The global economy is changing at an unprecedented rate. As a result, the traditional forecasting methods we’ve relied on for years are failing us.
Most of us run data predictions for our businesses manually in a spreadsheet. It takes time to update the forecast every month, and by the time we’re done, it’s probably already out of date. Plus, this method leaves plenty of room for error. We’re human, afterall.
It’s time to try a new approach to business forecasting.
It’s called: “Letting the machines do the hard work for us.”
In this article, we’ll cover:
Here we go.
A quick clarification to get started: is machine learning the same as AI?
Not quite. It’s a subfield of AI.
Machine learning is based on the idea that computer systems can learn from data and adapt without being explicitly programmed. AI refers to the broader concept of using machines to perform human tasks. So AI applies machine learning and other techniques to solve problems.
When it comes to running data predictions, machine learning really shines.
Many studies have compared machine learning forecasting methods with traditional statistical forecasting methods, and found machine learning to consistently produce more accurate outcomes.
Compared to traditional forecasting methods, machine learning algorithms:
Let’s look at these one by one.
The power of machine learning lies in its inherent pattern recognition capabilities. Machine learning predictive models examine the dataset in meticulous detail, analysing the relationship between variables in every possible combination — in a fraction of the time it would take a human with a spreadsheet.
The exciting thing about unsupervised learning in particular, is that you can throw any data at the model and it will figure out how the variables are related, if at all. Take Market Basket Analysis, for example, where machine learning models examine supermarket sales data to identify associations, such as “people who buy X (e.g. cheese) also often buy Y (ham)”.
Note: you can rely on machine learning algorithms to tell you how the data is correlated, but not why. Causation is up to you to interpret, as you are the expert in how your business works. The true magic happens when humans and AI work together. That’s when you get those ‘aha’ moments that lead to more innovative business models.
In today’s environment of uncertainty, the only way to run realistic business forecasts is to update them quickly and frequently. Traditional forecasting methods take too long, are quickly outdated, and leave room for human error.
Machine learning has the power to process vast quantities of data in seconds. It speeds up your data predictions so you can focus on analysing the results.
With machine learning algorithms doing the hard work for you, you can update your forecasts as soon as new information arrives, and make timely decisions.
Traditionally, most businesses look only at their historical data for financial forecasting. Since the pandemic though, many have realised the crucial importance of including external factors, such as market trends, consumer demand, supply, industry pricing, local employment statistics, and other economic data in their forecasts, in order to respond earlier to market cues.
Analysing how external data impacts the business is exactly the kind of complex calculation that machine learning algorithms excel at. With this technology, you can leave it to the machines to discover previously undetected correlations in the data, and focus on responding to what the information is telling you about the future.
By making full use of your own data, plus external market data, you have a better chance of identifying profitable opportunities and avoiding risks.
Over time, with more of the right kind of data, machine learning algorithms learn and improve. Machine learning is ideally suited for data with trends that are steady and systematic. It is designed to learn from the last iteration and adapt automatically, improving the accuracy of the forecasted outcome every time.
This means that there’s no need to manually backtest your predictions with actual data, because the model is automatically doing that for you. It creates a robust, self-learning system that improves in accuracy over time and quickly adapts to new trends.
The key to dealing with the extreme levels of volatility and uncertainty in the global market today is to gather more data more frequently and speed up your forecasting methods so you can keep pace with change.
Machine learning is ideally suited to industries that experience greater levels of uncertainty, because the algorithms are designed to examine exactly that: the rate of change of the data and how variables move together.
Take the fluctuating price of fuel, for example, which is one of the biggest costs for the airline industry. This is where machine learning algorithms truly excel: they can deal with this volatility far better than traditional forecasting methods can.
Not true. There are a plethora of low-code and no-code tools out there that make it easy for anyone to access and harness the power of machine learning without code. gini is one of them.
Run quick, accurate forecasts with machine learning predictive models built into a no-code spreadsheet. Try gini for free >
While machine learning does well with massive amounts of data, it doesn’t mean the technology is solely applicable to industry giants like Amazon, Netflix and Google. Small and medium-sized enterprises have a lot to gain from machine learning — particularly in regard to forecasting.
The truth is, only a small percentage of businesses are making full use of their data, however much they have. We’ve all become very good at collecting data, but precious few of us are putting it to good use. With machine learning, you can incorporate more of your company’s own data, as well as external market data, into your forecasts.
Many of the low-code and no-code tools mentioned above are priced for smaller companies with smaller budgets, ranging from as low as US$20 a month to around US$1,000 a month. It’s all part of the “democratisation of machine learning” movement we’re so obsessed about.
When you’re struggling with cash flow, you’re probably not thinking about adopting new tech, we get it. But the truth is, machine learning has incredible potential to drive cost efficiencies and business continuity — not only because it speeds up the forecasting process and generates early warning signals that help you make better decisions, but also because it has endless applications for improving customer interactions and powering a truly data-centred organisation.
There are many algorithms available for machine learning forecasting, such as Multi-Layer Perceptron (MLP), Time Series, SAGE, the Window Method, and Gaussian Processes.
Here, we’ll focus on time series forecasting, as that’s the most applicable for predicting time indexed data like profit, cash flow, revenue and sales.
The time series forecasting method analyses a sequence of historical data points taken at equally-spaced points in time to make predictions. Usually, time series analysis is divided into four components:
The most common time series models are:
1. ARIMA (Autoregressive Integrated Moving Average) models are linear regression models that use their own lags as predictors. ARIMA models make accurate predictions for short-term forecasts, such as sales and demand.
2. SARIMA (Seasonal Autoregressive Integrated Moving Average) models are seasonal ARIMA models that support uni-variate time series data involving backshifts of the seasonal period. This is best for predicting sales of seasonal products like vegetables, for example.
3. Exponential Smoothing models make predictions by using weighted averages of past observations to predict new values. These models combine trends, seasonality and error components into a smooth calculation.
4. Generalised Additive Models (GAMs) are based on a similar principle as that of regression, except that instead of summing effects of individual predictors, GAMs identify and sum multiple functions, which means they can model more complex patterns, and be averaged to obtain a more generalisable trend line that best fits the data.
Time series machine learning forecasting methods are used by millions of organisations around the world to get more accurate predictions, faster. Companies seeking investment use this approach to predict market share, media publications to predict web traffic, hospitals to predict the severity of seasonal diseases, city governments to predict the number of vehicles on the roads, and so the list goes on.
Now let’s look at how anyone can take advantage of this technology without learning to code.
Think about this for a second: Only 26.9 million people (0.8% of the global workforce) have the ability to code.
That’s why advanced technology is so often monopolised by giant organisations with giant budgets and giant developer teams.
Now think about this: at least 750 million people can use a spreadsheet.
That’s good enough for us! It’s the reason we designed our no-code forecasting tool as a simple spreadsheet add-on...
...with machine learning predictive models built in...
...that 750 million people can use…
...to run more automated, accurate forecasts on their business data…
gini’s customised machine learning model forecasts future trends based on historical time series data using highly predictive and highly adaptable GAM modelling techniques. It uses a divide-and-conquer strategy to model all four components of time series, fitting non-linear trends with yearly, weekly, and daily seasonality, plus holiday effects.
By examining the similarities and correlations between each variable in every possible combination, gini models linear and non-linear complex relationships for a flexible forecast. The model is also robust to outliers, shifting trends and missing data (to a certain extent).
You can use it to run forecasts on any quantitative time series data.
Basically, you export your data from any accounting or CRM software into a spreadsheet as normal, then run forecasts using gini from there.
With a few clicks, you’ll get a baseline forecast with a 95% confidence interval for guidance. gini also assigns an importance score to each variable based on the predictive power it contributes, so you can see which variables have the most impact on the forecasted variable.
To improve the accuracy of your forecasts with no-code machine learning tools like gini, bear these principles in mind:
Let’s look at two examples of insights you can get from your data with gini:
Using this publicly available dataset from a global hotel chain over a three-year period, let’s look at how easy it is to forecast net profit with machine learning using gini.
The dataset includes a variety of revenue and expense variables. On the gini sidebar, choose “Net profit” as your target variable and set the forecast period to, say, 12 months. You can exclude “Total revenue” and “Total expenses” to get a more granular view of how individual variables are impacting profit.
Here’s the forecasted outcome, showing a 95% confidence interval within the dotted lines. As you can see, there’s a peak in profit over the December holiday season, and there’s a predicted dip in the summer of 2021 — a seasonality trend picked up in the historical data.
To increase profit for the summer months of 2021, first have a look at which variables impact profit most. Revenue is a given. Let’s look at performance-based bonuses, instead. What would happen to profit if you capped bonuses at US$90,000 for those three months?
The new scenario-based forecast shows the updated net profit values plotted on the chart in green, with the original baseline forecast visible as a grey dotted line to compare against. As you can see, capping bonuses is likely to make a significant difference to profit.
Other insights you can get from this data:
As capping bonuses may not be great for morale, you could look at increasing marketing spend instead. The scenario planning feature is great for experimenting with different levers to understand the profit drivers for your business. You can also test a variety of budget allocations to see the impact on profit, and run scenarios for headcount changes before making any decisions. All this is possible in just a couple of clicks.
Marketing and sales teams need to know how user acquisition rates are likely to change, and particularly, which marketing activities most impact user acquisition. Let’s walk you through how to run predictions on your user data using machine learning with gini.
This dataset includes user registrations, as well as a variety of marketing expenses over a three-year period.
To find out which marketing expense has the most impact, you can run a forecast on “User registrations”.
gini’s machine learning algorithm analyses each variable’s importance based on the predictive power it contributes, while also accounting for complex interaction patterns.
In this case, the most impactful variable is “Apple Search Ads”, which makes sense.
To explore just how much predictive power it has, you can run a scenario to see the potential impact on user registrations of cutting all other marketing spend to zero for a month.
As you can see, the impact is not that big. So, in this case, you could get away with reducing your marketing budget, without having a significant impact on user registrations, and save a lot of money.
Other insights you can get from this data:
You can use the scenario planning feature to experiment with different marketing budget allocations until you find the optimal ROI. You could also look for correlations between marketing channels, such as Facebook ads and Google ads.
To sum up:
For businesses looking to become more efficient, agile and data-driven in order to better navigate uncertainty, machine learning forecasting offers a powerful solution.
Compared to traditional forecasting methods, machine learning algorithms:
And the good news is that more and more businesses are able to benefit from the speed and accuracy of machine learning, now that affordable no-code solutions like gini are on the market.
Any business can take advantage of this tech, regardless of size, budget, data quantity and talent. All you need is a spreadsheet.
So go on, level up.
Want to see more? Schedule a demo for some one-on-one time with our team.
By clicking submit, you consent to allow gini to store and process the personal information submitted above to provide you the content requested.
Take a comprehensive look at these two solutions—their specific features and their differences. You'll see what to consider before choosing between ERP and EPM for your business processes.
Small business tools can streamline and automate various business processes, helping businesses save time and increase efficiency. These are the 19 you absolutely need!