Every Forecast Needs a Why

Not Your Mother’s demand planning lead on why explainability is crucial in AI planning

A new Not Your Mother’s item was set to launch at Walmart in 2,000 stores. The award came back at 1,000. Eric Falkenmayer knew what came next: the go-forward forecast would drop, and his supply team would ask him why.

Every demand planner gets some version of that question each month, and the answer has to hold up SKU by SKU. For Eric, Senior Manager of Demand Planning at Not Your Mother’s, answering it is the part of the job that changed most after the company moved its forecasting to Omnifold.

Eric talked through that change with Ben Van Dyke, Omnifold’s VP of Applied AI, at the Gartner® Supply Chain Planning Summit in Denver in November. Their session was called “If I Wanted the Same Forecast, I’d Have Kept the Old System.” What follows is his firsthand account of what changes for a planner when you can ask the forecast why.

Not Your Mother’s is a masstige haircare brand with more than $250 million in revenue. Eric’s team forecasts about 300 SKUs across mass, drug, grocery and beauty retailers, from Walmart and Target to Ulta, CVS and Walgreens.

This conversation has been condensed and edited for length and clarity.

Why did Not Your Mother’s change how it forecasts?

Ben: You’ve worked with a lot of planning systems in your career, and you already had one. Why change, and why now?

Eric: When AI started coming into planning, demand planning was going to be one of the first things it tackled, whether we liked it or not. So either we adapt or we get left behind. And demand planners, me included, aren’t known for liking change.

The old system worked. It was good at forecasting our A and B SKUs. It wasn’t fantastic at forecasting our tail or at discontinuing product. That gap mattered, because innovation drives our growth now. For a long stretch we grew by getting new distribution on our core products. Now we depend on new products coming into the market, so a better new product forecast is a huge help.

My VP of supply chain sponsored it. He said he’d handle getting it signed up, and he just wanted me to play around with it and tell him how it worked.

Why does it matter which system explains the forecast?

Ben: You’ve said the biggest change for you is being able to tell the story behind the numbers. What makes that possible?

Eric: Being able to tell the story to leadership is probably the biggest thing about Omnifold. Before, I’d show them the numbers, and they’d say, that’s great, but why is the forecast up? Why is it down? What changed?

What’s different about Omnifold compared with demand planning systems that have an LLM on top is that Omnifold is the system that makes the forecast. When you ask it a question, you’re asking the same system that made the forecast. An LLM on top of another system is analyzing a statistical forecast that came from someplace else and guessing at how it got there.

Ben: So you get richer explainability and a better understanding of the number.

Eric: Exactly.

How do you explain a big forecast change to supply?

Ben: How does that play out with supply and the rest of S&OP?

Eric: Supply will say, Eric, my team’s going to want to know why these SKUs changed so much from the last forecast. Some of the reasons are legitimate, like that Walmart item that went from 2,000 stores to 1,000.

Now I can have Omnifold build a table of the largest variances from the last forecast, with a column that tells the story of why each one changed. I’ve been asked to do that for a very long time, and I’ve never been able to until now.

It mattered most during the migration. A forecast built in your old system can look completely different in the new one, and the last thing I wanted was to send a big broken signal to supply. So I had Omnifold show me the biggest month-over-month changes and explain why each SKU changed.

How does retailer inventory data change the forecast?

Ben: What can you do now that you always wanted to do but didn’t have the capacity for?

Eric: We get a lot of information from our customers, and it’s very hard to process it all and decide what to do with it.

We know Target targets six weeks of supply for their A SKUs and eight weeks for their B SKUs. Omnifold looks at how much inventory they have on hand against the forecast and will say, they look a bit heavy on this SKU, so I’m bringing your forecast below the consumption rate at that customer, because they’re going to drain. Coming into Christmas, it knows they’ll order more than they sell, and you can expect that increase.

I’ve never been able to do that with any other system, and it’s a huge help, because it lets me tell management why the forecast is up or down. Those are things I always knew would be great to do, but there’s not enough time in the day to do them with all the information that’s available.

How does knowledge from outside the system get into the forecast?

Ben: A lot of what planners know never makes it into a system. Have you been able to bring that in?

Eric: It’s really easy. We hold what we call feeder meetings with marketing, the commercial team, planning, supply and demand to hear what’s going on in the business. AI is great, but it doesn’t know what it doesn’t know. I need the commercial group to tell me what’s happening.

I record those meetings with AI and paste the transcripts into Omnifold. It turns that information into what Omnifold calls Enhancements, which are changes from the base AI forecast. If someone tells me we’re running a BOGO at Publix in August, that’s all I need to know. It looks at our BOGOs from prior years and applies that lift to August.

It’ll even point out that last year’s BOGO was in July, and ask whether I want it to take the July lift and put it into August this year. I say yes, and it’s done. It’s also very good at getting seasonality right from the start, just from your history.

Before, I’d look up last year’s lift at Publix, create a new event in August and type in a number. That sounds simple for one event, but across a whole bunch of customer groups and sales teams it adds up. So it’s a huge time savings, and a better forecast, because it’s analyzing how the item did around the same time last year.

How do you keep optimistic numbers out of the forecast?

Ben: Where are the next opportunities for your team?

Eric: We just implemented one. Nielsen doesn’t know your item’s distribution, but it knows how many stores are scanning it. Omnifold understands that for lower-volume SKUs, the store scan will be way below your actual distribution, while for high-velocity SKUs the two are very close.

That matters if you’ve had a lot of store expansions. Most statistical forecasts would see a clear upward trend. But the reason for that steep path upward is the added stores. Omnifold knows that, levels out your forecast and doesn’t shoot it into outer space. That’s a huge thing.

The other is the commercial group’s projections, which sometimes run a bit high. I take the spreadsheet they fill in with units per store per week, our velocity metric, and check it against Nielsen data for our SKUs and for competitive SKUs. If they expect three units per store per week at Target, it’ll flag that most projections look right, but this one is probably going to be 1.2. Those self-checks are an enormous help.

What does a planning month look like now?

Ben: What’s changed most in your day to day?

Eric: The biggest change for most demand planners is that you lose the tables. The statistical forecast and the puts and takes underneath it still exist, but you don’t forecast in them anymore. In my 15-plus years in demand planning, that’s probably the biggest change.

Most of the planning happens in the chat interface. If a product’s consumption at the store is here and my forecast is way up there, I used to calculate the percent difference, do the math and change all the boxes. Now I say, peg this SKU to the POS, and it does it. I don’t type numbers into tables, which saves time and keeps you from making key mistakes.

On the first of the month, once the prior month is actualized, Omnifold gives me a synopsis: the rest-of-year forecast changed by this much, and here are the main changes it made on its own. I can ask why it made any of them, and it explains.

Then if I want to tackle Target, I ask what needs my attention. It’ll tell me POS has dropped on an item for the last two months. Shipments still look stable, but it’s only a matter of time before they drop too. Then it checks weeks of supply at Target, finds it’s too high, and tells me orders will probably come in below consumption next month. That review used to take me a couple of hours for Walmart and Target. With Omnifold it takes maybe 20 minutes.

S&OP is easier too. I gave Omnifold the tables and charts management was used to seeing, it asked a couple of questions to understand the metrics, and now it populates them from the system. What used to be an archaic Excel shell is the press of a button. And with all my customers in Omnifold, I go to one system for any analysis, stories included.

What would this analysis take without AI?

Ben: If you tried to do all of this the old way, what would it take?

Eric: When it knows Target is going to drain for the next three months and takes the forecast below POS, or knows they’ll order more than they sell coming into Christmas, my demand planner and I just looked at each other. This would take us forever. I would need three or four demand planners to get that same level of accuracy.

The planners we have will still be planning. Their jobs will be different, with less of the monotony of building pivot tables and trying to figure out the story yourself. You spend that time explaining to leadership or supply why something is happening. The whys are what always get us in demand planning. We can give the numbers all day long, and then it’s, okay, why this number, especially if it’s one they don’t like. Being able to tell that story changes your job. You spend more time on the stuff that really matters.

How accurate is the forecast so far?

Ben: How are you judging whether it’s working?

Eric: It’s a little early to say exactly what our accuracy will be. I just finished migrating all my customers this past month, so I don’t have enough lag yet.

But we did some analysis. I compared the base statistical forecast and the adjusted forecast from my old system with the base forecast in Omnifold, since that’s all we had at the time. Omnifold was 20% more accurate using just its base forecast versus the base forecast from my prior system. And it was 10% more accurate than the forecast with the adjustments I made in the old system. With nothing done in Omnifold, it was still more accurate than the old system after my changes. That was very eye-opening for us.

What should a planning team expect in its first few months?

Ben: What would you tell someone starting where you were a year ago?

Eric: Give it a fair shot. It’s a big change from how I’ve planned for the past 15 years. A demand planner on my team has almost as much experience as I do, and we were both skeptical. Really? There’s a little flashing cursor, and I’m supposed to do all my demand planning in that? And it actually works really well. You start trusting it little by little, and then it’s, hey, this actually worked.

Ben: What skills should teams build?

Eric: The biggest is putting the right prompts in so it understands. Context is important. I spent the first couple of months adding our tribal knowledge, which is part of why the migration took a while. Omnifold doesn’t know how your business runs at first. Every time you tell it something, it creates a memory card for that business knowledge, and then it has a better idea of what you’re driving at. So when you’re starting, be very specific in your prompts, and make sure it’s creating those memories correctly.

It can make mistakes. It’s not perfect. I uploaded a receipt plan for one customer, where the date is when the customer places the order, and the order might ship two weeks to a month later. Omnifold didn’t know that and read it as the ship date. Once I told it, it created a knowledge card, and I never have to tell it again. It just knows.

It’s a learning process, and teaching it your business takes the longest. Once we had that, I migrated customers in stages, starting with Walmart and Target, checked that the forecasts looked right and kept going. It’s been great.

About the speakers

Eric Falkenmayer is Senior Manager of Demand Planning at Not Your Mother’s, where he has spent the past six years. He has close to 20 years of planning experience, beginning with about seven years in assortment planning at Bed Bath & Beyond, followed by demand planning roles at Marchon Eyewear and SheaMoisture. LinkedIn

Ben Van Dyke is VP of Applied AI at Omnifold. Before joining Omnifold, he was an executive in BCG X’s supply chain practice. LinkedIn

Want to learn more?

For more on Not Your Mother’s, read the case study, or What Not Your Mother’s Haircare Is Prioritizing in 2026, from Eric’s webinar with Omnifold’s Garrett Long. Omnifold builds reasoning AI for supply chain planning. To see how it explains the forecasts it makes, connect with us.

GARTNER is a trademark of Gartner, Inc. and its affiliates.

FREE WHITEPAPER

The next leap in supply chain AI isn’t automation.

It’s better decisions.

From Omnifold’s researchers (OpenAI, Google, MIT, Stanford): why history-only forecasts break, and how reasoning AI plans for what history can’t show.

HubSpot form loads in Preview and Published site.