5 Misconceptions When Launching a Digital Merchandising Project
Let's talk about the automation of trade audit based on neural networks. According to the open data of the FMCG market, the introduction of technology allows sales teams to obtain, for example, such effects:
- To reduce the duration of visits by 10-17%
- To improve the quality of work in sales points
- To improve the quality of field data
- To reduce the impact of the human factor on data
- To obtain a digital monitoring panel for 2500 SKU
- and 5 macro categories
- To monitor new metrics (the presence of price tags on all SKUs, the presence of promotional tags on promotional items, the number of the shelf on which the products are displayed, and others)
These indicators once again confirm the applicability of the technology in FMCG and the positive impact of audit digitization on OSA. But there are two sides to any coin, and today we will talk about the most popular misconceptions when launching the technology.
TOP 5 Misconceptions When Launching Image Recognition
1. If you achieve accuracy - you can relax
At the start of the project, it may seem that it is enough to train a neural network once and you can reap the benefits. It doesn't work
like that. New products appear, packaging changes, and recognition accuracy decreases.
Using the example of one of the FMCG market players, the recognition accuracy reached 82% during the pilot, and at the start of the project (31 weeks) it fell to 71%. The fact is that the design of one of the product has changed and new items have been added. By the 32nd week, the neural network was trained to recognise new positions, and the accuracy increased to 81%, by the 34th week - to 94%.
Using the example of one of the FMCG market players, the recognition accuracy reached 82% during the pilot, and at the start of the project (31 weeks) it fell to 71%. The fact is that the design of one of the product has changed and new items have been added. By the 32nd week, the neural network was trained to recognise new positions, and the accuracy increased to 81%, by the 34th week - to 94%.
2. We have enough data from the questionnaires
The questionnaire is a document with data on facing, shelf share and other basic layout parameters in the GROTEM/Agent application. Previously, the merchandiser entered this information manually, after connecting Image Recognition, the questionnaire is filled in automatically. But the questionnaire does not contain the entire assortment that is in the category, but only the key SKUs and key
metrics for them. At some point, we realized that we wanted to collect as much information as possible on all SKUs that the system recognizes. As a result, a BI system was connected to the project
contour. Now the customer receives a complete data package and builds specific reports with analytics on prices, competitors, etc. for the entire nomenclature.
3. IR will solve the problem with self-assessment
Measurements show that even an experienced conscientious employee makes up to 20% mistakes when auditing shelves.
Inexperienced and unscrupulous - more. At the same time, merchandisers find themselves in a self-assessment situation: the data they provide affects their own KPIs. It is not surprising that
Inexperienced and unscrupulous - more. At the same time, merchandisers find themselves in a self-assessment situation: the data they provide affects their own KPIs. It is not surprising that
people are tempted to embellish the situation or profitably "make a mistake". The introduction of technology makes it possible to reduce the impact of self-assessment on some indicators, but not completely. In some cases, office control is required.
Basic distribution evaluation metrics:
Basic distribution evaluation metrics:
- Shelf share
- Assortment share
- On Shelf Availability
- Price control
- Planogram execution
- Compliance with promotions
4. The Internet is everywhere
Not really so. And since recognition takes place online, nuances may arise. Fortunately, there is a solution: the photos taken during the visit will be processed in any case. When the Internet is available, the system will send the images for recognition. The only disadvantage is that the employee will not be able to see the result during a visit to the sales point and correct the situation. And the speed of the Internet connection significantly affects the comfort of using the technology.
5. In a month, everyone will learn how to take pictures correctly
The accuracy of recognition critically depends on compliance with the rules of photographing. Therefore, it is very important to regularly train employees, give feedback on compliance with the rules and involve the management of the sales service.
Unfortunately, these measures will not allow solving the problem once and for all. You need to be mentally prepared that some of the data will be lost, even when employees try to comply with the rules. For example, if a lamp is hanging over transparent bottles, then the labels are flashing. Price tags are often highlighted, overlapped.
Unfortunately, these measures will not allow solving the problem once and for all. You need to be mentally prepared that some of the data will be lost, even when employees try to comply with the rules. For example, if a lamp is hanging over transparent bottles, then the labels are flashing. Price tags are often highlighted, overlapped.
Let's summarize
Digitalization of merchandising using Image Recognition is an effective practice and, as we see it, a must have for the FMCG market in the very foreseeable future. But you need to understand that after the implementation, pilot operation will follow, at the stage of which it is important and necessary to monitor the nuances and find ways to solve them.
Do you have any questions?
We are here to help you
Leave your contact information and we will contact you promptly.