By Sue MacLure, Head of Data at Communisis
The perennial question about putting POS in retail outlets for FMCG is “how can I measure ROI?” Where should we invest in big and bold displays, what is the driver of success – brand splashing or simple price mechanic. And that’s on an over-arching performance basis, it would be even better to know what sort of POS works in which kinds of stores – out of town, local express, big retailer vs smaller boutique retailer. Does it work differently on different days of the week? How does it vary by product types, by luxury vs commodity etc? There are a wealth of ways we could cut learning about POS placement to print less (and save the planet) to put in the right stores (and generate more sales) and give the right information to consumers already faced with a dizzying array of choice (and give a better experience).
So why don’t we?
Because, fundamentally, FMCG businesses only have any real depth to 2 data points.
And that’s not enough.
To be fair, that does give some degree of insight – one would like to think that if you send lots of POS to a big store and you see lots of product sales, there’s a relationship between the two. And there probably is, but it’s not necessarily causal. It’s just that we tend to put the most POS where we get the most sales; it can be self-fulfilling. Control stores are notoriously hard to implement on a like-for like basis in stores that can stock thousands of products with the associated logistics and stock management that comes with that.
There are also many other factors at play which create a lot of noise in interpreting those simple 2 data points and our own unconscious bias in where we place the marketing in the first place.
Things like, did the POS even get to the store, did it make it out of the store room, did it get stocked adequately at the time it was expected to be, did it have the stand-out the creatives envisioned it having back in the concept stages. And if it did make it to the right place at the right time with the right level of product on it, did customers appear to interact with it, did it cause them to stop and enter the awareness, consideration or, dare we say it, purchase phase of the shopper journey.
And we’re not only talking about stock holding POS here – the same applies for any POS containing messaging and branding.
This isn’t a new challenge, it’s always been hard to do this at scale, we have tended to see that the stores with the best managers get the best results and that’s probably because they apply their judgement as to what the best POS is and use it in the right way for their store. But that’s difficult to quantify, and therefore measure and apply at scale across multiple stores, in multiple locations, in multiple markets.
In terms of technology, we’re seeing chips in the delivery pallets to see that it got there (if delivery pallet was the method of delivery), ibeacons in the POS, although that needs your phone, your Bluetooth on, the app, and the proximity range can open up questions. And again the scale and investment has been off-putting. It tends to be easier to use data that already exists rather than having to generate new data points.
So what might have changed that makes it more possible today than it has been before?
Perhaps this is an opportunity to use Artificial Intelligence to learn about what works.
What makes me think that? I recently attended a conference where the Director of Public Policy was talking about her business FiveAI. (As an aside, she’s a woman, and a young one, I couldn’t miss the opportunity to big-up the successful female story whilst I have the platform, but enough of that). FiveAI are an organisation working to understand how we’d organise ourselves to manage autonomous cars – using AI to inform the policies and regulations required to deliver safe, reliable, and socially responsible automated transportation services. And they are using TFL CCTV as the input data for that learning exercise – i.e. interpreting images on a mass scale to learn how people behave whilst driving their cars and what decisions they make – then extrapolating that learning.
Retailers have security cameras, they can see what is happening in store with their merchandise in a digital format. And if Retailers don’t want to share that data or they couldn’t be repurposed, there are organisations that already crowd-source imagery of POS stands or even without any additional costs, FMCG brands could use their existing field sales force to do the job (most do already!) – those images can be collated over time together with the meta data of where they are, when they are installed, what type of store they are in, geographically or demographically. Once you connect that with product sales you can associate the actual investment of what people actually saw with income – and once you have that dataset you can ask it anything you like in terms of performance metrics.
There’s no simple answer to finding the End to End performance measures of Point of Sale on a mass scale.
Brands need to decide between picking 2 or 3 stores, undertaking a deep-dive one off research – both Qual and Quant – project, and then assuming all other stores will behave the same.
Or they can look for a means of mass measurement, and to do that, you need mass data input. Maybe there’s an opportunity for AI to have a look at imagery to do that?