Forecasting sales is a crucial skill for any business, especially in uncertain and risky situations. How can you use data to make more accurate and confident predictions? In this article, we will show you how to forecast sales using data in the context of forecasting, and share some tips and best practices to improve your results.
Forecasting is the process of estimating future outcomes based on past and present data, trends, and assumptions. Forecasting can help you plan your budget, allocate your resources, set your goals, and evaluate your performance. Forecasting can also help you cope with uncertainty and risk, by providing a range of possible scenarios and outcomes, and allowing you to prepare for different contingencies.
There are multiple levels and angles of sophistication in creating forecasts. At the basic end of the spectrum are judgmental and naive statistical forecasts, such as moving averages and random walks. One level up are statistical forecasts that include additional information such as seasonal patterns and trends. These are however naive in terms of their uncertainty. Range forecasts are generally such forecasts where naive uncertainty is expressed as prediction ranges. Probabilistic forecasts are the most sophisticated in terms of uncertainty. Provided as time series of uncertainty distributions instead of points. AI/ML can be used standalone or add-on to probabilistic methods to include extraneous demand influences.
Sales forecasts inherently involve a degree of uncertainty and risk, as they're essentially predictions of future outcomes based on past and present data, trends, and assumptions. However, they're vital tools for budgeting, resource allocation, goal setting and performance evaluations. To manage the inherent uncertainty and risk, I provide a range of possible scenarios in my forecasts. This allows for preparedness in facing any contingency that may arise. ??
Forecasting is the process of predicting the future. In the B2B world, it is the process of forecasting revenue. Put simply, for a business to run properly, it requires robust revenue forecasts as far out as possible, with well-defined ranges of uncertainty fully communicated. A typical business will need to forecast new business sales, upsells, downgrades, renewals and churn. Parts or all of this will be a sales forecast, which will need to be converted into a revenue forecast for leadership. This requires recognizing the revenue properly, taking account of non-sales revenue, and including other financial adjustments such as for accruals.
Forecasting is predicting the future without total certainty. The biggest risk for most companies in forecasting is late surprises on forward revenue projections. With this in mind, accurate forecasting is critical for every growth company.
Dans le contexte financier, la prévision concerne l'estimation anticipée des performances économiques et des mouvements financiers futurs. Les professionnels de la finance utilisent des méthodes quantitatives et qualitatives pour prédire les tendances des marchés, les cours des actions, les taux d'intérêt, les devises, et d'autres indicateurs financiers. Les outils tels que les modèles financiers, l'analyse technique, et l'évaluation des risques sont employés pour formuler des projections. Ces prévisions aident les investisseurs, les gestionnaires de portefeuille, et les décideurs à prendre des décisions éclairées en matière d'investissement, de gestion des risques, et de stratégies financières.
Data is the foundation of any forecasting method. Data can help you identify patterns, trends, and relationships that affect your sales. Data can also help you validate your assumptions, test your hypotheses, and measure your accuracy. Data can come from various sources, such as your sales records, customer feedback, market research, competitor analysis, and industry reports. The more data you have, the more reliable your forecasts will be.
Data is not optional. Data is critical to get a useful forecast. Without data, your forecast is just an opinion. Data is needed to establish both an estimate of the level of future demand and the uncertainty around it. Both are key outputs of a useful forecast. The level directly drives cycle stock and with it efficiency. Uncertainty drives safety stock and other buffers and with it stability. Combined they drive all net requirements throughout the supply chain. Guesstimating the level is sometimes done without data (judgmental forecasts), but comes at significant impact to margins. Without data, uncertainty cannot be quantified, and this drives most expediting, lost sales, fire-fighting, and other instability in the supply chain.
Para manejar la incertidumbre y el riesgo en los pronósticos de ventas, es importante utilizar datos históricos, proyectar múltiples escenarios, priorizar los indicadores más importantes, identificar variables que influyen en las ventas y recopilar datos históricos. Es importantísimo, reunir información sobre las ventas pasadas, preferiblemente desglosada por producto, canal de ventas, región, etc. Esta información te ayudará a detectar patrones y tendencia.
Your data is a huge part of your competitive advantage, and how you combine it with your domain knowledge will determine your success.
Quality data is the cornerstone of a robust forecasting process. The definition of quality data is: - robust data on the current sales pipeline and connected data (such as contacts, accounts) - robust historic data tracking every change to the sales pipeline, as well as connected data - a system to track each forecasting submission, alongside all the context used to make the decision, subjective judgements made, and who made them - data on "macro" trends over time that are impacting the sales team and company which can be overlaid into future projections
In relation to revenue forecasting, data is critical and without it you can only use experience and 'gut' feel. Data can provide evidence to support assumptions and is critical for any AI predictions as it requires a base line to start.
Creating accurate sales forecasts requires various methods and tools, which generally involve defining the forecasting problem and objective, collecting and preparing data, choosing the right method and tool, building and testing a forecasting model, and generating and communicating the forecasts. Firstly, you need to define what you are trying to predict and why, as well as how far ahead you want to forecast. Secondly, collect reliable data from various sources and clean it. Thirdly, select the most suitable method and tool for your data, objective, and resources. Fourthly, build a mathematical or statistical model that represents the relationship between your sales and factors that affect it. Fifthly, generate forecasts for the future periods with your model and present them in an understandable way. Lastly, provide confidence intervals or error margins to show the uncertainty associated with your forecasts.
When diving into sales forecasting, the key is to adapt your approach to the data's complexity and volume. Initial models can be rapidly developed with Python libraries like darts, statsforecast, and neuralforecast, offering a quick start. As your system evolves, incorporating advanced ML techniques such as Gradient Boosted Trees or Neural Networks like DeepAR, can improve your models accuracy immensely. From my experience, the true value of a forecast lies in its reliability. That's where quantifying uncertainty becomes non-negotiable. Conformal prediction is a model-agnostic concept that's revolutionized this process, providing stakeholders with prediction intervals that bring depth and trust to your forecasts.
The reality is that most CRMs do not provide enough data to support the forecasting of sales. Most companies resort to using spreadsheets to forecast B2B sales and they usually struggle with capturing historical data to manage performance over time. I believe you need additional data points to forecast revenue outside of CRMs and even quantify sales methodologies including extended MEDDPICC
The best way to create accurate sales forecasts is with a process that starts early, repeats often, and delivers context. Forecast with multiple methods and combine them. Method 1: Rollup. Each sales rep categorizes deals as "Commit", "Best Case", etc. Forecasts per rep, per team, up to board level can be done by using historic win rates against those categories. 2: Pipeline-Based Multiply the sum of sales pipeline with historic "pipeline win rate", then by probability of closing on time. 3: Stage-Weighted Multiply sales stage amount by stage win rate 4: AI / stats Use powerful modelling to deliver a forecast. There are many more methods. Ensembling these methods and using them to deliver further context, in turn drives accuracy.
La prévision des ventes à l'aide des données implique l'analyse de paramètres tels que les historiques de ventes, les tendances du marché, et les facteurs saisonniers. Les modèles statistiques et les techniques d'apprentissage automatique sont utilisés pour identifier des corrélations et des schémas. L'intégration de données externes, comme les événements économiques ou les tendances de consommation, améliore la précision des prévisions. En continuant à affiner les modèles avec des données en temps réel, les entreprises peuvent anticiper la demande, ajuster les stocks, et mettre en place des stratégies de marketing plus ciblées pour optimiser les performances de vente.
La gestion des incertitudes dans les prévisions de vente requiert l'utilisation de modèles statistiques solides, l'analyse des tendances du marché, la collecte d'informations en temps réel et la collaboration avec les équipes opérationnelles. L ' adaptabilité aux changements du marché et la révision fréquente des prévisions sont cruciales pour atténuer les risques potentiels.
Forecasting sales using data is not an exact science, but rather an art that requires judgment, experience, and creativity. To improve your forecasting skills, try using multiple methods and sources of data to avoid bias, error, or overfitting in your forecasts. It is also important to update and revise your forecasts regularly to capture new trends, changes, or opportunities in your sales environment. Involving stakeholders and experts in the forecasting process can help gain valuable insights, feedback, and support for your forecasts. Finally, learning from mistakes and successes can help evaluate and improve your forecasting methods, models, and assumptions.
Model monitoring is so critical in forecasting applications! This helps determine what needs to be adapted before down stream users pick up issues.
To manage uncertainty and risk in sales forecasts: I would say 3 main things - Prediction Intervals: provide a range within which future values are expected to fall, reflecting uncertainty. For example, a 95% interval indicates the range within which future sales are likely to fall 95% of the time. - Annotation and Transparency: Clearly document and explain assumptions, data limitations, and model uncertainties. Provide annotations on forecasts to highlight factors affecting reliability and potential risks. - Scenario Mapping: Develop and analyze multiple scenarios (e.g., best-case, worst-case) to evaluate how different variables(market changes) impact forecasts. This helps in planning for various risk levels and making informed decisions.
For B2B revenue forecasting I recommend weekly cadence and forecasting to enable all stakeholders to track progress and leverage data points to support the development of forecasts.
Data-driven forecasting of sales can provide many benefits for your business, such as aiding in decision making and planning. You can use forecasts to optimize pricing, marketing, production, inventory, staffing, or investment strategies. Additionally, forecasting can help you enhance performance and competitiveness in your market by setting realistic goals and tracking progress. Lastly, data-driven forecasting can reduce uncertainty and risk in your sales environment by helping you anticipate and prepare for different scenarios and outcomes.
Without data, forecasting sales is solely based on 'gut' and sentiment. The risk of forecasting without data is borderline incompetence. Data can elicit intelligence that is not easily seen or remembered without it. Thus the rise of revenue forecasting platforms including Clari, Salesloft, Revenue.io and Prevu.