What are you trying to solve for? What is your business objective? These are questions we regularly ask in order to make sure the research we’re doing will provide actionable results for our clients. Sometimes a client will come to us and tell us they’re working on developing messaging for a new product and want to do some historical data analysis to determine what messages will increase likelihood of purchase. Well, unfortunately, the historical research strategy may not be the best way to answer their question. One question we face for many projects is, “should the research look backwards (historical data) OR be forwarding-thinking (predictive and prescriptive)?”
The best way to frame this discussion is to think about the disclaimer mentioned on every investment product…“Past performance is not a guarantee of future results.” In other words, diving into what happened in the past may not be the best predictor of the current or future market.
Both historical and forward-looking strategies have their place. Let’s quickly look at when to look backwards and when to be forward-thinking.
When to look backwards and use historical information
A good time to look backwards is when dealing with mature and stable markets. A mature and stable market doesn’t have a lot of new entrants or new products and changes are evolutionary … often a result of larger economic trends. Another good time is when the market is cyclical in a “normal pattern” and general trends in seasonality, holiday periods, etc. are understood.
Looking backwards is often referred to as descriptive analytics. It basically asks, “How did we do?” It’s a way of reporting what happened. For example, if you sell candy bars, descriptive analytics would tell you how many candy bars you sold in the past 12 months.
When is it dangerous to look too hard at the past?
On the flipside, there are times when looking backward may offer a misleading perspective. For example, when there is a lot of innovation and disruption in a market. Or when there are big “Black Swan” events that are unexpected and shake things up (e.g. the recent recession).
Perhaps the most dangerous time is when YOU (the client) are initiating change. A new strategy or new product can disrupt the historical status quo for YOUR brand disproportionally … and render the past less useful as a predictor of the future.
Visualize a hurricane path prediction. The past is clear; you can see exactly where the hurricane has been. The near term is pretty clear, too, as there is a very narrow range around the predicted path. But as you go further out (2 days … 3 days), the width of the bands that define the predicted path get wider and wider. In other words, the insights from the past become less and less relevant the further out you try and predict.
When to be forward-thinking
Forward-thinking research is often referred to as predictive or prescriptive analytics. Predictive analytics basically tell you, “If trends continue, this will happen…” Predictive analytics takes what you’ve learned from descriptive analytics and uses it to predict future outcomes. It’s great when looking at short-term tactics. For example, “How did our rebate offer on Product X work last year? Do we expect it to perform about the same this year?”
Prescriptive analytics asks, “What should we be doing?” Prescriptive analytics is about identifying what you can say or do to change future outcomes or obtain the optimal outcome. This type of analysis should be based on an underlying experimental design so that we can infer causality. If we change X, then Y is likely to happen. In other words, prescriptive analytics focuses on what can be changed in the ‘here and now’ to alter the current forecast.
- Backward-looking and forward-looking approaches are both valuable but each are best used under specific market conditions and to solve for different business objectives.
- Remember the hurricane path prediction example – historical information is great for looking at where you’ve been or where you might be going in the short-term but the further you look out, the more things not captured in the model can change the outcome.