Mastering Forecasting and in-Stock Availability

Paul Sonneveld
I'm your host, Paul Sonneveld, and today we’re tackling one of the most critical—but often misunderstood—disciplines for Amazon vendors, namely, forecasting. From anticipating your true sell-out to planning for Amazon’s notoriously opaque purchase orders, to handling unexpected surges from influencers or viral moments—forecasting on 1P can feel like flying blind.

That’s why I’m thrilled to be joined by Justin McMillan, co-founder of Rebelution, a vertically integrated Amazon 1P agency out of Chicago. Justin and his team help vendors build smarter, more resilient operations—and forecasting is a key part of that playbook. Justin, welcome to the show—it’s great to have you here!

Justin McMillan
Thanks for the invite, Paul. Looking forward to it.

Paul Sonneveld
Well, let's jump right in. Actually, let's not. Before I do that, I do want to remind our audience that this is a live session. So please, please post your questions in the LinkedIn comments section, and I'll do my best to get them in front of Justin during the course of this episode. So just an encouragement there. Okay, back to you, Justin. Tell us a little bit about your work at Rebelution, the kind of vendors you support, and particularly, I'm really curious, because I don't hear this often, how does forecasting really play into your service offering?

Justin McMillan
Yeah, at Rebelution, we work with a pretty broad range of both consumer and industrial commercial brands. We like to take on a mix of both emerging startups with some established players, but the vast majority of which exist on the Amazon 1P platform, Vendor Central. We feel that our value add in this space is to really kind of bridge the gap between what vendors expect, what Amazon actually does, and then kind of bring our proficiencies in there and plug and play. Helping vendors to forecast is a huge part of that. It's especially challenging with Amazon, as most people who work directly with them know. But it has absolutely been a big part of our day-to-day business.

Paul Sonneveld
So you mentioned forecasting challenging in the one sentence there. I mean, what makes forecasting in the Amazon 1P environment so uniquely challenging?

Justin McMillan
Yeah. One of the big pieces is Amazon doesn't share a traditional demand plan with vendors. For example, a traditional distribution partner is going to place POs months in advance. They're going to give you advanced purchase orders. You're going to operate on a very consistent demand planning forecast, whereas Amazon really operates on just-in-time purchase orders. They want to ensure that they have just enough inventory on hand not to get into over-inventory positions, essentially de-risking their overall business. 

And then also, unlike traditional distribution, Amazon could at times could be hedging against potential lost buy box issues, maybe some competitive pressure that you're seeing in the space. You might not be getting the same traffic that you were week over week. There could be a myriad of reasons that they're managing that different demand at different times. And you really are flying blind to some degree. You're forecasting on sellout, but you're getting paid on buy-in. And those two absolutely don't always align. And then mix that with system-driven constraints, maybe some retail readiness issues, and it becomes a big puzzle.

Paul Sonneveld
Yeah. Hopefully, we'll uncover some of it today, but it's certainly, it's a major headache. I mean, so that we probably get one or two of our customers, both events and agencies ask every, how about some solid forecasting module in your platform? And it's kind of easier said than done. Absolutely. Where do you start? So, you said sort of, you mentioned sort of sellout. When you're forecasting sellout, I mean, where do you start? What are the inputs you rely on?

Justin McMillan
Yeah, we really try and start with as much historical data as we can, right? You want to start 12 to 24 months, if possible. And then you want to layer in some, some x factors, you want to layer in advertising, seasonality, promotional plans, maybe some competitive trends, external media spend, social media spend, as many as you can manage, without it being too many. But, you know, Amazon's algorithm takes into account, uh, a lot of things, you know, most of which is traffic-driven, conversion rate-driven.

Those are the big two pieces that if you're trying to stay on top of, if you're trying to stay on top of how much traffic you're driving, as well as what your conversion rates are looking like, you can, you can paint a pretty good picture.But we also want to talk directly to the brand, right? They have a lot of visibility sometimes in the upcoming shifts that aren't necessarily in Amazon's data. Maybe there's some distribution changes or again, right, some planned off Amazon campaigns.

Paul Sonneveld
That makes sense. In terms of the exact sellout piece, I mean, maybe I'm getting too technical here, but are you looking at ordered revenue, ship revenue, sourcing, manufacturing? I mean, what's your sort of go-to baseline here?

Justin McMillan
Yeah, typically we're looking at COGS. So if you're looking at unit COGS on a rolling 12 month basis, you're trying to establish not only that baseline, but then you're also layering in growth rates of each individual ASIN, you're layering in if there's some strategic ad spend, you're layering in some strategic promotional spend. 

Oftentimes Amazon's gonna order well ahead of promotional seasonality. So if there's a marquee event coming up, Amazon's going to pull an additional inventory ahead of that. A lot of it has to do with talking with all kind of facets of your internal organisation, as well as whichever tool that you're using, so that you can get as good of a picture as possible. But start with COGS, look at growth, focus on units, and that can kind of get you a good start on it.

Paul Sonneveld
Yeah, so you touched on that. Obviously, that gives you the sellout forecast. Actually, just as an aside, when you go through that exercise, how closely aligned is it to Amazon's forecasts? You know, the P70, P80, P90. I mean, I'm just curious. I mean, because I expect the answer to be way off, but I'm just kind of interested, like when you actually have a proper comparison, what are the variances between what kind of you believe the best estimate of the sellout equation is versus what Amazon's, the piece that Amazon's given you?

Justin McMillan
Yeah, it depends. That's a big part of the reason why they don't commit to purchase orders, right? They're not going to give you that exact number because even that's going to change. From how consistent is that is that fulfillment rate against the P70, P80? Those are all probability algorithms, right? It's a probability, 70 probability, probability forecast and the challenge is that it's going to change every day. 

We've had, we've had folks that have had really great external marketing pushes that have driven that number up within a week, two weeks. And that affects the rest of the year because it's all, it's all really reactionary to a moment in time. But over time,  it's going to change. Amazon's never going to put a hard line, a hard historical kind of, match in place on that. But you're going to bounce between P 70, P 80, P 90. And it's going to change. Sometimes you'll be up 10%. Sometimes you'll be down 10%. Sometimes you could be completely off. And then the other layer behind that is that it's an ASIN level basis. So there could be instances where you're a hundred percent off on a probability estimate that they're giving you because you're growing the heck out of a certain product.

Paul Sonneveld
Yeah. Yeah. Yeah. At the total level, it looks reasonable, but the total level is not that useful in this situation. So yeah, no, I get it. So let's say we have our, um, sellout forecast. How do you go about translating that into a sell-in forecast or a purchase order forecast. I mean, that's kind of what we're interested in, right? So what are the steps you'd take there?

Justin McMillan
Yeah, what we'll take is we're not trying to base the recurrence necessarily on when Amazon is going to place a purchase order. There's a lot of historical contexts behind the scenes, whether it's a bulk buy, whether it's a marquee event that was an anniversary. That's going to change when Amazon is going to do that pull-in. There's also some black box activities that you're just not going to see. You're not going to see if they're having FC constraints.

You're not going to see if they're over-inventory because they overbought on a completely different vendor in the same space. If you're in a different category that requires a specific fulfillment center that you service out of like the hazmat facilities, there could be an instance where the hazmat facilities over-inventory on a product and unfortunately, you can't take up space. When it comes to marquee events again, Amazon is going to prioritise outbound shipments.

During those weeks, if you didn't participate, that could be another hindrance to how much inventory that they're going to bring in at that time because you're going to get both barrels with not having inventory coming ahead of time and not being able to get inbound shipments in when they're shipping out.

Paul Sonneveld
Is there any way to discover some of those signals? How much of the contextual information can you get out of your Amazon Vendor Manager? Or are there different things you can pick up? Or is there a way that Amazon communicates some of these things, ideally a little bit more in advance?

Justin McMillan
Yeah. Unless you've got a really good relationship with your Vendor Manager, you're not necessarily going to see some of those behind-the-scenes pieces of information, right? Depending on the acumen of that individual, how connected that they might be, right, maybe they haven't even heard of some of those instances. Maybe they're not necessarily being broadcast or shared internally. 

And then the challenge, too, with vendor managers, they're not necessarily tasked with forecasting, right? They're not tasked with that forecasting. They're really looking at optimising guardrails within top-line growth, profitability, driving the business, growing the catalogue for that brand. If you're a large enough vendor, you might have an opportunity to talk to an in-stock manager, and those folks are more closely aligned with PO placement and forecasting and such. But still, you're rarely gonna get a committed buyer or get that kind of hard committed purchase order run rate that you're looking for in that scenario.

Paul Sonneveld
So on that topic then, are there any other specific indicators that you tend to look for? And I'm thinking like really broadly here, but that can be used just to refine or predict that purchase order volume a little bit more accurately. Any kind of, sort of connecting to the dots and trying to translate it into a forecast. 

Justin McMillan
No, this might not sound like the most popular answer, but really in this space, it's not about perfection. It's about agility. The challenge with this scenario is you're not getting into a situation where you know, come this particular date, this particular week, you're going to get X in purchase orders. 

What you're operating off of is that sell-through and that sell-through is going to give you an idea of when you can expect a purchase order, right? So if you're tracking sell-through velocity, weeks of cover, and on-hand inventory, if you're running somewhere under four weeks, and sales are healthy, or sales are growing, there's a good chance appeal is coming, right? You're looking for lead times, you're looking for inbound flow, but at the end of the day, you're really working on how agile you can be with safety stock or speed to getting the product out the door when the purchase orders come in based on that velocity.

Paul Sonneveld
That makes sense. And we sort of shift more in terms of the unpredictability, right? Because forecasting is one thing, but I'm sure the actual purchase orders or the actual demand versus the forecast is always off by plus or minus a big percentage, probably the double digit at least. And that's just because things happen. So how do you… Well, actually, my first question is around like safety stock at your client's warehouse, right? What sort of contingency are you, and we'll get into the broader kind of contingencies and how you react to this, but from a safety stock point of view, what advice or what are you asking your clients? You know, we've got the forecast, we're feeling confident, it's as best that we can produce it. But then, you know, what's the safety buffer that we want to layer on top of that from an in-stock point of view in their warehouse to do some of that?

Justin McMillan
Yeah, from a safety SOC standpoint, we're advising somewhere between three and four weeks of safety SOC, right, beyond forecast. That's with the consideration that you've got your pipeline filled, right, you've got the forecast in place, you're building, you're shipping, it's on the water, it's incoming, but you've got on hand three to four weeks. That, in most instances, understandably, right, it's very different for each brand and very different for each product. 

But in those instances, you know, as you're growing, you're able to support a spike in demand, you're able to pivot as those changes in velocity count. As far as kind of building that safety stack to, you're looking to also support with direct fulfillment, you're looking to enable that strategically. Granted, there may be occasions where you run through what Amazon has on hand, and they're not picking up that safety stock fast enough. Maybe you have an amazing week, and you just absolutely move through all the inventory that they have on hand. 

Having that direct fulfillment as a backup, that's massively important. You can have it on at both times and run a hybrid model on direct fulfillment in network, but having the ability to quickly turn that on in instances like that can bridge the gap between the vendor lead time, you know, that occurs, happens to be anywhere between 14 and 28 days. So between 14 and 28 days from purchase order placement to when Amazon's going to get the product, depending on the time of the year, depending if it's a marquee event, all that kind of stuff. It could take 28 days from when that purchase order comes through to make it to Amazon. 

Direct fulfillment as a kind of a safety stock backup is a big deal. It keeps you agile. It keeps you prepared for those instances, not only from a highly trafficked merchandising event, but also every day. You know, in that too, with that safety stock, work with your team, work internally, change that safety stock on some type of rigour. Find some type of repeatable rigour where you can reevaluate where your safety stock's at. If you're moving through it, if it's kind of stagnant, just to make sure that you're keeping pace with Amazon sell-through and you're using that as your sort of true north with where you need to be.

Paul Sonneveld
Yeah. I understand. So just want to throw in two follow-up questions there. First of all, that scenario you talked about, but maybe Amazon is a little bit, maybe Amazon's not on the ball, you know, days cover inside Amazon's warehouses running down. You're running at the real risk of running out of stock. You know, you've spoken about kind of contingency options there, but is there anything you can do? I mean, should you pick up the phone to the vendor manager and say, Hey, you need to purchase, place a purchase order today? I mean, is there an escalation process to get Amazon to order more?

Justin McMillan
Yes, yeah, absolutely. I would say that the escalation process is any lever that you can pull on in instances like that, pull on it, right? Never hesitate, never get down to the 11th hour and then pick up the phone and make that call. Always feel obligated to ask. With that same understanding, though, like we were kind of discussing previously, there could be external factors affecting that. And I don't mean to harp on direct fulfillment as a solution. But there could be external factors that are making it difficult or preventing Amazon's order algorithm from pulling that inventory in. 

But from an escalation path, absolutely email, pick up the phone, submit a case, submit two cases, contact whoever you want to contact to get information on it, to make sure that it's not something systemically wrong with that product or systemically wrong with the account. Depending on the size of your team, there could be a lot of different things that can affect an account at different times. You never want to wait until the last minute to determine if it's something serious or if it's just a constraint that's going to work itself out.

Paul Sonneveld
Yeah, look, Amazon may be constrained, like you were saying before, around their warehousing capacity or they just can't bring in more inventory there. My second question was around direct fulfillment versus hybrid. Now, this is a really big topic. You and I can probably do another two episodes just on these things. 

Actually, I'm doing one on direct fulfillment in a few weeks. But just very briefly, setting up direct fulfillment, obviously, you know, some effort involved, and you need to have the capability to do essentially DTC shipping, different kind of warehousing setup, obviously. How do you think about the pros and cons very briefly on those things? All right, because traditionally, I've seen more and more vendors talk about, you know, hybrid is easier because, you know, you can just throw an offer up there. But how do you, why do you think Vendor Direct Fulfillment is a better option, or is it?

Justin McMillan
Yeah, it depends. Sometimes it's not. It really depends on the situation, depends on the brand's fulfillment abilities, how agile they are in that sense. And I do want to clarify that when we say hybrid, right, we're talking about direct fulfillment and in-network 1PO only. It's, right, not my, not necessarily my recommendation to split that up between 1P and 3P, but specifically with just 1P, hybrid would mean direct fulfillment and in-network. 

The benefits of direct fulfillment in network is there's opportunities where you can kind of supplement Amazon's inventory. You can supplement how much you can feasibly ship in a day to make sure that Amazon doesn't run out of stock. But granted, again, right, direct fulfillment doesn't necessarily work for everybody. It can be heavy lift for some. it could be low lift for others, and it can be preferred for others. Depending on the business model, depending on, you know, how efficient your warehouse is with shipping singles versus how efficient your warehouse is with shipping truckloads and pallets and such. It really depends on how well it fits your model. 

The only heavy pitch I would make for direct fulfillment would come at times of peak seasons. So if you're an in-network vendor, you know, traditionally, and you want to stay away from direct fulfillment. When it comes to a peak season, when it comes to, you know, Prime Day, back to school, back to business, you know, Black Friday, Cyber Monday, Q4, all that kind of stuff. Those are the opportunities where I would heavily encourage people to look at an opportunity of setting up maybe a temporary timeframe to do direct fulfillment as a supportive means in those events.

Not only is it beneficial for folks who are participating in a marquee event, but it's also beneficial for people who weren't participating in a marquee event and maybe didn't get fulfillment center space leading up to it. So now you're squeezed in your overall footprint for inventory because those that were participating are eating up that space to make sure that they can, Amazon can fulfill those promotional orders. The other side of that is too, that ensures that if you've got those promotions going on, you won't run out of the stock that they brought in. And if they start to get constraints with outbound shipping, you can support your deals, you can support your orders more effectively.

Paul Sonneveld
Makes sense. I just want to acknowledge that we have a lot of questions from the audience here. I'm probably going to throw you one or two of my own, and then we'll open up. So I just want to say thanks, guys, for keep posting your questions. I am seeing them. We'll get to them very shortly. So Justin, I just want to finish off on my side here before we open up on tools and tactics. First question is really around the tools of forecasting models. Beyond the G Sheet and Excel, is it as simple as that and just use your own logic? Or are there more sophisticated tools that you'd like to leverage when it comes to this topic?

Justin McMillan
Yeah. First off, there's nothing wrong with Excel. If that works for you, use it. It's really great for scenario planning. Flexible with how you can build it, you can do a lot with it. There's nothing wrong with using Excel. There's a lot of great tools out there without name dropping anybody and getting in trouble with folks that we work with. All the big software platforms have really good forecasting tools. There's a lot out there. We have our forecasting abilities as well. 

Whatever you choose, wherever you go with it, just avoid any tools that are almost entirely black box. You want to be able to challenge the assumptions. You want to be able to get into the details and do a little bit of modulating. You want to be able to play around with the numbers. You want to understand what kind of X factors are going to make a big difference based on how you're operating, based on different KPIs.

Paul Sonneveld
Very helpful. I do love a bit of Excel. Actually, I'm probably more Google Sheets fan these days, but for different reasons. But yeah, you're absolutely right. Thank you for that. Last one from me. Most common mistake. What is the most common mistake you see vendors make when forecasting sell-in or sell-out on their vendor platform?

Justin McMillan
Yeah. Don't operate in a silo. Whatever you do, don't operate in a silo. come up with kind of a think tank team, come up with an internal initiative team where you've got marketing ops, procurement, whomever has some form of influence over your stock and your inventory position, anybody that has some type of influence in that. Get them involved, right? Get what's going on from their side.

Have an understanding if there's going to be a social media post that's been, you know, in plan for three months, and it's gonna make a huge splash. But you weren't a part of that decision-making, or you weren't a part of that, that kind of initiative, just get people from different departments involved. Find some type of operating rhythm to work on. And then I would say that the second big one is make sure that your reviewing your numbers, right? There's going to be times throughout the year, every year with Amazon, where there's highs and lows. And while you're going to experience a pretty consistent growth in most instances across that year, just continually reanalyse. There's going to be a lot of dynamicism on the platform for both internal, external factors, competition, marketplace sellers, you name it. Make sure that you're reevaluating your forecasts on some type of rigour as well.

Paul Sonneveld
Thank you. That's great advice. OK, let me go to our audience. I'm going to go to Conor's question first. Thank you, Conor. He raised this when we were talking about sort of events happening and Amazon adjusting their sellout forecast, the P70 to P90 one. His question is, how quickly does Amazon's forecast really react to some of these trends?

Justin McMillan
That's a great question. I don't know the exact answer, but I can tell you that if I pull it from day to day, it will change.

Paul Sonneveld
Thank you. So that's quite likely. I guess we don't know whether they're reacting to a change from last week or this week, but that's an honest answer. Thank you. Let's go to Mike's question. Thanks, Mike. I'll read it out. It's quite long. If you have a regular and recurring delta between Amazon's placed purchased order quantity and your confirmed quantity, which historic data point do you weigh over the other? What you actually sold in 1P or what you could have sold if confirmed in full? So this goes to the question, what was your true demand signal? You know, requested or confirmed?

Justin McMillan
That is a fantastic question. To those that are initiated, that's a fantastic question. So the challenge is, is that if Amazon's ordering a million dollars from you each week and you're only fulfilling 500k, is your demand four million in a month or is it somewhere in between? What we have seen is that it's somewhere in between. What will happen is, is that there will be instances in which Amazon is trying to reorder what they've ordered. But there could be instances where the growth is so strong that it's on the higher end of what they're ordering. 

So as another example, if that product is growing 100% month over month and Amazon is just dying to get that inventory, you could be missing out on lost sales because your offer is completely falling out of the buy box, right? It's becoming temporarily unavailable. There could be instances where that's going on. But on the other ladder, the flip side of that, there could be instances where Amazon is just trying to get to a healthy inventory level. And they're just ordering more and more each week to get to that, you know, you know, one to two months' worth of inventory on hand for them that they would consider healthy.

Paul Sonneveld
Thank you. I'm switching to Guillaume. He's submitting a couple of questions via YouTube. Let's get on to the first one. Two questions. Bulk buys. Is it worth it, or will the POs come later anyway? He's got a second question in another card, so I'll get to that in a sec.

Justin McMillan
Yeah, it's a hot topic.That's another one that it, it depends. There's instances that we've seen that bulk buys have driven growth, uh, primarily because they're, they're contributing to, to PPM, uh, metrics. Um, right. You're giving Amazon an incentive. You're feeding the algorithm to get them to order more because you're giving them a more profitable offer. We've seen it where, you know, it contributes. We've also seen it where it hasn't. You're consistently growing maybe five, 10% year over year and signing up for a bulk buy is just really hitting your bottom line. 

I would encourage folks to test and learn in these instances. We've done tests and learns a lot because it can change from brand to brand. It can change from vendor to vendor category to category. There are instances where people haven't done bulk buys, they participated in them and then they started to see some growth. And then others, you know, we've done it a year-over-year analysis and participating in a bulk buy had little to no effect.

Paul Sonneveld
So I've just put up a second question as well, which is also quite interesting. I'm intrigued myself. We have heard, that is Guillaume and his team, of purchase order forecast reports, so not the sell-out forecast, but the sell-in forecast essentially, being available to some brands. Is this just a rumor, a myth? Have you come across anything on this front?

Justin McMillan
It is not a rumor, but it is not a completely open, for lack of a better term, it's not a completely open process or feature to Vendor Central. Amazon has some initiatives where they'll beta test on certain things, they'll open up initiatives, they'll close down initiatives. There's been a couple of different initiatives with forecasting and consistent purchase orders over the years. 

Sometimes they open up registration, sometimes they'll close registration. It can really depend. There may be the potential in the future that they do some of those things. There were times in the past, as well as now, where there might be a product that they consider absolutely a non-negotiable to have on hand at all times, and they'll get into contracts to to work on that, but it's very rare. 99% of what we're talking about really applies to most vendors.

Paul Sonneveld
Unlikely to see it in Vendor Central anytime soon, by the sounds of that.

Justin McMillan
Yeah, no, I can't make a commitment on that one. 

Paul Sonneveld
Yeah, no, I appreciate that. Okay, two final questions. Let's try and wrap this up. Narendra, thanks for your question. What about placing a BTR, I'm assuming Born to Run, order?

Justin McMillan
Yeah, Born to Runs, they can be a little bit tricky. Amazon has changed a couple things with them. Really, they're focused on bringing in new inventory. In the past, there was the opportunity to leverage them to put in additional inventory ahead of events. If you feel that Amazon is not ordering enough inventory for a certain time period, you could leverage the BTR system. They've restricted a lot of that ability lately, where you can't necessarily force as much as you want or use it for that same measure. It can be tricky to try and use that. What they have found too is that some people were using it as a means of inflating monthly numbers and that was causing challenges of its own.

Paul Sonneveld
Thank you. Last but not least, one from Nancy. I think it's a great one to close out on. It sort of takes us up a level again. How do you approach the conversation with client leadership about the unpredictable nature of the Amazon demand forecast? I'm sure there's you have to do a lot of explaining on this front at times. How do you approach that topic?

Justin McMillan
Yeah, that's a great one, Nancy. It constantly, I think the challenges with Amazon more so than others, is that it can be very unique. It can be very dynamic. It can be very difficult to kind of have an organisation, have a brand, have an ops team. A change of process that has worked for a long time, as well as get leadership to understand why it's so critical to support Amazon in a different way. 

As troubling of an answer as this might be, it is consistent messaging. It is not backing down in those conversations. You really have to be the ambassador of truth when it comes to it. As annoying as people might find it, I would say don't give an inch when you're trying to tell the truth on what it takes to do this stuff, because anything less than what you know to be the right answer is not going to be the solution you need.

Paul Sonneveld
Great advice, and it's a great note to finish on, Justin. Thanks for your question, Nancy, and thank you to all of our audience participants for asking us your very useful and very interesting questions today. Appreciate that. Now, that is a wrap of this episode of Marketplace Masters. Before we head off, Justin, I just want to thank you for really taking us inside of the world of Amazon 1P forecasting. Your perspective and experience is not only grounded really in that deep level of client experience, but incredibly actionable as well. 

So I know our vendor audience is walking away with really a new level of confidence around, you know, how can you really plan, prepare and do that forecasting piece? So for those of you watching live or catching the replay, we'll make sure to include Justin's contact info and links so that you can get hold of him in case you want to pick up this conversation or just go a bit deeper on some of these topics. I'm sure he's very open to that. Yeah. Thank you so much. Appreciate you being on the show today, Justin. Thank you.

Justin McMillan
Thank you. Same Paul. Thanks for the invite.

Paul Sonneveld
All right, everyone. That is a wrap of today's episode. Now, don't forget, you can find all of the past episodes and upcoming episodes on our website at merchantspring.io. And of course, if you're looking for better tools to monitor forecasts and optimise your Amazon vendor performance, of course, we'd love to show you what we're building here at merchantspring.io. I'm Paul Sonneveld. Thanks for tuning in and see you next time. Take care.