Perfectly predicting the future is impossible. But for ecommerce brands, accurately forecasting demand is still pivotal for success. Because when you can project future sales, you can prepare to fulfill that demand, avoid stockouts and overstocks, and unlock business growth.
But what is demand forecasting? Why is it critical to supply-chain management? And what’s the best way to predict customer demand? Let’s find out.
Demand forecasting (AKA inventory forecasting or sales forecasting) is a predictive analysis of future customer demand based on historical sales data and real-time inventory trends. By finding patterns in this data set, brands can accurately estimate how much product they’ll sell in the coming month, quarter, or year.
Accurate demand forecasting enables retail brands to confidently purchase enough inventory to fulfill upcoming demand. That way, they don’t stock out or over-invest in excess inventory — both of which tie up cash flow and put brands at financial risk.
When brands accurately forecast demand, they can leverage that information to make smarter inventory and supply chain management decisions. For instance, they can optimize inventory, increase turnover rates, reduce holding costs, and more.
Accurate demand forecasting outlines what inventory you’ll need (and roughly when you’ll need it). That way, you can find your economic order quantity (EOQ) and maintain ideal inventory levels. And with inventory control, you’ll reduce stockouts, dead stock, and lost revenue.
You can’t fix problems you don’t know you have – including slow-moving SKUs. By forecasting demand, you’ll identify what products aren’t selling quickly. That way, you can strategically boost inventory turnover rates.
How? By running targeted marketing campaigns or creating product bundles that feature that item alongside a bestseller. These marketing strategies speed up the sales velocity for stagnant SKUs, so you can boost your profit margins and free up vital warehouse space.
Brands that only order what they need avoid overspending on excess inventory. That’s because the more merchandise you hold in your warehouse, the more expensive your carrying costs. By only ordering stock you know will sell, you can eliminate unnecessary holding costs and free up working capital for other parts of the business.
Generally speaking, the longer items sit in inventory, the more likely they’ll become dead stock. And this can cost you big – dead inventory costs a shocking 30% more than the inventory’s value on average.
But with accurate demand forecasting, you avoid over-ordering in the first place, which lowers your risk of overstocking items that will become obsolete and go to waste.
When you stay on top of your real-time inventory data, you can detect and anticipate any upcoming operational issues.
For example, you’ll see when inventory levels start running low. This way, you can avoid stockouts (and missed sales) by replenishing inventory at the ideal reorder point. Or, if you have aged inventory, you can run a marketing promotion to get rid of it before holding costs wreck your margins.
Demand forecasting helps you anticipate upcoming costs and revenue based on anticipated sales. With this information, you can plan your budgets and resource allocation accordingly.
For instance, your operations team might use this information to plan future inventory needs and outline corresponding production orders. Meanwhile, your finance team might develop business projections to share with investors.
As I mentioned earlier, you can easily maintain optimal inventory levels when you forecast demand. These optimal inventory levels ensure you have what customers want when they want it.
This lends itself to a better customer experience and increases the likelihood that those customers will shop with your brand again. Every time they come back, their customer lifetime value (LTV) increases, and your revenue gets a boost.
Most retail brands will use some combination of quantitative and qualitative methods to increase their demand forecasting accuracy.
Qualitative methods of demand forecasting rely on intangible information, which in most cases means an expert opinion or unbiased, logical approach. It’s better suited for long-term forecasting. And while historical data can be included (if available), it’s not required.
Time series analysis makes future predictions based on historical, time-stamped data. Typically, brands build demand forecasting models based on fluctuations in historical sales and then use them as a guide to making observations and predictions about the future.
To get started with this method, collect and sanitize all your time-stamped data from a specific time interval. You want to pay attention to any outliers caused by unprecedented events, including viral campaigns and the pandemic. Then, use the data set to understand the “why” behind specific outcomes (like why sales spiked in Q1) and predict future demand.
The Delphi method relies on a group of experts who are either interviewed separately or provided with a questionnaire. Then, their feedback and predictions contribute to the forecasts. This method does not rely on factual data and is best used for long-term demand forecasting.
To begin, find your expert advisors who work in a similar industry (the DTC luggage brand Away might consult advisors from Baggu or Rimowa). Then, hire a 3rd-party facilitator to create questionnaires, gather answers, and create summary reports. As soon as the experts come to a consensus on the summary, you can use the information to influence forecasts.
Like the Delphi method, a customer survey analysis relies on individuals’ opinions (versus factual data). But in this scenario, you listen to target markets instead of industry experts to explore new markets, identify changing sales trends, and forecast future demand. This way, your target customers can impact decision-making.
For example, beauty brand Glossier collects feedback after every purchase. It then uses the data to gauge customer satisfaction and interest in new products. If the appeal is low, they can scale back forecasts (or hype up more excitement around the launch). And if the interest is high, they can ramp up forecasts to avoid product stockouts.
Brands with wholesale partnerships can also forecast demand based on their sales team’s experience, estimation, and deep understanding of changing customer needs.
But it’s important to aggregate as many opinions as possible (versus a few individuals). Otherwise, a limited data set might introduce biases and lead to underestimates or overzealousness.
Quantitative forecasting methods use tangible data (namely, historical sales data) to identify sales patterns and estimate future demand. This method is better fitted for short-term demand forecasting. And it is only available to mature brands since it relies on long-term data collection.
Trend projection forecasting looks at how historical sales and product demand change over time. By assuming that future sales will reflect past sales, retailers can leverage any identified patterns to forecast future metrics.
To use trend projections, start by inputting your historical sales into a spreadsheet and eliminate any outliers that might cloud your data (such as seasonality and unexpected surges). Then, find patterns in the sanitized data and use them to predict your upcoming demand.
The econometric forecasting method considers how market trends (like consumer spending, household income, and inflation) will affect future demand.
For example, home workout equipment brand Peloton saw an incredible spike in sales (172% YoY) when gyms shut down, and people opted for at-home workouts. An econometric forecasting method would’ve informed how Peloton should stock up to fulfill this demand.
To use this method, you will need to gather your external variables (economic trends) to plot your dependent variable (customer demand). Then, determine what relationship each external variable has with your customer demand (for example, gyms closing positively affected Peloton sales). Then, use this information to forecast future sales.
Similar to trend projections, the barometric forecasting method uses past sales to predict future demand.
But it also layers on additional economic data (called leading, lagging, and coincident indicators), which influence your future projections.
Other ways that brands forecast customer demand depend on the business’s unique needs. But generally speaking, it’ll be some combination of long- or short-term, macro- or micro-level, and active or passive forecasting models.
Long-term forecasting attempts to predict demand far into the future (more than 1 year out) and is typically used for budgeting. In contrast, short-term forecasting focuses on predicting demand for the next 3-12 months and is used to manage just-in-time supply chains.
Generally speaking, long-term forecasting is less accurate than short-term forecasting. However, it offers a bird’s eye view of what inventory the business should expect to need. Your operations team can then use this loose demand plan as leverage to negotiate better vendor terms.
Meanwhile, short-term forecasting is more accurate because it offers a ground-level view of your inventory needs. Brands can use this to inform inventory replenishment and supply chain decisions based on their average order lead times.
Macro-level forecasting looks at external factors (such as consumer trends and economic factors) to predict new product opportunities, potential financial barriers, and material shortages.
Meanwhile, micro-level forecasting still looks at external forces but drills them into specific industries or customer segments. For example, a clothing company might forecast demand specifically for Gen Z females in the United States purchasing via social media.
In other words, macro-level forecasting tries to identify “unprecedented” demand (like the homeware spike when the pandemic hit) as well as seasonal trends (like increased spending during the holidays). This information can then be used to inform micro-level forecasts, which ask: How will this macro-level projection impact my customer’s needs?
For instance, the homeware spike drove suburban box stores (like Office Max) to order more desk chairs. Why? Because a large percentage of their target demographic started working from home as their companies responded to Covid.
Active forecasting uses external factors (like market research) to predict future demand. Because this forecasting method doesn’t require historical data, it is typically used to forecast demand for new products or companies operating in a new market.
Meanwhile, passive forecasting relies on historical data to predict future demand, and it is often used by more established brands that have this past sales data available.
To promote accurate demand forecasting, brands must account for several internal and external factors (competitors, the economy, and the brand’s product offerings).
External factors happen outside the business’ retail operations and are typically not under a brand’s control — for example, seasonality, competitors, economy, and more.
Seasonality (or seasonal trends) refers to predictable variations in demand depending on the time of year. For instance, weather and holidays can cause seasonality, impacting customer buying habits.
Take Bask, for example. The DTC sunscreen brand sees most of its sales in summer when people vacation. By forecasting this seasonality into its demand planning strategy, the brand can ensure it has enough bottles of SPF to cover its peak demand.
But to accurately forecast seasonal demand, you’ll need to consider how the time of year affects volume orders.
When new competitors join the market, you’ll likely see demand drop for your existing offerings. Why? Because some of your customers will inevitably try the new competitor.
Likewise, when competitors go out of business, surviving brands see demand increase because consumers now have fewer choices.
Keeping a living list of all your direct competitors and their offerings is an easy way to gauge how the competition will affect demand.
Where do most of your audiences live? The answer to this question not only impacts consumer demand for your products, but customer purchase behavior, your order fulfillment strategies, and brick-and-mortar locations.
For instance, Arizona has a desert environment known for hot temperatures and sunshine all year round. So, they don’t have a huge demand for winter jackets. Meanwhile, places that get snow like New York does.
If your brand exclusively sells winter wear, a flagship store or fulfillment center in Phoenix, Arizona, probably won’t make sense. Likewise, an Arizona-based small business might not want to stock up on sub-Zero coats.
How well the economy is performing tremendously impacts consumer demand.
When a recession hits, consumer spending and product demand for non-essential items decrease. But when the economy is booming, pending lifts across all sectors — especially non-essentials like travel and luxury goods. Knowing these macro-trends can prepare you to weather hard financial times.
For example, inflation jumped 8.6% in May 2022 – a 40+ year record increase that motivated cost-conscious consumers to reevaluate their spending, cutting out non-essentials. And for many economists, it’s a warning that a post-pandemic recession is coming. Knowing this, it might not make sense to launch your most expensive product to date right now.
For most brands, going viral (on TikTok, for example) is the goal. But most retailers don’t anticipate the sudden spike in demand that comes with virality.
So, when most brands finally find internet fame, they’re left stocked out, missing out on revenue, and scrambling to recover. (That is unless they sell on backorder.)
Drew Fallon, COO of Mad Rabbit, explained on The Checkout podcast that going viral is an external factor that’s nearly impossible to bake into your demand forecasting accurately.
“We had 1 video on TikTok reach like 25m views,” Drew told us. “And you cannot prepare [your inventory] for that. That might be what you’re trying to do every time, but you don’t know when it will work.”
🤿 Dive deeper: Mad Rabbit’s Drew Fallon on demand planning.
Unlike external factors, internal factors happen within the company’s operations and are generally under its control — for example, product offerings, marketing events, and new product offerings.
Forecasting demand will be different for each specific product or service you offer. Everyday items tend to have higher demand (because people need them more frequently).
Let’s consider how this works for Campo, a DTC essential oil brand. Based on its product offerings, the brand likely replenishes its scents (the essential oil that goes into the diffuser) more often than its diffusers.
Why? Because most customers will buy several bottles of essential oil each year (after all, there’s only a limited amount in each bottle). But a quality diffuser has a longer lifecycle, even with daily wear. Meaning, customers won’t need to replace their diffuser as often as they do their essential oils (leading to lower demand for that item).
Any promotion you run or pricing change you make will affect demand. So, you’ll need to include these marketing events in your forecasting efforts.
Generally speaking, a deep discount increases demand for the promoted product. So, you’ll need to forecast how that spike affects your weeks of supply calculation (unless you’re sunsetting that SKU). Otherwise, you could sell out before the promotion ends, leaving revenue on the table.
Conversely, increasing prices (like many brands did to respond to inflation) can decrease demand. Knowing this, some brands will temporarily increase costs when they’re at risk of stockout to slow sales.
🔥 Tip: If a price increase doesn’t dip demand, it could indicate you’ve been undercharging.
New products don’t have historical data – an added challenge when forecasting demand for new products. But you can lean on data and trends from similar products to try and predict how much inventory you’ll need.
That said, you never really know how the new product will affect demand for your other product offerings until it’s on the market.
For instance, launching new products can cause product cannibalization and negatively affect demand for older products. Think: Apple releases new iPhones, and overnight, people stop buying last year’s model.
So, when launching a new product, consider re-forecasting demand for your existing products as well. That way, you don’t end up with dead stock that won’t sell.
Understanding the importance of demand forecasting is one thing, but successfully carrying out the demand forecasting process is another. Here’s how to forecast demand as accurately as possible:
First, you’ll need to choose the period of time you’re forecasting — whether it’s the next 30 days, 90 days, or 12 months.
The amount of time you choose will depend on your goals. A shorter time period (like the next 3 months) is best for planning inventory replenishment. Meanwhile, a longer time period (like the next 12 months) is better for production planning.
When you have your time period, start collecting and reviewing the historical data from the same period last year.
Next, you need to set some sort of educated guess or hypothesis based on the data. To do that, you’ll need to compare the historical sales data (that you collected from the same period last year) to your current sales data (the previous ~30 days).
Using both sets of data, ask yourself:
Also, if there are no recorded sales for a product, understand why before moving forward. It might be because consumers didn’t want to buy the product (AKA, no demand) or because you were out of stock.
To set your hypothesis, you have to determine if the trends you discovered will apply again during the same period of time (such as the next 30 or 90 days) or if they will differ. For instance, maybe you’ve gone viral on social media, or there’s a pandemic-driven demand.
Either way, your hypothesis should answer: Will the defined period perform similarly to what we’ve historically seen?
With your hypothesis, determine the best forecasting method for your brand (this will depend heavily on your available data). The best models are built from the bottom-up. Or in other words, you create the foundational forecast on quantitative data and then cautiously layer qualitative assumptions on top.
For example, let’s say you’ve been selling a product for 3 years, know your industry’s seasonal trends, and keep track of real-time stock levels.
You would use quantitative forecasting methods because you have a solid data set available (to find patterns and better predict future sales). Then, you can layer on growth assumptions using qualitative methods (like upcoming marketing promotions).
However, forecasting a new product (or for new brands) means much of this information won’t be available. And if that’s the case for you, it will require a qualitative approach over a quantitative one.
Now, you can create the first forecast model (called your base demand model) and remove any outliers. There are tons of statistical techniques available for doing this in spreadsheets — just run an online search for “how to forecast in [your specific forecasting tool].”
👉 Prefer to stick to Excel? No worries — here’s a complete guide to inventory forecasting with Excel.
Once you’ve collected and analyzed the data set, you’re left with a stripped-down demand forecast based exclusively on your historical and current sales data. That means if you needed 1,000 units in this time period last year (and this year’s sales are roughly the same), you’d probably need 1,000 units this year.
But you haven’t included your hypothesis at this point (you’ll do that in the next few steps). That makes this base model the safest forecast and the foundation on which you’ll cautiously layer other assumptions to increase forecast accuracy.
Look over the data again to remove any outliers. Plus, double-check that the data is entered correctly (namely, there are no human errors) and is still up to date.
It’s also helpful to calibrate the model and test that it works by entering data other than what you used. But the most effective way to eliminate human errors is using forecasting software.
For instance, Cogsy automates how you collect your inventory data and enter it into your forecasting model. Then, the tool automatically updates your forecast whenever something changes (like if you go viral on TikTok and sales suddenly spike). This makes human errors impossible, and your demand forecasts wildly more accurate.
Taking your base model, it’s time to cautiously layer on new assumptions (like upcoming marketing promotions, current fads, market landscape, and seasonality).
It’s important to consider how each of these factors impacts demand individually and in combination. To do that, run through alternative scenarios that might happen (sometimes referred to as setting parameters).
For example, if demand flops for a particular SKU, consider what happens to your inventory needs then. Or, if one of your products goes viral, think about what you’d do to meet demand.
While both scenarios are possible, they’re not necessarily likely. So it’s essential to layer the hypothesis onto the baseline forecast cautiously. Then, check these new alternative models for errors similar to your base model (but with an added focus on cutting improbable models).
Using your forecast, determine how much inventory you need and when. To do this, map out your operational plan for the entire period (30 days, 90 days, or 12 months) and consider building loose secondary plans.
If one of the other scenarios plays out (stockouts or viral sensation), consider how you will still meet those inventory needs. These plans allow you to pivot your strategy quickly if necessary.
Finally, confirm that your vendor and fulfillment center can provide what you need in the time frame you’ll need it. (Doing so can noticeably improve your supply chain relationships.)
Now that you have your forecast and operational plan, track the actual demand in real-time throughout your forecasted period. This way, you can compare what’s really happening (actual sales) versus what you predicted would happen with your “crystal ball” (predicted demand).
Then, adjust your forecasts and inventory plans to reflect this new data. Your predictions become more accurate the more “proven” information you can add.
Don’t forget to document what you learn. This can add additional color that would otherwise be missing the next time you forecast demand.
Your forecasted demand is not a plan that’s set in stone. That means every time new information is introduced, brands should re-forecast to eliminate models that won’t actually happen (and increase their forecast’s accuracy).
Then, dig into what this updated forecast means for your operational plan. For instance, when you see a sudden spike in demand for an SKU, you’ll want to consider the following questions:
The best demand planning software, like Cogsy, will automatically do this work for you.
An ecommerce brand has been selling cosmetics for the last 2 years and wants to project the next year of sales. Therefore, the team forecasts demand using the time series analysis (since they have a solid long-term sales database).
The brand collects and reviews its historical sales data to forecast an average demand of 20,000 units over the 12 months.
They also find that their best months are May and June, their worst months are September and November, and most of their customers are from California.
So, they plan a huge Black Friday marketing event in November to increase demand. They expect this promotion to have a 10% lift (meaning, they expect to sell roughly 2,000 more units than the previous November).
This data helps them forecast demand for the coming year (22,000 units). And it illustrates the optimal times to order most of that year’s inventory (before May). Plus, if they want to save on shipping time and costs strategically, they could consider moving their fulfillment closer to California.
Now, the cosmetic brand can optimize its inventory planning by only ordering what they need. This helps them avoid stockouts and avoid investing too much money in inventory before absolutely necessary.
And if they decide to move where they fulfill orders, the brand could also reduce its average lead time, lower its cost of goods sold (COGS), and subsequently increase its revenue.
🤿 Dive deeper: How to lower your COGS.
Estimating future sales is pivotal for your demand planning process, but forecasting has limitations. Here’s what to look for if you want to avoid false predictions:
Startups, newer retailers, and even established brands launching new products don’t always have long-term historical data. Without a foundation of past sales, predicting future trends is nearly impossible. And even if you try using your available information, it may not be very accurate.
But that doesn’t mean you can’t accurately forecast demand — you just need to choose a different planning method. For example, if you lack historical data, use a qualitative forecasting model (which doesn’t rely on concrete numbers).
It can feel like a complex and somewhat confusing process if you’re new to forecasting demand. And without a lot of practice, knowing which model to use, how exactly to calculate it, and what growth assumptions to layer on top are challenging. And it can result in unreliable predictions.
But you can combat this problem by relying on an ops optimization tool like Cogsy (instead of relying on a human who might lack expertise). That’s because the software uses your historical sales data and real-time inventory trends to predict customer demand accurately (no forecasting experience required).
Speaking of assumptions, sometimes brands can get a bit unrealistic and overcompensate with too much inventory. This leads to overstocks that eventually turn into dead stock, and that defeats the whole purpose of forecasting in the first place.
To combat unrealistic assumptions, consider pulling back on your PO quantities (even if it’s just a little) to avoid overstocking and racking up holding costs on unsellable items. Or, invest in a tool like Cogsy to know precisely what you need and when to place the PO.
Demand forecasting takes a lot of time and human resources, both costing money. And if you spend a lot of effort forecasting and then get it wrong, you waste precious time and spend more capital on inventory that won’t sell.
The best way to resolve this issue is to replace manual spreadsheets with a tool like Cogsy. That way, your team doesn’t waste time inputting (and reinputting) data, and there’s significantly less room for human error.
Luckily, you can avoid forecasting woes with Cogsy’s planning tool. With it, your operations team can project customer demand with pinpoint accuracy. Here’s how.
Cogsy collects, sanitizes, and correlates all your data from multiple sources to automatically and accurately forecast demand. So you can replace manually-updated spreadsheets with real-time demand data that shows you exactly what to order and when to place the PO.
Cogsy automatically forecasts demand using your historical and real-time demand data. That means there’s no need for forecasting expertise or gut assumptions. And while you can’t 100% predict the future, you can be confident that your demand forecasting is as accurate as possible.
Since Cogsy does the demand planning for you, your team gets back 20+ hours a week on average. And since your forecasting is more accurate, you’ll avoid costly mistakes and investing too much capital on inventory (whether it’s the wrong SKU or the wrong time).
So, are you ready to stop wasting money on overstocks and losing revenue to stockouts?
Demand forecasting is an exercise to predict what consumers want to purchase from your brand. Meanwhile, demand planning ensures your brand can fulfill that projected demand efficiently and cost-effectively.
Some of the benefits of demand forecasting include optimized inventory management, reduced holding costs, minimized waste, boosted brand loyalty, and more revenue.
Demand forecasting aims to predict what consumers will want to buy from your brand and when so you can prepare your operations accordingly. For instance, brands can use these forecasts to make smarter, more informed decisions that ensure they can fulfill all that demand without accumulating waste.
Quantitative forecasting relies on analyzing hard data. Meanwhile, qualitative forecasting leverages subjective soft data. Which method your brand should use will depend on the available data, but most brands use quantitative and qualitative methods in unison.