A lot less than you think.
We’re not just talking about money directly, there are many ways that Synthetic Image Recognition (IR) can bring down the cost of your operations. We’ll explain what they are in this blog post.
The cost of Traditional Image Recognition technology
Before we dive into the costs of synthetic image recognition, we need to establish the holistic costs of its inferior predecessor, traditional image recognition.
It costs too much time
One example of how time intensive traditional IR is that there is a dominant IR supplier based in Asia. This supplier employs thousands of staff to manually annotate images and validate predictions.
If you were to use this method you’ll be repeatedly held back from scaling your operations, simply because the process takes too long. With traditional image recognition the time it’d take to scale would become exponentially longer.
It costs too much in staff
As you saw in the previous example, it takes a lot of people to operate traditional image recognition technology, and all those people need to be paid. Manual annotation is a laborious , repetitive task, which pretty much every human, at minimum, doesn't enjoy. This results in a human-error rate of >5% in a human curated/annotated dataset.
Traditional IR is wholly dependent on this fragile process. The cost is also passed onto the end customer.
Having a human in the loop is a drain on both finances and time.
Poor data accuracy is expensive
As there is so much human intervention, the higher the number of SKUs to process, the higher the likelihood that your catalogue will have errors in it. This problem is exacerbated when annotating products that have little visual differences, bar a few minor packaging adjustments.
When building and managing a catalogue, data accuracy is everything. If you can’t rely on your data, you can’t operate with confidence, double checking and triple checking takes time and creates anxiety.
Scaling effectively is nigh on impossible
If you’re held back by spiralling financial costs, time costs, staff costs and data inaccuracies, how do you scale? Well, you can’t. Using traditional IR methods, by the time you scale to your desired goals, your competitors will have left you behind. The entire industry will have shifted and your new level of operations will be irrelevant.
On top of all these issues, to get to this level you would’ve spent a small fortune in time, staff and money - so there isn’t any ROI. This makes it impossible to remain competitive in the long term and can lead to total business failure.
The factors affecting the cost of traditional Image Recognition Technology
- Data Collection - The efficacy of how the data is collected is crucial to making a business case for IR. If data can’t be collected quickly and cost effectively your Shelf Auditing process will fail as you scale. On the flip side, efficient data collection can transform your process, by enabling you to scale your operations at your desired speed with no significant increase in effort.
- Data hygiene - Having to manually annotate your data reduces the likelihood you’ll be able to trust it. The quality of the data is vital to the level of performance you can achieve. If you can’t trust your data, there isn’t much point in collecting it in the first place.
- Model training - Your model is only as good as its training. Being able to use synthetic data to do so enhances this process.
- Model deployment - This involves keeping the model live, think of the hardware that makes the software work.
- Model maintenance - This consists of numbers 1-3. Additionally it includes model deployment, which ensures that the model deployed is live all the time, or at the desired time.
How traditional IR technology affects Retail Execution
Planogram Compliance
These will have to be manually checked by staff in order to achieve compliance. This can lead to the effectiveness of the displays declining as items aren’t arranged properly. Manual processes lack reliable accuracy, especially at scale, which contributes to the data quality issue we mentioned earlier in the blog post.
Stock Levels
The low level of accuracy when annotating high numbers of SKUs means that stock data is likely to become inaccurate, which negatively affects the ability to respond to any changes in consumer demands.
Pricing and Promotions
Similar to stock levels, if the data hygiene is poor the chances of price discrepancies is higher. In addition to customers being misled, pricing agreements between Consumer Packaged Goods brands ( CPGs) and retailers can end up being broken. Issues like this can cause contracts to be voided or terminated.
Product Placement
Having to carry this out manually can lead to mistakes and a decrease in KPIs for CPGs, which puts pressure back on Field Marketing Agencies (FMAs).
Point of Sale Materials
Similar to product placement, traditional IR means that the accuracy of these materials is at risk. This will also have a knock on negative effect on KPIs.
Competitor Analysis
As traditional IR has an increasingly low level of accuracy as you ramp up operations, it’s unlikely that you can rely on the data to perform any type of valuable competitor analysis.
How Synthetic Image Recognition technology affects Retail Execution
Planogram Compliance
Using the target planogram synthetic image recognition can quickly and accurately compare it against store layouts, identifying any discrepancies. This ensures that planograms are adhered to and can be accurately reported on.
Stock Levels
You can accurately detect when a particular product is out of stock, or is about to be. This makes it easier to respond to changes in consumer demands.
Product Placement
Products will now be placed where they should be, as Synthetic IR can recognise products and labels, even in disorganised environments.
Point of Sale Materials
Following on from product placement, the technology can accurately identify and locate POS materials. This helps to confirm that promotional materials are correctly positioned and visually appealing.
Competitor Analysis
By analysing shelf images from different stores or locations, synthetic image recognition can detect and compare products from different brands. This allows retailers and brands to conduct competitor analysis, understand shelf positioning, and identify market trends.
Quality Checks
Synthetic IR also helps to ensure that the right quality of product is available, not just the right product. It can recognise damaged and expired items which can be replaced with untouched and fresh products.
The cost of Synthetic Image Recognition
The Technology
This is the main thing you’ll have to pay for/with. Once you have this, it’s all systems go, because as you’ll read in the next section, and probably already know by now, Synthetic IR is far superior to its older sibling.
Synthetic Image Recognition vs Traditional Image Recognition
Cost
Unlike traditional IR the cost of Synthetic IR doesn’t increase exponentially as you increase the amount of SKUs. However, this is the case with traditional IR, the more SKUs you add, the more expensive it becomes.
This means you can quickly scale your catalogue without committing an exponentially large amount of investment.
This is largely down to synthetic IR not needing the process to increase in complexity of elements such as data curation, human in the loop, training etc.
Accuracy
Unlike traditional IR, data accuracy levels don’t exponentially deteriorate as you scale. This means you can trust your data more as you grow your operations vs. traditional IR.
Speed
Synthetic Image Recognition is significantly faster than the traditional version. It’ll take weeks as you expand your catalogue with the old technology, whereas, even with over 1000 SKUs it still won’t take more than a day doing things the synthetic way.
Remember the image below anytime you doubt/or have to convince anyone of the benefits of synthetic over traditional image recognition.
We've explained how synthetic IR can make your working life, more lucrative, and efficient. Still don't believe us? Try our free demo by clicking the link below.
At Neurolabs, we are revolutionising in-store retail performance with our advanced image recognition technology, ZIA. Our cutting-edge technology enables retailers, field marketing agencies and CPG brands to optimise store execution, enhance the customer experience, and boost revenue as we build the most comprehensive 3D asset library for product recognition in the CPG industry.
