you're dealing with, non-linear patterns in the data.

And then now throw in, a whole bunch of new tools, especially with

all these large language models.

and I think it, it really lands up being such a ripe space to help you

tools in, in forecasting, right?

Now, you know, traditionally, what, what we would have done, and this is

something that we do cover in, in our, intensive sessions as well, is to,

to kind of demonstrate and, really educate the practitioner on some of

these, statistical-based tools, right?

Machine learning tools.

but also kind of pulling in and drawing in how large, large language models,

can really be such a valuable tool, as far as, you know, playing this

role of this reasoning layer on top of a lot of your foundation models.

And part of the value is really, what we would call, you know,

narrative synthesis, right?

And multi-source fusion where it has the ability to pull

in qualitative information.

I mean, it's great if you have tons of historical structured data, but

pulling in, qualitative signals, right?

"Hey, there was an article in The Wall Street Journal that talks about

certain sentiment." Is that something you can pull into your forecast?

Or perhaps, even going, a little bit further and looking at things like

a consumer price index, you know, inflation, sentiment analysis, and then

using that to embellish the forecast, and really provide value, right?

Now, why is this such a ripe moment, if you will, for AI and forecasting?

Well, I mean, part of the reason is, you know, I talked about this immen-immense

amount of data that's available, and it's available real-time, right?

We have real-time access to data.

Computation is not as expensive as it used to be.

Models have matured.

I mean, you get into a large language model right now, and gone are the

days where we, we teach people how to, how to write Python code.

I, I-- we used to do that a few years back, go through the syntax.

and now you have these AI models that will say, "Hey, which, which model would

you like me to run?" And it generates a lot of the code for you, which really

takes, forecasting now, into the hands of managers and executives that perhaps

didn't have those skills before, right?

And so, couple that with the whole agentic piece that we were

talking about, Ricardo, earlier.

Now you have these agentic workflows, that can really elevate your

forecast, and, and put context around it, make it more meaningful.

and you don't have to wait for these, these windows, right?

Where we wait for, you know, six months of data before.

You've got information that's coming in on the fly that can be used to embellish

forecasts, and that's kind of, what the essence of, of the program is.

To kind of evaluate different methods, but also give, the manager or the executive,

the tools to be able to evaluate the, the effectiveness of different forecasts,

be able to compare and contrast them, on their prediction capabilities, and

kind of frame it in the context of how LLMs play a role, in, in forecasting.

Yeah, I think i-in my mind, you know, a, a big benefit here

and a big piece of it is n-not just the ability to pull in so much more of the

information data that's available to you, but the real-time nature of it, because

I feel like especially in, in retail for most his, organization historically,

you know, real time meant how fast can you put it into the spreadsheet, right?

To get an answer in your forecast model to give to a manager somewhere.

And that might mean it was days of effort of you manually putting things in, right?

And updating information or going and asking further into the org

for the next piece of information.

Well, now you're, you're kind of saying, well, you can set up an agent to go and

collect this for you and pull it all together, you know, and give you more

of that real-time capability that you wouldn't have been able to keep up with.

Absolutely.

E- exactly

So the other area you talked about was inventory

management as being a really crucial element in operations.

This is one of the ones I really like talking about, because it's just become

so much m- more important for every, every retail, right, to get their inventory

management right, to get it accurate, and to really make that connection

between what they're forecasting, what the inventory is today, so that they're

not having to face themselves with empty shelves and unhappy customers who

aren't getting what they want from them.

So how is AI going to help with this?

It, it's a great point.

I mean, inventory has, you know, there's a lot of money tied up in inventory.

And we talked about, you know, building these agents from day one

with the procurement and sourcing

Right

management.

And as we kind of move down, the supply chain, we've talked about, forecasting,

and now we are-- we're, we're talking about inventory, and they're all

kinda connected together, right?

based on your forecast, that kinda ties into the decisions that you make on

safety stock and so on and so forth.

I think the, the real value here, that AI brings, is moving away again from

the static reorder points and monthly forecasts that we used to do, right, to

something that's more real time, right?

Something that's really where you can make these decisions quickly.

And, the value has been tremendous.

I mean, a lot of companies have used AI, and you see a reduction in stock

outs, reduction in inventory costs.

I think Amazon, published something, I wanna say somewhere last year,

where they said they had a thirty-five percent reduction in stock outs

because of some of the inventory-- or some of the AI tools that they use

Yeah

with predictive inventory.

The same thing with Unilever, pulling in data directly out of, Walmart SKUs, right?

and then think of the, the, the fast fashion industry with Zara, right?

There's a lot of these, examples of, how AI helps with inventory management.

But I think at the, at the essence of what it is, is that, you

know, you have this ability, to dynamically look at safety stocks.

And the way you do it is these agents are able to kind of sense

when there's a demand spike, right?

"Hey, there's a demand spike in the market now.

Can I use the agent and automated workflows to perhaps, raise a

purchase order because I need more material coming in to support

Mm-hmm.

Right

oh, by the way, while it's creating a purchase order, it's

also looking at the contract with that supplier, to make sure that

we're not violating any budgetary restrictions, terms, and conditions.

And if it's something that's, pretty routine, that purchase

order goes out automatically.

But, if it's something that's, you know, a little more important, of course, now it's

where the human intervention comes in.

and I think that's, that's really the, the big value, right?

But the other thing that I do wanna talk about is I mean, this, this is probably

90% of the, the challenges with all these tactical decisions that we, we

make on a daily, weekly, monthly basis.

But, strategically, this notion of using AI, with, what we would call, you know,

multi-echelon, inventory optimization, and that's, in essence basically saying that,

it's not important to only know what your safety stocks are, but where they should

be held within that supply network, right?

Because where it's held, matters,

Right.

how quickly you can deliver, how much it

Yeah

So part of what we would try to do is to, is to also dabble into

that whole strategic decision-making, with, with discussions on, on

multi-echelon inventory optimization.

Well, I can-- I can tell that's gonna be a very

in- a very useful day, in the, in the, out of the three days.

I, I, I would love to be part, there to be able to listen to

the inventory management piece.

So Brett, thank you, for episode two.

Two down, one episode to go

I look forward, to the next one

Ruffles has made an entrance and took a folder.

I'm still working on being puppy-proofed.

Next episode, Ruffles will be back down for a nap, and we will

finish the map with logistics.

So routing, network design, control towers, last mile, and what can AI do

for you when a shipment goes sideways?

Hmm.

Yeah.

And then, then we'll widen the lens a bit to, to the part every executive I

think asks about eventually to every one of their supply chain managers.

You know, the, what's the infrastructure, the governance, and the ROI case that you

have to make before anything gets funded?

I need money

Always.

We'll also talk about procurement, operations and logistics,

and how they come together as one strategy instead of three siloed projects.

That's the whole point of the series.

Yeah, so that will be part three next, coming up, next week.

Follow the show and, don't forget to check the show notes for ASU Executive

Program registration link if you're interested for the full course, and of

course, to use our listener discount code.

But act fast because seats are limited.

Ricardo, this episode is a wrap

We'll see you all for part three!

Love this episode?

Drop us a five-star rating and review on Apple Podcasts, Spotify, or Goodpods.

And if you're watching on YouTube, like and subscribe before you go.

I'm Casey Golden.

Follow us on LinkedIn, Bluesky, Threads, and Instagram,

and subscribe to our Substack for highlights and bonus content.

For transcripts and guest info, visit retailrazor.com.

I'm Ricardo Belmar

Thanks for joining us on the Retail Razor Data Blades, part

of the Retail Razor podcast network

Until next time, stay sharp, be data-driven, and harness AI.

This is the Retail Razor Data Blades!