Most companies wish they had a magic crystal ball to allow them to see into the future. In reality, the best anyone can do is take a look back at what has happened to try and forecast what might be — humans and algorithms alike. This is not easy under normal circumstances, let alone the present. COVID-19 has affected the way we live, work, socialise and eat — the last one being relevant to Gousto.
With so much unpredictability, how do we begin to create a successful forecast?
In this post, I look at what you can do to increase your chances of creating a successful forecast. There is plenty of technical content on forecasting out there written by very clever people; I want to focus on the process that sits around it!
Forecasting at Gousto
Gousto produces several forecasts at various frequencies and for various purposes. These range from short-term forecasts for buying food stock, to long-term forecasts which span several years.
COVID-19 has impacted every forecast at every time horizon. We saw spikes in demand following initial lockdown announcements to long-term customer behaviour changes driving underlying demand.
Below are 6 tips that have enabled Gousto to achieve amazing success during this period.
But first some words of wisdom from a previous manager, which help set the context:
“Forecasting is a tricky business, there are only two outcomes. You’re either wrong or you’re lucky”
I can’t stress this one enough! Forecasting is often a black box. You feed in some inputs and produce some outputs, which more often than not, are rarely questioned.
During the height of the initial lockdown in March 2020, we went as far as having daily calls to review the forecast. These calls included everyone from the CFO to supervisors in the factory. Discussions included changes to the forecast, reasons for those changes and impact on the most affected area of the business, Operations.
Having everyone in the same room discussing the forecast created transparency. It allowed people to understand the challenges of forecasting in a volatile environment. And as a result, we made decisions that allowed for a larger margin of error and the greatest chance of success.
2. Understand your purpose & priority
As mentioned above, we have several forecasts with varying purposes. During times of volatility, it’s important to understand where the highest priority for accuracy lies. In most cases, this is the forecast produced for the shortest time horizon. This was the case for us during COVID-19 due to the sudden increase in demand and its impact on Operations.
Put your focus here and ensure you’re doing everything you can to support the areas most affected by volatility. For context, we didn’t revisit our long-term financial forecast until July 2020, almost 4 months after the lockdown began.
3. Reduce your cycle time
Cycle time refers to the amount of time between forecasts for the same period, i.e. a re-forecast.
Businesses set cycle times dependent on their proposition. For example, in pharmaceuticals, cycle times can be over 3 months due to the length of time taken to produce the product. At Gousto, we re-forecast everything on a minimum weekly basis as the menu changes, creating new data ready for the forecast.
However, during COVID-19, we reduced our cycle time to daily. We updated forecasts multiple times a day depending on Government announcements, issues with supply or anything else that might influence the final demand. Reducing cycle time allows you to be flexible and adapt to changes in circumstances and gives everyone the most up to date information.
Again, I would stress communication is key here. There is a balance to be drawn between being flexible and creating confusion. It’s important your comms are clear and any impact is understood.
4. Scenario Modelling
Scenario modelling was a habit Gousto had already adopted prior to COVID-19, but it became even more useful during the lockdown period.
It refers to forecasting several eventual possibilities depending on changes to your input variables. For example, we had different scenarios modelled depending on the duration of lockdown, the possibility of future lockdowns as well as scenarios dependent on our own production capacity.
A word of advice here, scenarios can proliferate and you can lose track of what’s still a possibility and what needs to be ruled out. Try to minimise the number of scenarios you model and create a labelling convention that makes each scenario easy to understand.
5. Create an understanding
Once you have tackled any immediate concerns around forecasting from an unpredictable situation, the next step is to understand how the external environment has impacted your input variables.
In our case, lockdown created 3 main changes in customer behaviour:
- Increased demand from new customers
- Reduced churn from active customers
- Reactivation from previously churned customers
Using this new understanding, we were able to remodel our short-term and long-term forecasts and improve our future forecasting accuracy.
6. Spread responsibility (diversify your risk)
As my ex-manager said, you’re either wrong or lucky with forecasting. In an unpredictable scenario like COVID, the chances of you being lucky decrease massively.
When the inputs are so unknown, it’s best to get consensus from the people who the forecast ultimately affects. If everyone is aligned on the inputs (and your forecasting model has previously been verified) then the outputs are just part of a calculation.
Getting input from senior leaders or a wide range of people allows you to reduce anomalies, excessive noise in your input variables which will, in turn, improve your output.
As a closing note, I hope you found the above tips helpful. Hopefully, you don’t need to employ them under another pandemic.