It’s very easy to see image recognition in retail and and think one of two things:
- This technology can solve all of our problems – it can do everything we need!
- It’s too good to be true – all of these features are over exaggerated and won’t last long
The truth of IR lies somewhere in the middle. It isn’t the answer to all of your prayers, but it shouldn’t be underestimated either. In reality, the secret to getting the most out of image recognition is going in with your eyes wide open; a full understanding of the technology will lead to a scalable, successful and profitable endeavour.
So, where does image recognition hit some walls?
IR is a tool that can be used in any country without a problem, but this flexibility does come at a cost.
A CPG may have the same product in every continent, but the packaging is bound to be different, the category may be switched up, the display may literally be a foreign concept compared to the first one introduced etc.
Because of this, IR isn’t a pick up and go solution. Image recognition uses AI – know as a neural network – to work its magic, but even artificial intelligence needs a minute to catch up.
In order to be accurate and more efficient than the manual alternative, it needs time to learn the different scenarios of each location. This doesn’t take long and shouldn’t hinder the progress of any retail execution strategies, but is important know note for any prospective IR adopter.
Luckily, as real-time image recognition and smarter algorithms emerge, this ‘training’ phase for the algorithm is getting shorter and shorter. Even better still, present day IR has been made to be adaptive, meaning it can learn on the job and any changes in display won’t hinder accuracy.
Let’s get this straight: image recognition provides your team with the greatest possible actionability of any retail execution solution.
That said, external variables – mainly the type of store and country it’s used in – can have a big impact how useful the provided insights are in the moment.
Some countries are pretty strict with the rules of their shelves; no touching, no moving, no changes without approval. Even if real-time IR tells your rep that there’s a serious issue with the display they’re stood in front of, fixing this may take a lot longer than expected.
The same also applies to the type of store that’s being audited. A privately owned shop may be more receptive to changes than a chain, and a hypermarket is bound to have more regulations than a corner shop.
Obviously, this doesn’t apply to every country and every store, but is worth researching if real-time actionability is your main reason for switching to image recognition.
One of the big benefits of IR is that it doesn’t require your sales rep to be an expert in the technology. AI cues tell them exactly what to do and how to do it, so human error is almost completely eliminated – emphasis on almost.
If a picture isn’t clear, image recognition isn’t able to give the most accurate insights possible. If continually fed unclear pictures, KPIs will be wrong across the board, which can paint IR in a bad light. Luckily, because it works using pictures, you can easily see if blur or bad angles are the problem, so this shouldn’t be too much of an issue in the long run.
Moving on from human error – human capacity can be another problem.
There are thousands of in-store KPIs that can be collected, and for some CPGs, recording as many as they can is a main priority. The issue is, while image recognition can calculate several KPIs at once, there is a limit to how many images it can handle – and how many images your sales rep can take.
An intense, picture-heavy audit would take as much time as a manual audit, so negates the quick and efficient selling points of IR. It may sound counter-intuitive, but in the case of image recognition, getting your rep to do ‘just enough’ is more than enough, and maximum effort is wasted energy.
Image recognition can calculate several KPIs from just one picture, but there is a limit to how many KPIs it can handle in one audit.
Expecting IR to provide you with hundreds of KPIs is a one way ticket to disappointment. Instead, knowing that it can provide granular, accurate results on the 10-15 KPIs that truly matter is what will show you that it’s the right choice for your field team.
It also pays to be sensible when considering what KPIs you want the algorithm to track. Clearly visible points such as share of shelf, assortment compliance, out of stocks etc. are usually fair game, but small or unusually placed items can cause problems; a big item such as a carton of milk is a lot easier to recognise than pen.
Unfortunately, there isn’t a work around to this, but any manual audit would face the same hurdles.
If you’d like to learn more about how your team can use image recognition to complete more efficient stores audits, read the full outlook in our white paper “SCANNING THE SHELVES – How Image Recognition Helps Consumer Brands To Boost Their Retail Execution” here.