Name: Moby Live Session: Forecasting with Moby
Uploaded: 2026-09-23T16:43:05.047Z
Duration: 30 min 8 s
Description: Moby Live Session: Forecasting with Moby

Transcript for "Moby Live Session: Forecasting with Moby": Hey, everyone. Thanks for joining. What's up? What's up? Alright. We will give everyone a few minutes to join. We have a fun session packed for today. And, Ashley, yeah, feel free to throw in the chat where you're coming in from. Let us know too. Are you, there's a combination of, like, people that are triple l customers in here, also people that are, not triple l customers yet. So feel free to let us know. Customer in Miami. That's fun. I'm in Columbus. Alright. Amazing. Do we have anybody okay. We have somebody in The UK, Puerto Rico, or worldwide. Dylan's Moby's mister worldwide, basically. That's what I'm picking up. Alright. Awesome. Awesome. Dylan, do you wanna, kick it off for us? Yeah. Let's keep going. I'll screen share, and we can kinda dive into it. So, topic of the day, forecasting. Can you see this? Does this pull up? Yes, sir. I see here. Yeah. Excel. Perfect. Okay. Great. So for for most ecom brands, this is this is forecasting. You you create an Excel sheet with some previous order data. You come up with, you know, a few metrics that you wanna control or influence here. Here, we've got, hey, sessions, conversion rate, orders, and our AOV, and then we're gonna make an a revenue number based on that. And then we're gonna come up with some assumptions. So we're gonna say, hey. In September, we're gonna get 2% more traffic. In October, 4%. And then November for BFCM, you know, that's gonna be 20% conversion. It's gonna be 10% higher. And we're gonna say, you know, great. Here is our our forecast, for the next three months. There's no problem with this. It's just not the best thing that's possible out there to actually predict and forecast your revenue. It's lightweight. It's simple. It's easy to do. If you asked, say, ChatGPT, hey, help me forecast my revenue. This is basically what you're gonna get back, an Excel sheet with some assumptions based on numbers, and that is your revenue forecast. And so, you know, there are much better ways to do this, and Mobi has access to two what are essentially models, to do the revenue forecasting. So the first one is one called Times FM. This is a model from Google Research. There's a lot of nerdy science y stuff in here that not even I can explain. But these are full models built, not AI models, but actually predictive data science models to do forecasting. I'll kinda get into this a little bit in a second, this concept of univariate versus multivariate. But this is one of the models we have access to that is directly built into Moby is this model Times FM, and then another one from the Amazon research team called Kronos two. You do not have to know or care or worry about any of these models or how they work or what they do or how to use them. I'm just simply showing them because they're cool and science y and look interesting. If you wanna go just, like, read these pages after, we'll share them because it's just it's just super interesting if you're into data science at all. But the best piece is Mobi knows how these tools work, and it is fully connected to them. And so when you try to do forecasting in Mobi, it's going to decide which one of these tools is best. In Mobi, it's going to decide which one of these tools is best, and then it's going to pick the best one and use that. And so there's a concept I wanna get into, real quick. It talks about it just a little bit right here. Up until times 2.5 released in September 2025, our models were strictly limited to univariate forecasting. And then this model, which came out 08/31/2026, is actually a multivariate model. And so what does that mean? Univariate versus multivariate. Univariate is is essentially just you get one set of data. So a good example might be that you you use thirty months of order revenue data to estimate the next three months of of revenue data. And that's just univariate. You have one source of data in the example. For multivariate, we can actually include a bunch of different context and data points and, you know, pieces of of data to actually create our prediction here. So rather than just using historical order data, we could say, hey. Let's use order data. Let's use site data. Let's use ad spend data. And then we're also gonna put in our, you know, marketing calendar. We're gonna put in when we're running promotions. We're gonna put in as many different variables or variants as they call them here into the model to actually come up with your your forecast. So, again, very data science y. What's great is you don't have to worry about it. You basically just tell Mobi, hey. Here is what I'm trying to make a forecast for, and here is the data that I want to use. And so for this first one here, walking through the flow of how we actually create, a forecast, Step one is just to find the question, and we're just telling Mobi, hey. What do we wanna forecast? How much data do we wanna use, and how far ahead are we wanting to look? So an example question here is forecast monthly order revenue for the next three months using the last thirty months of history. Something that's really important is, like, let's say your store is two years old and you ask for thirty months of data like it is here, that six months of empty data is going to skew the results. So anytime you're giving Mobi a dataset of something to use, let's say, you know, a product was out of stock for four months and you're wanting to forecast the next, you know, twelve months of that product's purchases, always make sure to include that in of of, hey. We were out of stock, so exclude this dataset or whatever it is. Mobi will try to figure that out exclude this dataset or whatever it is. Moby will try to figure that out for you so you don't have to go and figure it out. But anywhere you have stale data, it's super important to just let Moby know, hey. This data is stale because, otherwise, it'll influence the results of the forecast. So let's come back up kinda to building it. First one, like I said, define your question. This is tell Mobi what you wanna forecast, how far ahead you wanna forecast, and then what data you wanted to use. Next thing is what I talked about. Check the history. So make sure all of your data is good. Make sure that you're not missing any any, time stamps. Mobi will pull the correct data that you tell it. Just make sure to give it any context that is necessary, and then tell it how you want the organization in the forecast to be done. So, hey. We want to get, you know, yearly revenue. We want to get an hourly snapshot for BFCM. We have a lot of people who have, you know, seven years of BFCM data, and they're super concerned about ad spend over Black Friday. And so they want an hourly, revenue forecast that they can then use to influence, hey. We're ahead of revenue forecast. Let's scale up our ad spend. We're behind. We might need to scale back. Mentioning here two models, Times FM and Kronos two. You don't have to worry about these. Moby will take care of it for you. And then here's kind of an example. So we have the input history, which is the thirty month window, and then the horizon, which is how far ahead we're wanting to look. Then we have the estimates here and then the comparison. The forecast will get much deeper, and Ethan can kinda show you those, and then we can walk through some questions after. But this is just kind of some background into how the product actually works from a a forecasting perspective and what tools we're actually using to do it and how it differs from just using a Google Sheet or even using ChatGPT where you're basically just going to get an output like this. That's essentially a Google Sheet with just some assumptions. So we'll hand it off to Ethan to actually walk through some examples of how this works. Incredible. Alright. Thank you, Dylan. It it looks like too we have a question, for Matt here that I'll address. If you've been in Triple l for only one year, can you only go back one year for your data? You you you should have more than a year of data. When you join Triple Whale, we upload, like, the historical, Shopify data as well, so you should have lots more than just one year of data that you can use to forecast. So check that out. But, yeah, if the the amount of data you have to forecast should not be limited by the amount of time you've been on triple l. However, for, like, pixel data, like, that will be limited to how long you've been on Triple l. But you should have all of the order revenue data history from your Shopify store. Awesome. So I'm gonna jump through a few really cool examples. And and for some more context too here, I guess Dylan mentioned, like, we had a we have a whole team, like, data science team building out these tools and and building out all of this. And there was actually a period of time too where we were considering, like, do we, like, build it all out into its own product? Like, how do we, like, bring these tools to life? And then we realized that it's actually just, like, so simple and beautiful and powerful to just build these into Mobi, and and that really just makes an intuitive, amazing way to build these forecasts. And I and and, Dylan, you actually probably don't know this, but I when I actually started my career several years back, when I was still, like, a analyst, I was building out, like, forecasts, every single day for it wasn't ecommerce. It was other stuff, but it was actually a like, some businesses, when they get to the scale of larger companies, it's like you have to use these models. You have to manually do all this, manually build the daily pacing. Everything's an Excel sheet, and it's actually just like a brutal process, to do all of this. And it's insane how simple Mobi makes it. So with that, let's jump in and, like, let's see let's see what Mobi actually looks like here and what we're talking about. Alright. So I'm gonna share, yeah, we're gonna share this window down here. Alright. Okay. So we're gonna start really simple first with an example. And and this shows, like, just how this is, like, the simplest example we can show. You can literally tell Mobi to build a monthly revenue forecast for the last six months. You can see here, like, I'm not even following the best practices for the prompt yet, but that's okay because out of the box, Mobi's gonna know that, okay. Like, I'm gonna analyze the last thirty six months of data. I'm gonna look at these metrics, and I'm gonna build that forecast. So it gets even better when you clarify, like, exactly what historical period you want and, and more. But you can see out of the box, Moby is gonna look at your last three years of data to build this forecast. And you can see here just from this simple question, Moby is gonna think for ten minutes. You know, it's gonna take a little bit. It's gonna go through 76 steps, 68 tools, and it's really gonna bring this forecast to life. And along with getting the wrong numbers, I wanted to get a report too. So you can see here that I'm getting a report about this monthly forecast, and I'm getting a view of it too. So I can see the historical data leading up to, today, and I can see what that forecast looks like. And then I can see that breakdown too as well here. Now this is a simple report. Like, sometimes people want to view their forecasts in Google Sheets, and they wanted that data there too. So you could easily ask Mobi to deliver the data that way, and then we can easily get something more live than a report. But this is, like, the starting really simple example I wanted to show you of, of how you can bring a forecast to life simply. Now let's go a little bit deeper. Like, let's have a little bit for more fun now. So not only can you forecast, like, one metric at a time as as Dylan mentioned because Moby's, multivariate forecasting that it takes into account multiple variables, and it can also forecast, multiple things at one time. So here's an example of where I wanna create a daily forecast. So we're going from monthly to daily now for our top 10 products for the rest of the year. In this example, I wanted to look at the top 10 product SKUs. You could also have it just forecast across all of your product SKUs if you wanna do this. Obviously, that'll take a little bit longer, but, you know, maybe an hour later, you're gonna have that forecast. And then I want to flag, like, any potential inventory risks as well. Like, hey. Are there any of these products that are going to stock out later this year? So you could see Mobi thought through this. It brought the forecast to life. And as I'm looking through the report here, it's actually going to flag as well. I guess it's in the left here. It's actually gonna flag as well starting over here that we have, a product, Whoosh Wilson, that's going to stock out on December 6 if we don't order more. And we can see even Splish Splash Sophie is gonna stock out on December 23, if we don't order more. So I think this is a way too of being able to forecast far in advance and being able to even, you know, set up an automation where you can identify which products are you're at risk of running out of, and you can use that to just inform your, inventory ordering strategy. You can see here that I can even just toggle and see the exact forecast for each of these products, and and go into that level. And then here, you can see as well which products are gonna run out. So we can see base shortage, and more from there. And this is this is our demo data too. So you can see these products have, like, funny names because, because they're fake. But as you guys use this too, it'll be super clear which product, the product SKU, the SKU forecast for that product, and then exactly how much inventory you have on hand and when it'll stock out. Alright. Let's go even deeper. Let's show you another even more powerful example. So right now, BFCM is really top of mind for everybody. It's really important for us all to just, like, absolutely crush this season. This is a really powerful UK use case for this tool. So I'm gonna forecast my daily revenue for the BFCM holiday shopping season, and I wanna forecast November 1 through December 31. Here, you can see that Mobi is gonna bring this forecast to life. It's gonna give us some high level metrics. And then here, we have a really awesome daily outlook with the numbers for each. And this is great, and this is nice, but I think how this could be even more valuable if it's if if I turn this into a live pacing dashboard that I can use during this period, and then every single day, I can understand how we're pacing versus the forecast. One note too is that, let's say, Mobi creates the initial forecast, but you wanna be, like, you know, a little bit more aggressive this season to hit your goals. You can always, you know, create a target based on that forecast, let's say, 20% higher, and then you can use that as your aggressive target, for the season. So you can see here on the left, we have the data, and I tell Moby, awesome. Now can you build a live pacing dashboard that we can use to track our pacing with this forecast, once we hit, November I guess I meant to say November 1, but you can see Mobius smart and figured this out. Now I'm gonna open this dashboard. Now here, we have a live pacing dashboard. You can see it's blank right now because we haven't hit, the start of it yet. But you can see once we hit that period, October 1, it's going to start to update this dashboard every single day in real time. It'll update every five minutes with our revenue and how we're pacing for this. So I think this is a really simple, powerful use case, for how you can use the forecasting to inform a live pacing dashboard that you use for something like BFCM or a critical sale period or even just, like, daily pacing tracking? Alright. And then, a few other examples, and then we can yeah. We'll we'll look at some of these questions too in a second. I think one one that's also on top of people's minds right now or or will be soon is just starting to, like, plan ahead for the next year. I think using this tool is a really awesome way to jump start that planning. You can see here I just had Mobi build out a twelve month forecast, for 2027. And each month, I just wanted to start high level. I wanna have our total revenue and total ad spend for that month so I can start the plan. And we can see here that Mobi's building out these numbers, for each month and really just keeping it simple, high level showing, hey. 4,700,000 for January, three around 3,000,000 of ad spend, all based on the historicals. And this is a really great way of just, like, using Mobi to bring the data, the forecast together quickly, and start to bring this to life for that planning. And, again, like, if your team wants to operate out of Google Sheets or Excel, you can very easily export that data, keep working with it there. Alright. Let's see, let's see what questions we have. Yeah. Yeah. You can you can easily input the targets, and you can see all your pacing to that monthly target. That's a really great use case of Mobi. You can just give it those targets, tell it to build you, a live pacing dashboard, and you can track that. Daniel is asking to share out the prompts, for these. Yeah. We we are 100% gonna share out the prompts after this. And on actually, on that subject too, let me show you. We actually have a new I'm gonna grab it real quick in real time. We have a new, like, prompt thing that we're sharing each time, but I'm just gonna make sure you have it all here as well. Let me grab it. Here it is. While while you grab that, Ethan, a. question was asked, which is a good one. Just does model selection or higher thinking effort in Mobi produce significantly different results to the forecast in your experience? Generally speaking, no. Like, within within a few percentage points difference of of effectiveness of the forecast. You know, we we did some benchmarks today just as well, yesterday, because Sol, Sol six came out a much cheaper, lower cost model compared to Astra six from ChatGPT. It's about at a fourth of the cost. We ran it through our forecasting benchmarks, and Sol was within two or 3% of Astra. So with the models that have come out in the last, I don't know, two or three months, there's there's very little difference because the the model that's actually doing the work is Kronos or Times FM. So So if you're running these forecasts, don't feel like you need to use the most expensive, high highest credit cost model. Using something like Sol six, which we just came out with, is is plenty good to get you the results you need and can also be much more cost effective when it comes to to credits. Yes. And, and that's exactly what I used for these, and that's a great call out, Dylan. Yeah. New we're super excited about this model. It's super powerful and super efficient, and it's it's more than plenty, good enough to use for stuff like forecasting. Let's see. We have some other questions. So I know we had a question about the prompts, and then I think Shauna brought up a good point that we haven't shared the prompts the past few weeks. That's because we honestly, we've been moving so fast. We forgot to add this link for a few weeks, and we've been working hard to update it. So I just shared this here. We will send this in the follow ups every time now. And you can see here, we're gonna update this each day we have a live session. You can see the recording, and then you can see the prompts below here as well. So we're gonna make sure to include all of that here. Yeah. And if you missed a missed a past session, it's easy to go back and see that too. For okay. Somebody had a good question on forecasting accuracy and confidence levels. They we will share those confidence levels, and you'll get, like, with a forecast, Moby will share its, like, percent confidence around that forecast based on, the the data that has access to. So, for example, if it has access to less data and it's a little bit all over the place, Moby will tell you if it's if it's less than 95 or, 80% confident in a forecast, and all of that information is provided just like standard forecasting, best practices. And you can you can ask Moe to backtest the actuals versus the forecast. That is definitely something you should do, and we have had several customers really be just, like, surprised by how accurate, the forecasts end up being. A question was asked. Can we model forecast demand of some products as well, like, for fall season? So something important to keep in mind is that the the the forecasting models we're using are just predictive modeling algorithms, basically. So they need some type of input data to forecast something. So if we're trying to forecast demand of a product, it's totally capable of doing that. But to get a super accurate forecast, it's going to need some type of data to model off of. So if you have a data source of, you know, demand of a certain product, you can feed that in as a data source and then get a prediction of of future future sales of that product. If if no data source exists, Mobi can try to find outside data sources and use those. So Mobi could look and see, hey. You know, maybe there's a data source it can find on on winter product sales over the last ten years, and it can feed that data in and try to forecast that way. But it it's it needs some data source to to make a prediction on something. So, if you have the data, that's obviously much better. If you don't and you just ask Moby to to forecast something, it can go try to find data accessible data points online, but you you run into a little bit of grayer area of what the results will look like. Another good question. Someone asked, hey. We have a Google Calendar by month for our planned sales. How do we connect that? Mobi has a full connection to your G Suite. So you can connect Drive, Sheets, Docs, whatever you want, and then you could just say, hey, Mobi. We track, forecasting in this Google Sheet. Do a revenue forecast and then go update this Google Sheet. The Google Sheet that I showed you at the beginning of this, I didn't make. I just told Moby, make a Google Sheet that's a revenue like, a lightweight revenue forecast that's just not using any, not using any of our of our modeling, just to kinda show an external example. But, yeah, you can connect your full G Suite and just have it do your your forecasting all through your existing systems that you're using. Yeah. And related to that, Dylan, somebody asked about, your Google Calendar. Like, that that's another one you can connect, and Mobi can use everything you have on the calendar, to inform as well. Yeah. And you should, Jean, you should be able to just, add the connector right here, and then Moby should be able to access that. But definitely test that out. And then if you do other. question, Ethan. And someone said, can it be uploaded if we were migrating from a different system? So you can just upload CSVs directly into Mobi, and it can store those files and then use those. So you could say, hey. Here's our 2023, whatever, inventory data that, like, might not be in Triple Whale. You can just upload that into Moby and say, keep track of this. Remember that you have our 2023 order data here. And then anytime Moby is using those forecasts, they can pull those in. If that data is stored, like, in Drive, you can just do the same thing. Hey, Moby. Remember, you have access to my twenty twenty three order data in Drive. When you're doing revenue forecast, pull that data in, and it'll pull in a CSV or, you know, Excel sheet, whatever it is, from any of these outside sources and then help use those in its in its predictive modeling. Sweet. Alright, Dylan. Let me know if there's any other questions here right now. One one I wanna get to is someone. said, can. you project revenue at different ad spend levels versus historical ad spend levels? The answer is yes. Absolutely. It's probably one of the best use cases that, like, we should have highlighted better in this is Mobi has access to all of your historical ad spend, and it's one of the biggest levers you can pull in your predictive modeling. So you can tell Mobi, hey. Go look at my last twenty four months of ad spend alongside my order data, and then predict what happens if we're able to scale ad spend by ten, twenty, and 40%, whatever it is. And and it'll actually use all of your historical ad spend data to predict future, future sales. So having it pull in your ad spend data, it's it's a perfect use case, and it's something it's really built for. Yeah. That's a great point. Like, any of those key metrics, especially ad spend. And then if we're planning too, like, hey, Moby. If we were able to improve our ROAS this little bit, like, how would that adjust the plan? What can we do there? But, yeah, great great example. Yeah. There's no this is a straight forecast. This is not this is not marketing mix modeling. We will add these prompts to the link that we shared, which we are making sure to share in every follow-up. We can share it again. These prompts will be added here by the end of day today. Last thing I'll say is, like, there's a lot of questions coming in around, like, can it can it forecast this? Can it forecast this? Someone said, you know, how would you incorporate a planned price increase into the forecast? There's so much been work that's been done in on the Mobi side that you can just say, hey. Forecast our BFCM 2026, and in October, we're gonna be raising the price of our best selling product by 10%. And it'll work all of that into the forecast pretty easily. So any of these outside variables that you're curious about or wondering about, just throw those into the prompt. There isn't anything special you need to do. You know, Moby is basically taking all of these pieces that you're giving it and then organizing them and then feeding them into either Kronos or Times FM. It's picking which model it thinks will be best. So you don't really have to worry about how you're organizing the data or the pieces that you're throwing it. It's Moby's job to figure that out, and then it's just showing the outputs. So any variables that you have, throw them in. If it can use them, it will. If it can't, it'll tell you. Totally. Like, push you I think the more you push Mobi on use cases like this, the more you'll be really blown away, by what it can do. I think it's it's stuff like this where people really realize just the full breadth of what we built and and what's possible. Amy asked a good question too. She said she used Mobi to, in a different way, generate a forecast this year. How would you approach auditing that process to learn for this year? I I I do really think a starting point there is asking Mobi to audit that to review as well. But, yeah, I think it's another example of how you can use Moby to just kind of lean in and and learn from these things. That's what we do a lot as well. The feature is is already rolled out and completely live for everybody that has access to Mobi. Alright. Amazing. Any final questions, anybody? I do have some closing. Let me share some closing plugs. We have, yeah, Keith. But if if just asking Moby fails, then then you can come to our team too. Don't worry. We have we have lots of support here as well. Alright. So one thing too I wanna share is this is so we have a session tomorrow, if you haven't registered yet, where we're joining with a real customer, Felix, who is, the cofounder and chief experience officer under Outfit. He's he's a really great example of a Mobi power user, really using it in just, like, all aspects. And so that definitely recommend joining that session. AJ is gonna join. We're gonna talk to him. It's gonna be super fun. And then another fun one, so not Mobi specific, but ChatGPT ads, we know are new and very top of mind for everybody. We're actually co hosting a session with them next week all about, like, how you can use ChatGPT ads and Triple Whale to start scaling on that platform in a way that you can, trust. Because we know, you know, new ad platforms are exciting, but also a little bit scary. So just wanted to share that here. The team asked me to share that. But lots of lots of cool valuable live sessions in the pipeline. But, yeah, I think that's all all I got. Thanks, everyone. Anything else still in? No? Thanks, everyone. Awesome. Just last thing as always, ethan@tripleoil.com, dylan@tripleoil.com. If you have questions, shoot us an email. We're here. Thanks, Yeah. everyone. Hit us up. Thank you all.