Forecasting is often thought of as a solo pursuit. Yet experience in the recent VN1 forecasting competition demonstrated again to me how teamwork can significantly enhance accuracy.
1️⃣ Experience in the VN1
In the VN1 forecasting competition, Jakub Figura and I competed as a duo, and our collaboration helped us win. We both believe that together we delivered a more accurate model than either of us could have done alone. Achieving this level of accuracy individually might have been possible, but it would have required significantly more time and carried a lower likelihood of success.
We challenged each other, testing assumptions and refining methods. We approached the problem from different perspectives, offering techniques and best practice from our individual expertise. This made our combined forecasts more resilient to potential inaccuracy and bias.
This was most apparent when we assembled and critiqued our final submission. We made decisions together on how to treat outliers, whether we should zero out very low forecasts, whether we should perform smoothing and also how to ensemble our best models.
2️⃣ Real world forecasting
In a business context, accurate forecasting requires a great deal of relationship building, for example to acquire accurate data; to understand historical variation; and to obtain reliable past and future values of exogenous variables, such as promo activity.
I have often found myself dissatisfied with the accuracy of future values of exogenous variables, causing errors in my models due to Garbage-In, Garbage-Out. It has been managing relationships and negotiation with other functions that helped improve accuracy.
Further, having a colleague acting as a challenger can be hugely useful for identifying issues with forecasts. And building good relationships with users of your forecasts can help inspire confidence and increase buy-in.
3️⃣ Improving forecasting through relationships
📚 Build collaborative networks, both within and outside your organisation, to promote shared learning. Seek support for specific challenges and invest in opportunities for joint learning.
🤝 Foster relationships with those who can offer insight into what you’re forecasting and provide reliable, useful data.
🗯️ Encourage open dialogue & critique. Create an environment where constructive feedback is welcomed, strengthening your models and uncovering areas for improvement.
🔍 Engage with users of the forecasts. Understand their needs, address concerns, and ensure forecasts are actionable, relevant, and clearly communicated.
Forecasting is not just a technical exercise, it’s also a human one. Focus on relationships and fostering collaboration and unlock insights that drive better accuracy.