A Definitive Guide on How Image Recognition Can Transform In-Store Retail Execution
We present an ‘all you need to know’ guide on how in-store digitalization and image recognition can optimize your store execution and help boost same-store sales!
This story has been originally published on Infilect blog.
What is image recognition?
Image recognition is the ability of a system, combining IoT and AI, to accurately process and interpret content from any visual media, say an image or a video. This is made possible through technologies such as Computer Vision and trained Deep Learning algorithms to decode every image down to its pixels, and identify all objects, people, places and convert them into intelligent and comprehensible data. The images or videos can be captured through a digital camera or mobile camera at any location.
How does Image Recognition function?
Studies have found that the human brain can comprehend images in less than 13 milliseconds! Our brains have been trained subconsciously to develop this rapid processing capability and interpret the real world around us effortlessly.
Contrary to humans, machines are yet to possess such an extraordinary processing speed. A computer system still views digital images as a series of numerical arrays and interprets them based on patterns using AI algorithms to decode images into their pixels and comprehend their content.
For simplicity, Image recognition can be broken down into three main systems; “Digital sight” through camera vision, image recognition algorithms, and intelligent yet comprehensible insights.
For example, algorithms can be trained to identify certain objects in an image say, a car. But what if you want your algorithm to not only identify a car but, the make, model, and type?
Comparison requires the algorithm to be trained on various automobile brands, the variants, the type, and other crucial attributes required for classifying them.
Now, imagine this scenario in a retail store with hundreds of items located on the store shelf., Here an algorithm has to not only identify a specific product or a Stock-Keeping-Unit (SKU) but also identify their variant, type, price, and more. This requires extensive algorithm training to identify not only the SKU assortments of your brand but assortments in other product categories belonging to other competing brands!
Image recognition in retail supply chain and store execution
Retail consumers today care less about the variety in assortments and more about product availability. In a study, almost 32% of consumers frequently encounter out-of-stock scenarios at stores. In such cases, consumers typically do four things:
- They pick the same product of a competitor brand over yours
- They substitute another product in the same brand
- They purchase the same from somewhere else — say online
- Or wait for a few days until they are back in stock
The result is always the same — Lost sales and dissatisfied consumers.
For retail sales and category leaders, stockouts can be a nightmare. They can manifest out of insufficient ordering due to inaccurate demand forecasting. With a constant war being waged for maximum shelf space per category, brands find it challenging to understand how their products are stocked, displayed, and positioned to their consumers when they shop at these stores. Retail leaders want to understand how easily do consumers discover their products when they shop at stores and how often do their products go out of stock. Lack of in-store visibility can lead to a significant delay in fixing execution errors thus resulting in lost sales and poor RoI on any promotions being executed.
According to Retail Drive, Retail brands see almost 70% deviation in the strategy that was planned Vs what is being executed in stores.
When you sit out of an office, it is impossible to gain visibility into every store and every shelf. So, without the liberty to see things in action, how do you accurately monitor the performance of your display and promotional strategies at any given moment, when they hit the road?
Manual audit processes — Outdated?
Retail manufacturers solve the problem of in-store visibility by relying on market research firms to conduct extensive store audits by deploying a large network of merchandisers or field agents to visit a set of stores and manually collect on-shelf data.
Why this will fail going forward?
- Slow and subjective: A merchandiser or field agent spends a minimum of 40 mins inside a store manually counting the SKUs on the shelf and logging them in.
- Covers only a fraction of stores: You gather data from only a fraction of stores only to end up with incomplete and inconsistent data. Manual audits offer a limited scale and will get more expensive as you expand your operations.
- Prone to human error: Manual gathering of data is a laborious process for agents visiting each store. Having to manually count SKUs and their variants belonging to different categories is tedious and time-consuming. With fatigue setting in, merchandisers often log in incorrect data with an error margin of over 20%.
Manual store audits partially resolve the in-store visibility problem, no doubt. But, considering the increase in product categories and sudden unforeseen peaks in demand, retail manufacturers will have to adopt a tech stack that transforms their supply chain to be more agile and efficient to:
- Rapidly respond to demand changes and optimally fulfill stores to avoid stock-outs and improve On-Shelf-Availability
- Optimize store distribution to better manage inventory wastages
Manual auditing offers very limited visibility which will become time-consuming and expensive and will not suit your business demands going forward. The time to switch is now!
AI-powered Image Recognition for optimizing in-store execution!
In retrospect, manual audits became the MO considering the dearth of tools and technology platforms available to deliver accurate in-store data. But, AI technology and IoT, have come a long way and offer compelling and credible value to retail manufacturers.
In-store automation combined with image recognition helps retail stakeholders to gain real-time visibility into their stores by capturing just a few images iof retail shelves! Retail stakeholders can now track every SKU on a shelf across thousands of stores in any geo-location and time zones, at any given moment.
In-store automation can be achieved with two methods:
- By enabling your merchandizers to capture in-store image using their mobile devices, or
- By installing low form-factor-based IoT-based cameras on the retail shelves to capture images of the store shelves frequently throughout the day.
Retail analytics delivered to you straight from thousands of stores
- Shelf intelligence: Track on-shelf presence, position, pricing, and planogram. Gather accurate retail analytics such as On-Shelf-Availability and Share of Shelf to significantly improve SKU visibility inside stores.
- Compliance monitoring: Monitor if stores display appropriate pricing of your products on the shelves. Work with the category manager and store managers to fix violations.
- Competitive Intelligence: Get competitive insights on how your display, promotional, and pricing strategies stack up against competitors in-store strategies
- Trade marketing performance: Monitor where and how your in-store promotional and Point of Sale materials are positioned and displayed to the consumers.
- Tracking Customer interaction: Monitor millions of consumer touchpoints at the store shelves to gain insights into their preferences and behaviors.
There are important factors you need to consider before you invest in image recognition and understand how they can deliver high value to your business. Some important considerations are as follows:
- Accuracy: Accommodation of any type of SKU in any retail scenario, be it Modern or General trade stores belonging to any sector ranging from consumer goods to the grocery to fashion.
- Instant availability of store analytics: Time taken by the AI to process in-store images and transform them into business insights should be within a few seconds or minutes.
- Instant optimization plan: Once your merchandizers capture images at stores, instant retail analytics and action plan must be made available to them which can help fix execution errors before they even exit the store.
- Correlation with sales data: Correlating execution insights with your historical and projected sales data must be available to help you track the direct impact on same-store sales.
- Sustained confidence in the data accuracy: AI may not be the new shiny object anymore, but ensuring the platform always delivers action-ready insights of the highest accuracy is paramount.
- Minimal change management with a minimal learning curve: The platform should easily integrate into your operational and business process with no new learning.
Before and After
Constraints your image recognition must account for in both Modern and General trade
Brick and Mortar stores do not always offer an ideal scenario to capture in-store images. This is especially relevant in General trade stores or small-scale high-frequency mom and pop or Kirana shops.
- Image quality: Merchnadizers or your field agents who visit a large number of stores may encounter stores that have very little room to move around to capture images of the shelves and poor lighting conditions can impact the photo quality.
- Image capturing: Merchandizers should find in-store photo capturing easy and almost a no-brainer activity, while still being able to capture good quality photos that enable the AI to produce the highest accuracy in data. This is important as the AI must generally account for various photo constraints such as slant, blur, and poor lighting conditions.
- Offline capability: Image recognition platforms are mostly Cloud-based which involves uploading images onto the AI. Far-flung stores can have erratic and spotty mobile networks which makes it important for the system to work with nimble or no network connectivity.
Image Recognition in different retail sectors
Image recognition platforms can be an indispensable tool for any retail sector ranging from consumer goods to fashion to grocery to consumer electronics. For example, FMCG and CPG manufacturers are constantly challenged to track in real-time hundreds of SKU assortments per category per segment across thousands of stores which is not so simple.
With rising competition and an increase in different product categories to address niche consumer demands, CPG manufacturers are actively adopting AI, ML, and IoT technologies inside stores. This enables them to better understand their store’s performance, understand consumer pulse and take necessary steps to improve product sales, and incentivize stores to ensure better RoI on in-store promotions.
Store Analytics? Done! But what should you expect going forward?
It is safe to say that, Image recognition can overcome several roadblocks posed by manual-store audit processes. While image recognition can deliver instant store analytics that enables you to not only gain complete visibility but add tangible value to your business. Going forward image recognition system should adopt with your growing business needs and offer a more pro-active recommendations and uncover the unseen. Some of them are as below:
- Pro-active sales recommendation: Insights that go beyond telling you where and how your products are placed on the shelf, to which category mix will have a higher sell-rate in which store and in which location.
- Precision product distribution: Proactive recommendations to distribute the right product assortments to the right stores in the right quantity.
For a demo on how image recognition can help you optimize your retail execution, visit our INFIVIZ product page.