Forecasting Metrics and KPIs

The Best Forecast Metric is the One Everyone Understands The goal of a successful S&OP process is a shared reality – a common view of the potential future demand along with the risks and opportunities that exist within the planning horizon. A common obstacle to creating this shared reality is KPI’s that not all S&OP team members understand. Unless everyone sees how often and how much an item’s demand varies from its forecast, it is nearly impossible for the team to work together to find and resolve any issues. It is also important to understand that forecasts will always vary from actual demand, and how much variation is acceptable for your organization. This is why I am a huge fan of using forecast bias as a common metric for S&OP discussions. It is easy to calculate and explain, and it gives a clear picture of recent forecast performance. As a demand planner, I am supposed to be the expert in forecasting; expecting everyone on the S&OP team to be equally versed in the complexities of forecasting demand is, to me, not realistic. I think of it this way: everyone can understand if an item is over- or under shooting the forecast. And if an item consistently exceeds or falls short of the planned demand, it needs attention. The example below is the bias model I use. It looks at the last 3 months forecast performance and calculates a bias percent for the 3 months in total. (I’m actually allergic to measuring bias month-by-month, as this only shows how much noise the demand contains.) It also shows the variance by month, which allows for examining and explaining one-off large variations in specific periods – variations which can often be explained by unexpected purchases or promotions that were unplanned. The goal here is to quickly find and show the items with the largest consistent variation, so that the team can analyze these items and see if the forecast needs to be adjusted. This helps create the shared reality that everyone can understand and if necessary, agree that action is required.

Most teams struggle with predicting the impact of future bets because it’s too complex and labor-intensive. As a result, they miss out on continuously growing their impact through data-driven learning loops. So I'm trying to figure out a lightweight workflow for teams to simulate the quantitative impact of their future bets. To be practical, the workflow must be conceptually sound while not requiring an onerous amount of data collection or ad hoc data science. The attached gif shows a tool prototype I'm playing with to power this workflow. Here's how I'm thinking this works: (1) Start by building an algebraic KPI tree for your business—this simplifies the impact of various factors into a clear model. An algebraic KPI tree breaks down your primary metric (could be revenue or a customer-oriented north star) into logical components (e.g., Revenue = Visitors * Revenue per visitor). (At DoubleLoop we have AI that helps with fast creation of algebraic KPI trees.) Note: algebraic KPI trees are a good place to start because the relationships are deterministic. While some teams want to create probabilistic models with soft influencer relationships between metrics, it requires more data science resources to get insight from these models. We're working on making this easier with DoubleLoop. (2) For a future period of work (e.g., Q1 2025) plug baseline values into the KPI tree. You could use a previous period's values or just use your judgment to pick something reasonable. It doesn't need to be perfect. (3) Based on the above, you can immediately do sensitivity analysis on the KPI tree to see where 1% changes to metrics will have the highest impact on your primary metric. This helps inform which levers to target with your bets. (4) Add your planned future bets to the canvas and connect each one to the input KPI you think that bet will influence. (5) Add other factors to the KPI tree; e.g., holidays, seasonal influences, or anything external that might impact your metrics. (6) At each connector between bet/factor and KPI, estimate how much you think that bet/factor will change the metric with a percentage. For example, a marketing campaign might both increase the # of new visitors and decrease conversion given lower intent. (7) Based on the formulas of the KPI tree, you will now be able to see the total predicted impact to your primary KPI across your whole portfolio of bets. (8) You will also have a framework to quantify the impact of each of your bets, even when external factors add noise. For example, sales might be down YoY, but you could still show how your bets had a positive impact in the face of headwinds. The first time you try this, your predictions will probably be far off. Your goal is to make better predictions with each cycle. The is unlimited potential to make your predictions more accurate, but this shouldn't stop you from getting started. Would you want to try this workflow for simulating bet impact? Why or why not?

I used to think forecasting was all about crunching numbers. Boy, was I wrong. The game-changer? Integrating non-financial data. Here's what I've learned: 1. Employee satisfaction scores → predict productivity trends 2. Website traffic patterns → indicate future sales 3. Supplier performance metrics → forecast potential disruptions By combining these with traditional financial data, we've improved forecast accuracy by 35%. It's not just about better numbers. It's about making smarter decisions. What non-financial data has surprised you with its predictive power? #FinancialForecasting #CFOStrategy #FractionalCFO #StartupFinance #Growth #CFOInsights #CFOServices #Strategy #SMBgrowth #StrategicFinance #SmallBusinessSupport #StartupFinance #SMBfinance #ScalingUp