By Tal Eden, Chief Data Scientist at


Machine Learning is the ability of computers or machines to learn without being explicitly programmed. Arthur Samuel

Supervised learning is a specific field of machine learning in which a machine learns to label data based on a set of prelabeled examples. Sales prediction is a specific example of supervised learning. We have a labeled data set of past sales, and we generate a prediction for future sales.

In the example of retailers – we want to infer the future sales volume of each product in each point of sale based on historical data such as past sales, item price, date (whether it is a holiday or not), etc. In the past several years, we have seen a spike in the number of businesses using such predictive components in their planning processes as technology gradually became more accessible.

One of the greatest adversaries of sales prediction (and of any prediction engine in general) are black swan events. The machine cannot reliably predict any longer because the historical (labeled) data is no longer relevant. For example, historical sales data after a financial crisis like the one that occurred after the collapse of Lehmann Brothers in 2008 can no longer serve as a basis for prediction because people have less money and therefore tend to buy less. The environment in which the predictive engine was created has changed so drastically that it has rendered the engine unusable. Another prominent example is the Coronavirus outbreak that we are currently experiencing.

Our machine cannot predict accurately because history is irrelevant; here, history cannot and does not repeat itself.

Sales have been altered completely – whether online or offline.

Fashion retailers have suffered a significant decrease in sales volume. A prominent example in this industry is Adidas; the German sportswear company reported that it expects first-quarter sales to drop by $1.14 billion in greater China. Like most fashion retail stores, they have had to close physical stores, and people are buying less sportswear and leisurewear.

While sales in the fashion industry have plummeted, sales in the online food industry are skyrocketing. According to Nielsen, last year, just 4% of grocery sales in the US came online. However, with people avoiding crowds or being quarantined, the alternative is to buy online. Downloads of Instacart, Walmart’s grocery app, and Shipt increased 218%, 160%, and 124%, respectively compared with the prior year.

Can we still predict?

Do not despair! Although times are changing, businesses who can adapt rapidly have the opportunity to make the best out of the crisis and leverage it for future growth. Below are a few suggestions on how to deal with this situation in terms of demand forecasts – and all its implications – sales, financial planning, orders, and deliveries?

  1. Understand the New World Order
    Create a comprehensive, real-time BI dashboard. You can see which products are seeing an increase or a decrease in sales. Walmart has reported an increase in sales for tops, whereas a decline in sales for bottoms (thank you video conferencing).
  2. Integrate external data sources – You need to understand how different events influence this new “order” – the number of infected people, whether schools have been closed or not, is there a curfew in place and more.
  3. Working without Historical Data
    When there is little historical data to rely on, data should be sampled at a higher frequency. If you were generating predictions on a monthly basis, you should switch to daily predictions, and if you were doing daily predictions, we recommend real-time analysis. 

    This will benefit the company twofold:
  • Information will be gained quicker
  • Sensitivity to changes will be higher even without long term predictions

Accordingly, we will need a different set of ACTIONS than those for long term predictions:

  1. Inventory: Your inventory strategy in times of crisis should be continuously changing based on data-driven insights. A few examples are mobilizing products intended for offline stores to online stores or making sure you have enough stock in locations in which the stock is likely to be consumed.
  2. Marketing strategy: Your marketing strategy should reflect the changing environment. Identify strong sellers, feature them, and use your marketing budget for them rather than “weaker” products or try to push products that display a strong negative trend before the trend sets a new baseline.
  3. Be ready to draw rapid insights and translate them to quick actions – When the environment is changing rapidly, businesses who cannot adapt in a timely fashion will be rendered irrelevant. You should be ready to make business-impacting decisions based on your new insights and predictions.

While we cannot predict the future reliably in times of uncertainty, we can draw meaningful conclusions regarding the rapidly changing environment.  This helps us harness the technology to help us identify immediate trends, plan our next steps, and how to weather the storm the best way possible.

Do you want to hear more about how you can use machine learning in your business? reach out to us for a consultation at

Despite what some may believe, SMBs (small and medium-sized businesses) are not simply smaller versions of enterprise organizations. The size of a company has a deep impact on how a company is organized, but most importantly, it impacts the purchase processes. Since we are talking about different sales processes, we also need to define a different forecasting approach for each type of company. In the following post, we will go through the most important factors that dictate the differences between enterprise and SMB forecasts.

The sales cycle

The complexity of the sales cycle is the main factor that affects the differences between enterprise and SMB sales forecasts. For starters, enterprise sales have more complex sales processes and longer sales cycles. This happens because the decision-making process is often divided between several stakeholders, not to mention the fact that decision-makers are harder to reach in an enterprise. Moreover, each enterprise can have its own purchase process, which can be more or less complex, affecting the length of a sales cycle. On average, an enterprise purchase decision requires the consensus of 7 stakeholders. The longer a sales cycle is, the harder it is to predict its outcome. SMBs have shorter cycles because their needs are more urgent, and their purchase processes are less complex. On the other hand, with enterprises, a cycle can take more than 6 months, so it will be harder to make quarterly forecasts for these companies.

The accuracy of historical data

Historical data is the heart of a forecast, but the accuracy of this indicator can vary based on the size of the company in question. The reason for this is very simple: smaller companies tend to have shorter and less complex sales cycles, which leads to more concise and more accurate historical data. Enterprises, on the other hand, have complex sales cycles which can be affected by numerous variables such as the availability of decision makers, internal purchase processes or last-minute budget cuts. Due to the complexity and volatility of enterprise sales, enterprise forecasts need to be more complex and include as many variables as possible to increase the accuracy of their historical data.

The sales reps

As the sales processes differ for enterprises and SMBs, so do the skills and the goals of the sales reps. Enterprise sales representatives are often more analytical as they use more market date to state their points and stay ahead of the competition. Moreover, enterprise reps are also farming sellers, which means that their end goal is not just to make a sale, but also to get as much business as possible out of a new or existing client. SMB reps, on the other hand, are more aggressive, they deal with less competition, but they also have less experience and less data to make informed predictions. Since there are several forecasting methods which consider the predictions of the sales reps, it makes sense that the different skills of the sales rep will impact the final predictions.

Another way in which sales reps can impact forecasts is by abandoning leads in the middle of a sales cycle. This has a bigger impact on enterprise sales which can take months to finalize. Having an enterprise sales rep leave in a middle of a deal will not only impact the morale of the whole team, but that rep will take with him all the business relationships that he nurtured, relationships which will have to be rebuilt from scratch.

The complexity of the deals

Enterprise sales are far more complex than SMB sales. When selling to SMBs, you could close a deal from the first contact with a new lead, but this would never happen with an enterprise client. The more people there are in a company, the more complex the deal will be. Enterprise sales also leave more room for upselling or cross-selling, not to mention the fact that most enterprises prefer customized products.
Conclusion: As you can see, enterprise and SMB sales deal vary greatly. The differences in the sales processes require a different set of indicators which will affect your options in terms of forecasting strategies. To know which metrics, you need to track and to increase the accuracy of your forecasts, it is essential to understand the needs of your clients and the complexity of your deals.

In sales forecasting, mistakes are to be expected. However, with each mistake that you make, you should learn something, in order to refine your forecasting methods, and improve the accuracy of your forecasts. As a rule of thumb, the more metrics you track and include in your forecast, the more accurate the forecast will be. In the following post, we will discuss some key metrics that all businesses should track.


Accuracy is not only a characteristic of your forecast but an actual metric that you need to keep track of. There are several ways in which you can measure the accuracy of a forecast.
By the measure of error
Each sales team should measure the error of the forecast which is calculated by taking the absolute value of the difference between your forecast and your actual sales, and divide it by the divisor (the largest value between the forecast and the actual numbers)
Measure of error = (F-A)/F
F – forecast
A – actual sales
By product type
Accuracy can also be calculated by product, a strategy which can be a lot easier for your sales team to implement. This strategy is best suited for off-the-shelf products that don’t have customization options that can complicate the sales cycle. When it comes to complex products, you can simplify the process, by breaking the prediction down into several parts.
By size of the deal
For this strategy, you will start with a less accurate forecast in the early stages of a deal, which you will refine along the way. In the later stages of the sales cycle, you will have a better understanding of the needs of the buyers, which will help you refine your numbers.
By time period
There are two critical stages where you can measure accurately. The first stage is 90 days out, a stage at which your accuracy should be quite low, as the deal is a long way from closing. Nonetheless, there are a lot of early signs that can help your sales reps predict the chances of closing a deal. The second stage is 30 days out. At this point, the sales cycle should be at the final stages, and your accuracy should be above 90% at this point.  


When it comes to complex transactions, it is not uncommon for planned close dates to be pushed back. It is important to track these types of sales and discover indicators that show a deal’s chances of getting postponed. A pushed deal is not a lost deal. In most cases, it is just a deal that is postponed into another quarter, but this will heavily impact your predictions.  Moreover, pushed rates can also indicate a problem with certain sales reps.


The variance is calculated as the distance between the commit and the pipeline upside. This indicator will help you understand how an initial forecast can change from the beginning to the end of a quarter. To improve the accuracy of your variance, you need to be very involved in the deal activity of your reps, but you must also consider their gut predictions. In the long term, tracking the variance will not only help you understand the changes in your forecast, but you will also be able to narrow down the factors that influence these changes.


Depending on your business, and the complexity of your sales cycle, a sales forecast can be balanced throughout a quarter, or it can weight more heavily towards the end of the quarter. To measure the linearity of your forecast, you will have to track the stages of a sales cycle and the close dates. The point of tracking this metric is to adjust your sales strategy so that in the long run you can rely on an accurate and linear forecast.


As most of your forecasting strategies rely on historical data, it is important to keep an eye on your team’s commitment to recording all valuable data. The compliance numbers will show you if all the members of your team have submitted accurate forecasts for the relevant opportunities. Obviously, the sales rep forecasts are refined by a team leader, but unless all the members of your team are committed to tracking all relevant metrics, you won’t be able to improve the accuracy of your forecast

According to eMarketer, online shopping now accounts for almost 10% of total retail sales, making e-commerce a significant factor in traditional retail and this trend is likely to continue in the years ahead.
The US Census Bureau figures show that over the last two decades, online retail sales in the US have grown rapidly, rising from 5 million in 1998 to 389 million in 2016. Additionally, in Europe, Enterprises Total Turnover from E-commerce represented 15% in 2014 and grew to 19% in 2017 (Source: Eurostat).

Growth per category

As one would expect, electronics dominate e-commerce sales, followed by miscellaneous products such as office supplies, gifts, novelty, and clothing, based on the US Census Bureau figures. On the flip slide, ranking very low, are categories such as food and beverages, that have a very small share of online sales. This will surely evolve in the coming years, as it already did in 2016 when shares of online sales nearly doubled in that year alone for consumable products.

Amazon e-commerce statistics

  • Amazon is the leading online retailer with net revenue of $232.88 billion in 2018. (Statista)
  • Amazon has over 100 million Amazon Prime members. (Jeff Bezos in a letter to shareholders)
  • Other sources estimate that there are over 95 million Amazon Prime members in the United States. (Statista)
  • On average, two in five US consumers (41%) receive one to two packages from Amazon per week. That number jumps to half (50%) for consumers ages 18-25, and 57% for consumers ages 26-35. (Walker Sands)
  • In the last six months, 83% of US consumers have made a purchase on Amazon. (BigCommerce)

Recent Trends

More recently, the retail industry is undergoing two significant shifts. The first is technological, and the other a result of changes in consumer behavior. It’s the stores that understand and overcome both of these major shifts that will thrive. Alternatively, retailers that don’t will go the way of Circuit City, Borders, and Blockbusters.
Online retail has grown 300% between 2000 and 2018, according to the U.S. Commerce Department. During the same time period, however, department store sales have dropped almost 50%. In 2018, JCPenney, Gap, and Victoria’s Secret announced the closures of 300 stores.
Although shoppers will probably never wholly abandon brick-and-mortar stores, they expect retailers to offer a convenient online alternative. Most brand names are responding while still trying to get shoppers into their stores for pick-up of large items. In order to keep customers coming back to physical store locations, retailers must use a combination of branding, service, and pricing to convince shoppers to get dressed, get in their cars and drive to pick up merchandise.

E-commerce cart abandonment statistics

  • The average global cart abandonment rate in Q3 of 2018 was 76.9%. (SaleCycle)
  • 58.6% of US online shoppers have abandoned a cart within the last 3 months because “I was just browsing/not ready to buy.” (Baymard Institute)
  • The top three reasons US online shoppers give for abandoning a cart during checkout are high extra costs, the need to create an account, and a complicated checkout process (these are the survey results after removing the “I was just browsing/not ready to buy” segment). (Baymard Institute)
  • The average open rate for an abandoned cart email is 15.21%, and the average click-through rate is 21.12% for SmartrMail users. (SmartrMail)
  • The average revenue per email for an abandoned cart email is $27.12 (for SmartrMail users). (SmartrMail)

US vs. The Rest of the World

Despite the assumption, the US is one of the largest consumers of the online marketplace, alas, it is not. In the end, China ranks first place for the largest online marketplace followed by the United States respectively.
According to Nielsen’s Global Connected Commerce, The foothold the US has in the global economy is slowly declining with global shares expected to decline by 16.9% by 2020. In terms of industry, South Korea leads the way in fashion and beauty products while Japan remains ahead of the curve for music and stationery purchases.
Sales forecasting accuracy is crucial across the retail chain and Curve’s powerful Sale’s Prediction engine often exceeds traditional forecasting methods by determining the impact of sales history, consumer trends, holidays and more…
“Retail demand forecasting is one of the hardest analyses to get right: Forecast too little and you have empty shelves, and forecast too much and you may be stuck with excess inventory. Did you know that Amazon earns more than one-fifth of its North America retail revenue because local stores can’t forecast accurately? Customers try to purchase the product at a store in these scenarios, but the stores are out-of-stock and so shoppers look to Amazon.

Curve’s Sales Forecasting Solution Results

As made clear in the points above, In the retail industry, time is money. For enterprise retail business owners of rapidly changing consumer goods, speed and accuracy of decision making are critically dependent upon accessing relevant and accurate sales forecasts. Data compiled through machine learning based sales forecasts can translate into the significant and valuable edge needed to win in your marketplace. Curve helps enterprise retail businesses minimize the manual guesswork while gaining a better understanding of future sales patterns and anomalies at a more detailed level.
Curve has helped retail businesses:
– Decrease overstocking by 14%
– Improve understocking by 16%
– Increase sales by 10%
– Grow profitability by 11%

In Conclusion

Developing an effective sales strategy for your online store requires knowledge about e-commerce statistics, trends, and consumer habits. I hope this article gave you the insight to help you get started on that strategy, and that you found some of the answers you were looking for. Have you stumbled upon any surprising e-commerce statistics lately? Reach out to us and let us know!
[ictabs alias=”cta-1″]

Sales are the driving force of any business and being able to accurately forecast your sales is essential in order to for your business to evolve. A proper sales forecast is the foundation of your sale strategy as it reveals valuable info into how your organization needs to plan in order to hit your profit targets, how much profit new products will bring or whether or not you need to hire more staff to reach your sales goals.
For a forecast to be as accurate as possible it needs to be based on as much statistical data as possible, and this is where things get tricky as hundreds of product SKUs may be spread across multiple locations and sales channels.
However, there are ways around this challenge. Stay with us as we are about to reveal the top 10 most effective methods for retail sales forecasting.

The Weighted Pipeline Technique

When you lack relevant statistical data, the best thing to do is to start with probability-based forecasting methods. The easiest to implement ‘probability-based’ method is the weighted pipeline technique. This method requires you to give each sales stage a percentage-based probability of closing. To calculate the weighted value, you multiply the probability of closing to the deal value. This process is repeated for every stage of the sales process, and while there might be some overestimations or underestimations in your forecast, they should weight themselves out. Given the fact that it is based on probability, the accuracy of this method is debatable, but it is a great starting point for a good forecast.

The Length Of The Sales Cycle Method

As the name suggests, the foundation of this method is the time that it usually takes a lead to convert. This method is very easy to implement. All you have to do is divide the average sales cycle, by the amount of time that a sales rep has been working on a lead. Keep in mind that different sales channels will have different cycle lengths, so you will have to separate your leads into different categories, such as referrals, cold emails and so on. Moreover, for this method to work, you need to be very consistent with tracking the first moment that the leads enter the sales pipeline.

The Intuitive Method

This method is based on the opinions of your sales reps. Obviously enough, this method is rather subjective, but you can improve it by building the confidence of your sales team. A great way to do this is to ask them to avoid 50% probabilities, and to motivate their answers.

The Test Market Analysis Technique

This method is ideal for new products and services. To implement this technique, you must segregate your market, and release the new product to a certain group of people. This will give you a glimpse of the market’s response to the new product, while also increase brand awareness. Moreover, this is a great way to gain some feedback, and maybe do some improvements before the official product release.

Curve Can Help Your Sales Team Predict The Unpredictable. Find Out How

The Lead-driven Method

This method is more accurate than the previously discussed technique, but it does require more in-depth data such as the number of leads per months for a given period, the conversion rate by source, and the average sales price by source. To apply this method, you must assign each lead a value, based on the past behavior of similar leads. The challenge with this method is that the final forecast can be affected by changes in the lead generation strategy, so you must be very consistent with keeping your database updated with all the changes in the sales strategy.

Historical Data Techniques

This method relies heavily on historical business performance, assuming that future performance will at least equal the past performance. This method can also incorporate variables such as yearly growth rates, or anticipated growth rates based on recent investments such as new lead generation tools or the growth of the sales team. The problem with this method is that it can’t consider market changes, so it is best to only use historical forecasting as a benchmark for a more analytical forecast.

The Opportunity Stage Method

This method is also based on historical data, in order to estimate the rate of success of each stage of the sales pipeline. Based on that data, each stage will be assigned a percentage that represents the chances of closing the deal. Obviously enough, the further a lead is down the pipeline, the higher the conversion chances will be. Furthermore, the sales stages can differ for each company, based on individual lead generation and sales strategies.

The Opportunity Creation Model

This method uses demographic and behavioral data in order to predict which opportunities are more likely to convert. There are numerous variables that affect the conversion chances of an opportunity, from the size of the business to the decision power of the lead, previous interactions with your company, the leads decision-making process and so on. For this model to be accurate, you must consider the historical data for closed deals, but also for referrals or retained leads.

The Pipeline Technique

This method is a combination of the historical data and the opportunity stage methods. It is rather hard to implement, as it differs for each business, and it requires a lot of hard data. Ideally, you should work with a software solution to handle the complex calculations. The pipeline technique will calculate the chances of conversion of each lead, based on unique company indicators, such as the success rate of each individual sales representative or the individual value of each opportunity.

The Multivariable Approach

This is by far the most accurate forecasting solution, as it incorporates data from other forecasting techniques, such as the cycle length or the opportunity stage. Obviously enough, this approach is harder to implement, and it requires some religious data tracking. Moreover, this is not a method per-say, but a forecasting approach that can bring together numerous methods and complex analyses such as single equation models, cross-correlations, vector autoregressions and so on. The downside of this method is that it requires statistical data specialists as well as advanced analytical tools which can be quite expensive.

Curve’s Machine Learning Based Forecasts

Curve provides retailers and e-Merchants with an unbiased sales prediction technology that delivers clearer estimates of future sales and product inventory predictions.
Curve uses machine-learning based sales prediction technology, allowing companies to accurately forecast sales, products, and support requests, to increase revenue and optimize profitability. Our unique technology goes beyond traditional business intelligence, by recommending the right solutions based on use cases and customer segments.
[ictabs alias=”cta-1″]

When it comes to operating retail sales at a high level in 2019, sales forecasting is critical to your business. Whether you run your business as a brick and mortar or operate strictly online, forecasting technology is a must-have if you want to keep up with your competitors in today’s age.
In the past, we have labored long and hard with sales and inventory forecasting through endless paperwork, spreadsheets, and of course, the cost of the labor to do the forecasting. More recently BI was helmed as an efficient means in which retailers can conduct sales forecasts tasks, however, that still only delivers intelligence based on business hindsight.
Today, businesses that are leading the pack have integrated smart forecasting technology allowing them to operate at with greater efficiency and speed while simultaneously cutting labor costs down dramatically.
So why is Sales Forecasting such an essential part of the retail industry today? Essentially, when done right it will deliver the following benefits for your organization.

  • Improved decision-making about the future
  • Reduction of sales pipeline and forecast risks
  • Alignment of sales quotas and revenue expectations
  • Reduction of time spent planning territory coverage and setting quota assignments
  • Benchmarks that can be used to assess trends in the future
  • Ability to focus on a sales team on high-revenue, high-profit sales pipeline opportunities, resulting in improved win rates

Here are three tips you can take advantage of today if you’re debating the value of new sales forecasting technology:

Leverage The Data You Have

You may think that you need to start collecting and organizing a ton of different metrics before you can really start to forecast your future sales trends effectively. Don’t worry, you don’t need to do a ton of work before getting started. You can use what you have today.
That being said, the greater the data you have, the more accurate your future sales forecasts will be, and you must weigh this against the result you need with the forecasting itself.

Clearer Data, Less Noise

When it comes to utilizing forecasting technology, you will get the best results when you have the right data. The great thing is with our solution Sales Prediction by Curve, you will be assisted in exactly what data is essential for the algorithms to provide clear forecasts.
Additionally, with frictionless sales forecasting, you don’t need to worry about the “behind the scenes” forecasting technology. The sophisticated solution was designed with retailers and sales teams in mind.
In other words, making data available to you should be the primary goal, while understanding the technology behind it shouldn’t set you or your team back.

Data Can Only Take You So Far

Having access to sales forecasts and product demand data is imperative, as described above, however, having a clear plan on how to effectively utilize your sales trends is equally important.
To effectively create a sales forecast action plan we recommend starting with the right set of questions:

  • What are you going to focus on?
  • What are you going to change?
  • In practical terms, what steps are involved?
  • What territories and targets are you going to give each salesperson or team?

Keep in mind that the main purpose of sales forecasting is to provide information that you can use to make intelligent business decisions.
For example, if your forecast indicates a 30% increase in sales of products or services you may wish to begin searching for larger business premises and/or increase additional stock on your hot items. Conversely, a forecast of excess stock or low sales can allow you to mitigate the effect by taking advance measures such as reducing expenses or reorienting your marketing efforts.

In Conclusion

For leading retailers, sales forecasting is not only crucial to surviving, but also to thrive. Not only will it allow you to predict upcoming sales trends accurately and efficiently, but it will allow you to manage product demand.
Over the next decade, the difference between businesses that advance to the next level and those that fall behind will be greatly based upon the businesses that adopt new technology systems in their business.
You can still get an upper hand on the competition by allowing forecasting technology to push your business forward today.

There is nothing worse for an online business than running out of stock or alternatively being stuck with unsold inventory. Both of these scenarios will potentially impact your bottom line.
Regardless of how great your marketing plan is, or how well your eCommerce website may be converting, if you fail to plan ahead, your customer will find an alternative, no question about it.
Sales and inventory forecasting delivers the insights to easily identify obsolete stock in order to promptly liquidate it – thus lowering directly lowering the costs associated with keeping it on your shelves.
For these reasons and so many others, sales and inventory forecasting are extremely important, and the fact is that it really isn’t as complicated as some may assume. Using a combination of sales history and future trends in sales, Curve’s machine-leading based sale prediction solution can supply today’s top retailers with future sales forecasts, based on a variety of point for every product.
An important benefit to leveraging Machine Learning for sales forecasting accuracy is the ability of Machine Learning to ingest data and present that information at a granular level. Today’s leading retailers and marketers are using machine learning to understand, anticipate and act on their sales faster and with more clarity than their competitors.


Sales forecasting is for everyone anyone who has a stake in a retail or eCommerce business and takes interest in predicting its future sales. At Curve, we assist eCommerce and brick-and-mortar merchants make informed business decisions, using our machine learning sales forecasting technology.



  • Optimal Stock – Avoid being out of stock or having too much stock.
  • Advanced Analytics  – Get weekly, monthly and quarterly sales forecasts.
  • Sales Predictions – Totals per category or per individual SKU.
  • Location-based Data –  Forecast your sales by stores or sales channel.
  • Cut time and Cost – Automate time-consuming sales prediction tasks.

To Recap, best practice demand forecasting helps a business succeed in having the right product in the right place, at the right time. Curve helps you forecast sales by time frames and variables such as geographic locations, individual SKUs and much more. Sales forecasting is an essential business technique that delivers a visual representation of where your business is heading. It’s not just a fancy top-level solution, but can be used by everyone who has a stake in the business. If you haven’t done so yet, schedule a free demo with Curve today.