A lot of people talk about AI in planning - we apply AI.
Almost all of our clients faced the same issue before they decided for smartPM.solutions: Time consuming data-collecting processes and inaccurate forecasts. According to BARC’s 2019 Planning and Forecasting Study this is a problem known and identified by 65% of their respondents (n=316).
From a technical ‘how to …’ perspective, a typical regression problem would be revenue forecasting. Forecasts can be generated using statistical methods, usually starting with simple methods like Linear Trends. More advanced methods like Holt-Winters Exponential Smoothing or Auto-Regressive Integrated Moving Average (ARIMA) might be the next evolution step. In some cases, where data is almost linear or has a nice seasonal pattern, statistical methods do a ‘good enough’ job and deliver reliable forecasts. Unfortunately, in most cases the pattern is way more complex, and/or some external factors need to be included in the forecast algorithms. Moreover, statistical modelling is always based on a number of assumptions, like linear relation between dependent and independent variables, independence of observations and a normal distribution of errors. Machine learning methods can work better without these assumptions, though some methods like Neural Networks are really ‘hungry’ for train data. smartPM.solutions uses both statistical and AI methods and picks those who perform best on particular data, based on train error. With our analysis approach, more than 40 networks with different structures are tested and the best one for a solid forecast is selected automatically.
By using the smartPM Sales Performance Forecasting Module, different forecast outcomes are clearly visualized, suggesting the forecast with the best fit. Scenario based what-if-analysis with auto-detection of the best forecast method fit allows for better decisions. Big data sets can be easily handled and the forecast precision is notable. Another big advantage of machine learning methods is, that a lot of parameters like region, company size, products etc. are considered and weighted. The most relevant parameters are automatically selected by mathematical methods like Principal Component Analysis to maximize accuracy and avoid over-fitting. Plans and forecasts are created at the push of a button, hence allowing for true agile decision making.
Especially in times of increased volatility and uncertainty, precise forecasts can be crucial to make appropriate decisions rapidly.
Learn how we have helped other companies
Sales Performance: Forecast closing chances for leads and opportunities
smartPM.Solutions use Machine Learning and Artificial Intelligence methods to deliver the most accurate probabilities of winning an opportunity and for estimating the actual closing month. Contrary to common closing month estimation made by the opportunity owners, based only on their experience and subjective feelings, ML methods use all the relevant factors and previous experience of all opportunity owners to deliver the most accurate winning probability estimation. We leverage our expert mathematical and statistical knowledge and tools like R and Python with their most up-to-date libraries.
Flexibility and robustness
Depending on the data and its dimensionality, some methods may be more, some less accurate. For the winning probability estimation, up to three different ML-methods are compared. SPM experts who think that artificial intelligence should be supported by human intelligence, may fine tune the methods by choosing the most appropriate kernel and activation functions or number of layers and nodes for Neural Networks. Users with less mathematical background may use the default functions to let the system automatically choose the most accurate method.
While choosing the most appropriate method for revenue forecasting, there is no ultimate recipe – it is all depends on the data. That is why we use different statistical and ML-methods to be able to compare them and choose the most accurate one for the future forecasting. Also here we give the freedom of fine-tuning and adjustment of parameters of the number of trained networks, or letting the system to make the entire work for you.
In many cases, sales revenue depends on external factors like weather or some macroeconomic values. The multivariate forecasting method allows for taking several external factors into account. AI will automatically pick the most relevant of them and adjust the forecast to make it even more accurate.