You have to build a forecast and work on predicting what the turnover of the company or a particular line of revenue will look like in the next few months. Do you need to use time series models to build your forecast - or do you think that a more let's say 'manual' approach would be enough for your forecasting? Start by reading our central post on sales forecasting.
Here are all the good reasons why you would need to use a forecasting algorithm such as Prophet to help you with your forecating process. There a many powerful tools to develop automated or semi-automated forecasts, leveraging the power of time series and machine learning to detect patterns in historical data, to foresee what the future will look like. The beauty of using statistics is that this is methodology can be applied even if you don't know anything about a market, a company or a product line.
DataInsightOut makes this forecasting process very easy, just by uploading a CSV file filled up with the data you want to forecast, in case you need something easy to use.
I have *many* forecasts to generate
Let's say you need to build bottom-up forecasts to finally aggregate them.
A forecasting tool will help you save a LOT of time. If you decide to forecast the activity at a customer level, product level, account manager portfolio level, chances are that you need a tool helping to come quickly to a relatively fair number. Unless you're ready to spend day and nights doing it, you will need to get some statistical tool which will extract the very essence of them: strip out seasonality, get the general trend, include holidays and bank holidays.
Manually reviewing the data will be a critical and mandatory step still, but instead of spending 20 minutes on each forecast, having used a statistically driven forecast methodology will reduce the time required by this step dramatically.
Upload a CSV and start generating sales forecasts at scale with time series analysis.
Imagine now that your are in charge of managing an e-commerce website. Imagine that you manager is asking you to develop a forecast for next year at a product category level. So if your business is about selling shoes online, you might end-up having multiple forecasts: one for the "women boots", another forecast for "men loefer" and so on, assuming that men and women have distinct consumption habits and shoe model style.
Get a reliable base forecast to start
So why your sales forecast need to be right? In the case of the e-commerce shop, this forecast will have multiple direct consequences: in terms of purchasing, you'll need to have the quantity right, otherwise you might end up over-forecasting which will impact directly the cashflow and the working capital requirement of your company, ending up in massive sales and overall lower margin. If you under-forecast, you'll miss sales and probably increase your customer's churn rate as many of them will be frustrated not having the pair of shoes they want (or feet size).
Having an automated statistical based forecast tool at hand is like having a robot which will do 80% of the job for you - and you'll however to finish the 20% remaining and get to assess the output and correct them, based on your experience and company strategic goals.
A great amount of time will be saved if you run automated forecasts!
I have a limited experience in forecating - and no data expert available
... and actually using a model backed-up with some geeky maths we'll definitely help.
Forecasting requires skills
Never heard about time series? Not really sure if what you're projecting actually makes sense - and is the projection would be correct using past data? Where should you even start?
Times series analysis is almost a full time job in large listed companies. Getting the financial forecast right is important, especially when you commit a target number to investors. These forecasting specialists need to spend a huge amount of their time understanding underlying trends in the data and detect growth or softness pattern, to be able to predict the future.
You have financial analysts specialized in forecasting, reaching out throughout the company to capture as many as possible business variations and factor them into their prediction to get a little variations as possible at the end of the day.
Understand calendar effects
One of the aspect of analyzing time series, is to understand the year over year growth and understand all the calendar effects: 1st day of the week versus weekend activity, seasonality patterns such as sales or sport event which can affect the prediction, leap year, early month effect versus end of the month - when customers get their paycheck.
You might also need to breakdown your overall forecast into smaller components which will allow you to understand the underlying growth drivers. Let's say a new product line which is skyrocketting in terms of performance or a special product following a distinct pattern. This forecasting exercise requires a very deep comprehension of the fine mechanisms and drivers of your activity - hence a lot of time invested.
Using a statistical forecast will allow you to spare some time investigating these year over year growth effects, calendar effects such as leap year or holidays to take into consideration or weekly seasonal patterns. You'll still be able to interpert the result to see if it makes sense or not and course-correct the results, based on your overall business knowledge. You can also breakdown your global forecast dataset into multiple sub-dataset which you will forecast separately and then aggregate into a global one.
I need to cross-check what I already forecasted
You already worked on a forecast but would like some neutral third-party check. Using a forecast model is good to cross-check if you haven't exagerated or under-assessed a trend. Cross-checking, validating... Forecasting a general level of activity is not a hard-science; lots of variability can happen. You can forget a parameter, have over-estimated a general trend or got the seasonality wrong. Getting a double check, like an external eye, is always a good idea.
Using multiple ways to forecast is definitely a sound approach, especially if you're operating in a tight margin business where demand forecast needs to be accurate. A forecast answer is never black or right.
Forecast accurracy can also be checked if you use some techniques like the MAPE score. These techniques allows you to perform "back-test" analysis, ensuring that the variation between forecasted data and actuals will be as minimal as possible. You can know more about how to check that a forecast is accurate in this post.
I need to have a granular forecast, at a daily or weekly level
If all you need is a forecast aggregated at a monthly level, then you might think - why bothering using such a tool. A quick and dirty projection might do the job, right? When accuracy is not required, for a high level projection and quick estimation, some mental arithmetic are enough. To quickly estimate the ramp of a new product or a new line of revenue, then Excel and a little bit of brain juice will be ok.
So in some cases, definitely.
Get forecasts at daily level
When you need however to do a forecast at a daily level, then a forecasting tool can help you save time. If you want something more robust, understand if you're on track or not at a daily or weekly level, get more accuracy on your monthly aggregated forecast by summing up all the tools you have, a more statistically driven approach is required. Breaking down a forecast at a daily level can be a very tedious task to perform - and to get it right. Weekdays and weekends patterns will be included.
You will need to run a forecast many times a year
Why you need to re-forecast
Your business is changing - so will be your forecast. Every time you will have a major evolution of your activity, you might wonder what is its impact on the overall performance of your activity. Some changes are planned in advanced and their effect can be foreseen and modelized such as the launch of a new product or the recent penetration of a new market.
However, some macroeconomics such as the COVID19 will require to re-forecast your activity and get quickly new sales predictions to understand what are the new trends - and what your should expect in the future.
Re-run a forecast easily
A quarterly review of your prediction to understand long term trends and variations - Running a forecast on the same data perimeter (E.g - a product line or a country) is also a great thing to understand if the trend is upwards or downwards over time. You might indeed lose the big picture while running your daily operations and not be aware of the latest evolution of a customer portfolio for instance. Having a statistical forecast can help you reveal observed shifts in trends. Establishing comparisons and understand why the forecast has revised itself upwards or downards help deciding if further business actions are required.
Drop us an email if you want to know more about how DataInsightOut can help you building a sales forecast!