Image recognition: Does it actually work for South African retail?

In-store trade compliance, category management and the measurement of promotional execution have come a long way in the past five years. Whether FMCG brands use the services of merchandising agencies or deploy their own sales force automation technology, the advancements in this arena have certainly created a more efficient sales force.

The evolution of this new technology invites in some disrupters. Phrases like ‘artificial intelligence’ and ‘machine learning’ are not immune to the FMCG and retail world. With the advent of the new decade, brands are asking whether image recognition technology works in the South African retail landscape? Does it assist brands in measuring in-store marketing efforts? Is it worth pursuing?

Yes, because we tested it out and it works!

The ConnectGroup has spent the last few months trialling QuickScan image recognition technology. This artificial intelligence (AI) is integrated into its locally developed sales force automation platform called ConnectFMCG. Working closely on the pilot with the world’s largest consumer goods brand, QuickScan was trialled in 100 stores across a single category with great success.

How does it work?

Setup: The ConnectGroup loads all planograms by channel and store size against an already existing route list on the app. This means that once the sales rep has checked into their stores through mobile location services, the correct planogram is automatically displayed. The artificial intelligence (AI) is then matched to the geo-location of the store, rep details, time checked in and the other data like promotional feedback, strike rates and surveys. Once the photographs of the shelf have been uploaded, the AI analyses the image calculates the KPIs and produces a report to the rep’s device within two minutes.
Results: QuickScan assisted one of the biggest consumer goods companies in the world to improve merchandising visibility by 30%.

How did it happen?

It was critical for the ConnectGroup to work with a brand that is internationally recognised, spread across multiple categories and is available, prevalent and dominant in both high-end and low-end stores. The ConnectGroup sought to partner with a brand that embraces technology and is not afraid of taking risks with big upsides. The goal was to measure four main KPIs through a series of stitched photographs; category flow compliance, ranging and assortment, forward share and out of stocks.

Benefits

1. Measure multiple KPIs with one photograph per gantry/drop
2. Get data on a SKU level and category level
3. Two minute turnaround time for results
4. Wholistic data across segments, channels and regions
5. Ability to measure a large footprint nationally
6. ROI on time and coverage turnaround is measurable
7. Ability to match image recognition data with sales out data from retailers

Challenges and solutions

Much was learnt during this trial and the ConnectGroup iterated quickly to improve the technology for the test users on the ground. Image recognition is setup across different channels, segments and store sizes. Due to the sometimes chaotic nature of trade, not all SKUs were placed in the ideal orientation for artificial intelligence (AI) to detect easily. To combat this, the ConnectGroup implemented ‘image angle detection’ in real-time – which prompted the rep to retake a photograph if the angle was not right. This improved successful input by 60%.

Sometimes stores are dimly lit and this can affect the quality of the photograph. It is important to note here that having the right hardware is critical – a good camera will make a good photograph even in low-light stores.

Things to keep in mind:

1. Training your teams is crucial. There are do’s and don’ts when it comes to taking acceptable photographs to feed into the system
2. The store variables are extensive when it comes to planograms and number of gantries. This means to set up can be quite a tedious process. However once done, it is easier to maintain
3. The technology is new. And therefore, it is relatively costly compared to existing sales force automation technology which many brands compare it to
4. The technology is constantly improving and will get cheaper over time
5. The potential of this technology will expand in the next 12–18 months and it is worth getting started sooner rather than later

Would you like to see how QuickScan image recognition can help your brands perform better on the shelf? Click here to schedule a live demo or contact jane@weareconnect.co.za | 0722527718.