Full path: Home, Digital, AI, Insights, and Solutions, Demand Forecasting
How we can help
You can’t forecast demand in a silo. Our systemic approach and analytics expertise span every business function from supply chains to marketing. We’ll help you focus your efforts, improve your approach, test your results, and scale a superior forecasting capability.
The most intelligent demand algorithms are only as smart as the data that goes into them and how well they are understood by their users. Our end-to-end approach considers all these factors, consistently generating more effective plans and valuable business outcomes.
Most companies build strategies on heavily-biased, “most likely” assessments of the future based on presynthesized analyses and reports. We uncover creative primary sources and model scenarios along with point forecasts. We then apply a disciplined process to add judgment while limiting bias. The result: greater alignment on key decisions with better hedging on important risks.
We identify the places where forecast improvement will produce the most value
We develop better forecasts with a combination of quality data, algorithms, and machine learning
We test improved forecasts with in-market trials to measure real-world business impact
We distill requirements, support vendor selection, and mentor in-house staff for a smooth roll-out
increase in group operating profit
reduced time to market
increase in stores ordering daily, increasing product freshness
Grupo Bimbo, a $15 billion packaged food company, wished to create a system that would reduce waste while still driving growth. We helped them achieve this goal by building a new front line tool and an algorithm to adjust behaviors and improve order accuracy. By taking the guesswork out of ordering, we helped exceed the organization’s initial waste reduction target by 30%. Time and energy previously spent on making and adjusting orders was now redirected to focus on growth.
Results:
CPGCo is an international market leader in consumer packaged goods with more than $30 billion in annual revenue. But the company’s potential was hampered by decentralized forecasting models and siloed data hubs that made it impossible to generate accurate demand forecasts or draw quality, enterprise-wide insights. Using our artificial intelligence and machine learning expertise, we helped develop an AI-fueled, top-line forecasting system that can deliver an 18-month outlook for 180+ markets, by month. We also automated data feeds, designed a user interface portal, and trained the team to create seamless data flows across the organization.
Results:
PetChem Co, a large petrochemical company in Latin America, relied on a labor-intensive demand forecasting process that used rudimentary statistics, leading to poor accuracy. In turn, this generated high inventory levels. We helped revise their process to include relevant input, such as pricing strategy, and other market dynamics, like macroeconomic changes, import prices and industry indicators. Incorporating user input, updating analytics techniques, and identifying improvement levers for implementation reduced over 80 drivers of forecasting errors.
Results:
Grocery Co, a regional APAC supermarket chain with over 100 outlets, struggled with maintaining consistent stock availability. We helped them reduce the number of out-of-stock scenarios by adopting a data-driven and scientific approach to demand forecasting. A data-driven model now forecasts the top 1000 SKUs each week, decreasing both labor costs and the number of out-of-stock SKUs. We also helped develop smaller grab-and-go stores with fewer SKUs, and used space optimization to examine under-performing stores to further lower costs.
Results:
Grupo Bimbo, a $15 billion packaged food company, wished to create a system that would reduce waste while still driving growth. We helped them achieve this goal by building a new front line tool and an algorithm to adjust behaviors and improve order accuracy. By taking the guesswork out of ordering, we helped exceed the organization’s initial waste reduction target by 30%. Time and energy previously spent on making and adjusting orders was now redirected to focus on growth.
Results:
CPGCo is an international market leader in consumer packaged goods with more than $30 billion in annual revenue. But the company’s potential was hampered by decentralized forecasting models and siloed data hubs that made it impossible to generate accurate demand forecasts or draw quality, enterprise-wide insights. Using our artificial intelligence and machine learning expertise, we helped develop an AI-fueled, top-line forecasting system that can deliver an 18-month outlook for 180+ markets, by month. We also automated data feeds, designed a user interface portal, and trained the team to create seamless data flows across the organization.
Results:
PetChem Co, a large petrochemical company in Latin America, relied on a labor-intensive demand forecasting process that used rudimentary statistics, leading to poor accuracy. In turn, this generated high inventory levels. We helped revise their process to include relevant input, such as pricing strategy, and other market dynamics, like macroeconomic changes, import prices and industry indicators. Incorporating user input, updating analytics techniques, and identifying improvement levers for implementation reduced over 80 drivers of forecasting errors.
Results:
Grocery Co, a regional APAC supermarket chain with over 100 outlets, struggled with maintaining consistent stock availability. We helped them reduce the number of out-of-stock scenarios by adopting a data-driven and scientific approach to demand forecasting. A data-driven model now forecasts the top 1000 SKUs each week, decreasing both labor costs and the number of out-of-stock SKUs. We also helped develop smaller grab-and-go stores with fewer SKUs, and used space optimization to examine under-performing stores to further lower costs.
Results:
A large consumer health company found that its typical demand forecasting process could not stand up to pandemic-fueled supply constraints and overinflated orders. To improve accuracy, our hybrid team of consultants and Advanced Analytics experts helped ConsumerHealthCo establish a single, stable source of truth. After identifying issues such as data governance gaps and manual adjustments, we worked with the client to redesign the data collection and modeling process, using machine learning to identify patterns and select better algorithms. We tested a prototype with sample SKUs and continually drew on demand planners’ feedback over time to refine the process.
Results:
Demand forecasts break down during disruptions not because of the algorithm, but the data.
Standard forecasting pipelines rely on stable, recurring demand signals. But those patterns become irrelevant the moment conditions shift. Supply shocks, behavioral shifts, and external events scramble the signals that complex algorithms—machine learning models in particular—depend on. The more automation has displaced human judgment, the more correction disruptions require.
The fix isn’t to abandon modeling. It’s to rebuild for adaptability. Here are four ways to do so:
To improve forecast accuracy, effective teams prioritize SKUs and markets by classifying demand series according to volume, which indicates importance, and volatility, which indicates forecastability. After all, not all forecast errors carry the same cost. Demand series classification can focus improvement efforts where accuracy gains deliver the most business value.
Within that framework, high-volume, high-volatility demand series come first, since that’s where better forecasting pays off most. High-volume, low-volatility series follow, and low-volume, low-volatility, then low-volume, high-volatility last.
This classification should account for special demand signals. New product series have limited history and need tailored approaches such as attribute-based models or life cycle analysis. Retired product series may need to be filtered out or adjusted. As the assortment changes, so does the prioritization.
Determining whether to build vs. buy a demand forecasting capability is a critical decision, and the right answer depends on the organization. A custom build is as much a software engineering challenge as a statistical one. Without careful planning, cloud computing costs can quickly outpace expectations. The most effective approach is to validate before you commit to a path by running early prototyping efforts with defined accuracy targets.
Even the most accurate forecast algorithm won’t provide value if frontline workers don't trust it, understand it, or find it difficult to use. When they don’t, they often default to their gut feeling. This is why demand forecasting is a change management challenge as much as a technical one.
Behavior change starts with the tool itself. Making it easy to act on the forecast is a must-have design principle. A purpose-built interface designed to surface the right information clearly and quickly at the moment of decision is what translates model accuracy into actual ordering behavior.
Extensive staff training and change management programs are equally critical. In fact, they are prerequisites to scaling. Teams need the skills not only to produce and interpret forecasts, but also to understand and apply them. And when new tools are in trial, giving users the opportunity to provide feedback and adjust based on their understanding of the business doesn’t just build trust—it improves the model.
External data sources can become more valuable when demand shifts suddenly and historical internal data becomes less useful. External data can give forecasting models a more reliable read of what consumers are doing right now, improving accuracy when familiar demand patterns break down.
The guiding principle is that better data beats better algorithms. It’s probably the single most important investment a company can make in its forecasting capability.
Some of the external data categories with the strongest track record include:
