Happy anniversary Insights! In this latest video, Matt Owen explains top down and bottom up approaches to search campaign forecasting.
Top down and bottom up forecasting, it's a thing that you'll come across a lot when thinking about forecasting and search planning. The top down version is usually like an ordained target. A CEO says we need to sell 10,000 widgets this month because that's what the business needs, and that's often quite useful because it gives us a sense of clarity about what's required and what the business needs are. But, from a point of view of being realistic, it's not necessarily the best way.
The bottom up forecast, sometimes called this sort of demand forecast or forecasting to the market, is the opposite of that approach. In other words, we're looking at preferably a large quantity of known data, known performance trends, and extrapolating that forwards into a view of what can occur over the next 6 months, 12 months, 18 months, or whatever.
Marketing issues are a key thing to bear in mind in the forecasting process. So having kind of agreed that we feel best practice should be a kind of bottom up forecast, a forecast to what's actually going on to what's known, the really major consideration strategically is to get a handle on whether the forecast is really accounting for the sort of brand and category demand.
So it's what I mean by that. Think of an example. Let's say you're forecasting for a bank and people understand what a current account looks like. So there will be a search demand pattern around a brand and the category that it exists within. Now, if that is for that kind of business, we can clearly see whether the brand has grown over time historically, whether the category has grown over time or indeed shrunk over time historically, and kind of map that into the future based on those search trends.
The other side of that is if you're looking at a much less well-known brand or a very sort of fledging category. So let's take an example of an innovative fitness device, for example. In that situation, your forecasting assumption is very different because the brand may not exist, but there may be plans to invest in brand. The category may not exist, but, due to category popularity, the category may, in fact, grow significantly over time. So that brand/category debate is a really key issue to understand and to put into the forecasting mix as soon as possible.
A great adjunct to that brand conversation is around brand marketing. So some brands plan to invest significant sums in above-the-line marketing, and that can be a very regular cycle. So certain brands invest at Valentine's Day, Easter, Christmas, and so on. So the first thing to bear in mind, from a forecasting perspective, is how well-known that plan is in advance and if we know that there is an appetite to invest significantly at certain points in the year to build that into the forecasting relation to the brand element of your forecast.
Another angle is to think around whether the brand itself is going to invest in tactical marketing at certain points and how, for example, regionalized that might be. So actually, that regional local angle is a key thing. So we do see examples where brands invest in things like outdoor and radio in a highly regional way, and that can be built into the forecast in a very granular and very realistic manner as well.
So far we've talked about some broader marketing issues around brand and category and above-the-line stuff, but the real engine room of the forecast is going to be around data, and the key issue here is having as much data as you possibly can. The more we have, the better the forecast is going to be, the more robust it's going to be. So, in a best practice scenario, we would like to pull data from analytics, from networks about impressions, clicks, click-through rates, conversion rates, and really pile all that data into a forecast and to look back as far as we possibly can, because the better the look back window, the more accurate, the more we can smooth the data out over time.
So, ideally, at least two years' worth of historical data is great because it allows us to identify either seasonality or exceptions to that seasonality. So having loads of data is great.
The other thing to bear in mind from a data point of view is understanding where perhaps goal posts might have moved in that process. So in some situations that won't be relevant. Impression data kind of is impression data, as is click data, but conversion data could change due to some change in internal accounting policy. So your client might suddenly regard a new client as somebody who has been taken on board within the last 12 months as opposed to the last 36 months, and that could significantly change the conversion data that you're basing your forecast on. So understanding that goal post and where those goal posts have moved is also a key part of the accurate data analysis process.
So seasonality plays a major, major role in developing a really good bottom up forecast, and there are really obvious things. So for retailers, for example, Christmas is a big time. Boxing Day is a big time. Easter could be a great time as well. But we see unexpected seasonality across all kinds of different categories, some at a sort of quarterly level, some at a monthly level, some even weekly or day of week or time of day.
So really getting to grips with seasonality and building that into the forecasting is a really significant thing, especially at a run rate level. In other words, if your client, if our client is very sensitive to sales fluctuation, being able to map the run rate against known seasonality is an absolutely critical component of successful forecasting.
The summary for me is really all about having as much data as possible, is the first point. There's a great sort of saying. The acronym is GIGO, but it means garbage in, garbage out. You'll never get a good forecast unless you put good data into that forecast in the first instance, so that's the first thing.
The second thing is, that when the forecast is being created, to have agreement that the methodology is appropriate to the business needs. So going back to our top down versus bottom up, if ultimately the CEOs going to say, "I don't care; I want 10,000," then the whole process is in fact a waste of time, or you might spend valuable time batting the numbers backwards and forwards in a really sort of unconstructive manner.
I think the third point is to make sure that the methodology behind the forecast is clearly understood and agreed before going ahead. That way, everyone understands what the output is going to be, the basis for the calculation, and they will buy into the pragmatism and the ultimate value of the forecast.