AI-powered Optimal Forecasts for all Seasons
Demand Planners have one of the toughest and thankless jobs. They are expected to deliver precise and accurate estimates of future demand every month in face of highly volatile and seemingly unknowable demand across thousands of Product/Location level planning combinations. What makes their job even more complex is inflexible and primitive planning solutions from the pre-AI era, which are only marginally better than Microsoft Excel. Demand Planners are unfairly judged either way: either because they went conservative resulting in stockouts and lost sales or because they went aggressive resulting in excess inventory. They just can’t thread the needle every month across thousands of combinations, but that’s the expectation from them.
Demand Planners need a Digital Planning Twin: an AI engine that can evaluate demand patterns at a granular Product/Location level relevant for Supply Chain decision-making (what inventory where when), learn from both historical data as well as forward-looking demand drivers to produce an Optimal Forecast which both senses and optimally shapes Demand.
Pre-AI era S&OP processes and supporting tools still are trapped in sequential processes and resulting silo efforts: Demand Planning & Review first to lock Unconstrained Consensus Demand, followed by Inventory & Supply Planning to identify the least cost way to respond to the locked demand signal. This rigid sequential approach to planning results in silo optimization, and leaves significant margin optimization value on the table. Imagine if Supply constraints and current excess inventory could inform your Demand Shaping activities (pricing, promotions, new product launch schedules, etc.) to generate a demand forecast and a supply plan simultaneously that optimizes Enterprise level P&L, Balance Sheet, and Cash Flows.
The first step in this AI-powered Value Chain Optimization process is to understand Demand Patterns deeply and identify Forecastability aspects in terms of Planning Hierarchies. Given product portfolio proliferation, it is typical to see high intermittency levels at an individual Product / Location or Product / Customer level. It's traditional wisdom that Demand Signals should be generated at higher levels in the Hierarchy to aggregate and stabilize the historical data, thus making the time series more Forecastable. However, most companies are stuck getting such aggregated forecasts to disaggregate accurately down to the level at which Operational decisions are to be made: Product/Location. Vyan.ai enables Forecast Level Optimization by evaluating all combinations of Hierarchy Levels (e.g., Product ID - Customer ID, Product Group - Customer Group, Product Family - Customer Region, Business Unit - Customer Group, etc.), then leverages the power of AI to disaggregate all such top-down forecasts to Product/Customer or Product/Location level accurately.
The second step in this AI-powered Value Chain Optimization process is to blend Demand Forecast across a variety of Forecast Models: including traditional models, such as Exponential Smoothing, Seasonal ARIMA, as well as Machine Learning models, such as AdaBoost, Neural Networks, etc. Vyan.ai differentiates itself by autonomously and optimally blending all machine forecasts (across all models and bottoms-up / top-down forecasts) as well as human forecasts from the last cycle (prior cycle sales forecast, prior cycle marketing forecast, prior cycle demand planner forecast, prior cycle consensus demand plan, etc.). Most Integrated Business Planning solutions including market leaders are missing this optimal blending capability and expect Demand Planners to provide % weights explicitly to produce a hybrid forecast mindlessly. There are multiple challenges with such planner-dependent approaches: 1. we cannot expect the Demand Planner to know how to provide weights across tens or hundreds of forecast candidates in the forecast competition; 2. such weights will most likely be different for different Product/Customer or Product/Location level planning combinations; and 3. such weights will also likely be different across different lags (lag 1 vs. lag 3 vs. lag 5). This last point is very important as we align the right lag forecast for a given product based on the replenishment lead time.
The 'pick-best' functionality available in the pre-AI solutions is limited to picking one specific model as opposed to optimally blending across a large set of models; a large set of top-down forecasts for a given model; across lags; and across both machine and human forecasts. Vyan.ai delivers a glass-box experience for Demand Planners, as they can see all the specific weights given to each lag / forecast level / forecast model as well as disaggregation factors used to bring aggregate forecasts down to granular levels.
This AI-powered Optimal Forecast from Vyan.ai cuts 10-25% absolute error at the granular levels of Product/Customer or Product/Location level forecast. Most organizations game the Forecast Error metric by reporting aggregate level forecast performance, which is meaningless, as we do not ship product families globally once every month, but need an accurate demand signal at product / location / week level to have any chance of optimizing inventory or supply response in a way that helps with the enterprise level margin and can sustain over time. The other aspect of metric gaming is to report Forecast Accuracy instead of Forecast Error. Most enterprises we deal with cap forecast error artificially at 100% error, and report 0% accuracy. In reality, error of 100% vs. 500% have very different cost impacts for the supply chain. Analytics teams in most enterprises are well aware of the high raw error at granular levels, it is usually a number that is guaranteed to ruffle executives, and hence poor forecasting techniques hide behind dressed-up metrics: forecast accuracy at aggregate levels is reported as 70-80% or even higher. In reality, most enterprises struggle to cut raw granular level error below 50%, this is where Vyan.ai comes in. It cuts error where it helps, it does this autonomously, and it sustains the improvement over time.
Vyan.ai does not stop at generating an Ai-powered Optimal Forecast. It also tracks overrides made to this touchless forecast to learn when sales / marketing / demand planners are successful with their overrides. Vyan.ai continually learns from humans to produce a highly accurate starting point for Demand Review. Vyan.ai also alerts Planners real-time when their overrides are likely to destroy value (increase forecast error), based on override direction, scale, and underlying error/noise in the specific planning combination.
You can reach out to Vyan team at info@vyan.ai to learn more about AI-powered Demand Forecasting Best Practices and to see Vyan.ai in action. Our team can also conduct a Proof of Value with your data, where we conclusively prove the forecast error reduction potential compared to your forecast performance baseline. We help you understand the Cost Of Forecast Error (COFE), which gives you a data-driven business case for Demand Planning Transformation.
