1 post
• Page 1 of 1
SoftServe Business Systems
One Technology – Different Outcomes
Image Recognition (IR) in the CPG market for shelf audit is no longer a novelty – it has become a standard part of many companies’ strategic toolkit. Most consumer goods manufacturers are already piloting or implementing CPG Image Recognition solutions in multiple markets. However, their experiences vary significantly: while some companies successfully scale and transition to fully automated audits, others remain stuck in pilot mode for years or abandon the initiative after early roadblocks.
So why does the same technology of Image Recognition in the CPG market deliver such inconsistent results?
The answer lies not in the technology itself, but in how it is implemented. Let’s explore the main reasons why companies fail to unlock the full value of IR in CPG, and what can be done to change that.
Reason #1: Incomplete Neural Network Training Cycle
Image Recognition for CPG is not a “set it and forget it” tool. For the system to accurately recognize products, it must first be trained on real market photos, and that’s just the beginning.
In many cases, vendors or companies focus on initial training but don’t establish a clear retraining process. But the reality is dynamic: new brand extensions and SKUs appear, packaging changes, seasonal promos are launched, and brands enter new categories. Without timely updates, model accuracy drops, and trust in the system fades.
image recognition for cpg accuracy during the implementation
At SoftServe Business Systems, we’ve designed a process that ensures continuous model retraining based on real-world usage. With sufficient image data, retraining can be completed in just one working week, making improvements seamless for end users while maintaining high accuracy.
Read More: AI for CPG: Transforming Innovation and Growth in 2025
Reason #2: Poor Photo Quality – The Weakest Link
The system can only recognize what it sees. If photos are taken from the wrong angle, overexposed, blurry, or incomplete, the results will be poor.
This is not a people issue – it’s a process issue. In many cases, field teams lack clear photo-taking guidelines or don’t fully understand why quality matters. Without standardization, the CPG Image Recognition system won’t receive a consistent stream of usable inputs.
This is why, after successful training and pilot phases, the quality of results can drop dramatically during full rollout. A large number of untrained users flood the system with poor-quality data, undermining recognition accuracy and, ultimately, user trust. Usage rates fall, and the perceived value of the solution erodes.
To avoid this, we place a strong focus on training. Clients receive detailed learning materials and visual instructions. The mobile app also guides users in real time, helping them take proper photos and blocking low-quality submissions.
Another way to improve data quality is by implementing Augmented Reality AI Image Recognition for FMCG, which provides real-time feedback on image quality before processing.
Learn more about how AR Shelf Recognition drives business value here.
A proper implementation approach ensures stable recognition accuracy and long-term sustainability of the solution.
Reason #3: Data Exists – But Isn’t Used to Drive Action
A common mistake is using IR “after the fact.” Photos are taken, data is processed, and reports are sent a day later. But decisions should’ve been made by the sales team in-store, instantly. If they’re not, the value of IR drops sharply.
For IR to become a true business tool, it must be integrated into the actual store visit. A rep scans the shelf and immediately sees product availability, planogram compliance, missing SKUs, and prioritized action points. This level of execution is only possible with seamless integration into the SFA and near-instant data processing.
image recognition for cpg recommended steps
Watch webinar: FMCG Sales 2025: Augmented Reality & Guided Store Visits
In SSBS implementations, IR is embedded directly into the mobile visit flow. Results are updated in near real time, turning a traditional sales call into a guided, insight-driven visit, improving the overall customer experience. Meanwhile, headquarters teams get immediate insights from the field, allowing them to respond faster to retail conditions and market shifts and to more actionable analytics at HQ.
Reason #4: The Vendor Can’t Scale
The growing maturity of Image Recognition in Consumer Goods technology has lowered the barrier to entry. As a result, new startups offering shelf Image Recognition solutions emerge frequently. While creating an MVP is easy thanks to the maturity of technologies like machine learning and computer vision, scaling it to thousands of stores and millions of images, with weekly SKU updates, is a completely different challenge.
In practice, sometimes we see companies selecting vendors without properly assessing their ability to scale or the adaptability of their recognition algorithms to complex FMCG environments. Common issues include:
Lack of experience handling large data volumes or complex assortments
No offline mode – problematic in stores with unstable connectivity
No automatic model updates or retraining workflow
Or, simply – the model doesn’t work as expected
At SSBS, we fully control the entire process, from annotation and training to model support and customer service. This is the only way to ensure consistent AI product recognition quality across different markets and regions.
Reason #5: Internal Resistance to Change
One of the most underestimated obstacles is the human factor. Image Recognition in consumer packaged goods introduces objectivity through Artificial Intelligence, reducing human bias in execution data. This can be perceived as a threat, exposing gaps in execution that were previously overlooked or manually adjusted.
In some cases, field teams may distrust the system or feel that insights don’t reflect their effort. In other teams, fear of the change may affect bonuses or lead to micromanagement.
That’s why clear communication is crucial. Teams must understand that IR doesn’t punish – it helps. It eliminates unnecessary bureaucracy, provides accurate performance data, and increases team satisfaction through transparency.
Image Recognition (IR) in the CPG market for shelf audit is no longer a novelty – it has become a standard part of many companies’ strategic toolkit. Most consumer goods manufacturers are already piloting or implementing CPG Image Recognition solutions in multiple markets. However, their experiences vary significantly: while some companies successfully scale and transition to fully automated audits, others remain stuck in pilot mode for years or abandon the initiative after early roadblocks.
So why does the same technology of Image Recognition in the CPG market deliver such inconsistent results?
The answer lies not in the technology itself, but in how it is implemented. Let’s explore the main reasons why companies fail to unlock the full value of IR in CPG, and what can be done to change that.
Reason #1: Incomplete Neural Network Training Cycle
Image Recognition for CPG is not a “set it and forget it” tool. For the system to accurately recognize products, it must first be trained on real market photos, and that’s just the beginning.
In many cases, vendors or companies focus on initial training but don’t establish a clear retraining process. But the reality is dynamic: new brand extensions and SKUs appear, packaging changes, seasonal promos are launched, and brands enter new categories. Without timely updates, model accuracy drops, and trust in the system fades.
image recognition for cpg accuracy during the implementation
At SoftServe Business Systems, we’ve designed a process that ensures continuous model retraining based on real-world usage. With sufficient image data, retraining can be completed in just one working week, making improvements seamless for end users while maintaining high accuracy.
Read More: AI for CPG: Transforming Innovation and Growth in 2025
Reason #2: Poor Photo Quality – The Weakest Link
The system can only recognize what it sees. If photos are taken from the wrong angle, overexposed, blurry, or incomplete, the results will be poor.
This is not a people issue – it’s a process issue. In many cases, field teams lack clear photo-taking guidelines or don’t fully understand why quality matters. Without standardization, the CPG Image Recognition system won’t receive a consistent stream of usable inputs.
This is why, after successful training and pilot phases, the quality of results can drop dramatically during full rollout. A large number of untrained users flood the system with poor-quality data, undermining recognition accuracy and, ultimately, user trust. Usage rates fall, and the perceived value of the solution erodes.
To avoid this, we place a strong focus on training. Clients receive detailed learning materials and visual instructions. The mobile app also guides users in real time, helping them take proper photos and blocking low-quality submissions.
Another way to improve data quality is by implementing Augmented Reality AI Image Recognition for FMCG, which provides real-time feedback on image quality before processing.
Learn more about how AR Shelf Recognition drives business value here.
A proper implementation approach ensures stable recognition accuracy and long-term sustainability of the solution.
Reason #3: Data Exists – But Isn’t Used to Drive Action
A common mistake is using IR “after the fact.” Photos are taken, data is processed, and reports are sent a day later. But decisions should’ve been made by the sales team in-store, instantly. If they’re not, the value of IR drops sharply.
For IR to become a true business tool, it must be integrated into the actual store visit. A rep scans the shelf and immediately sees product availability, planogram compliance, missing SKUs, and prioritized action points. This level of execution is only possible with seamless integration into the SFA and near-instant data processing.
image recognition for cpg recommended steps
Watch webinar: FMCG Sales 2025: Augmented Reality & Guided Store Visits
In SSBS implementations, IR is embedded directly into the mobile visit flow. Results are updated in near real time, turning a traditional sales call into a guided, insight-driven visit, improving the overall customer experience. Meanwhile, headquarters teams get immediate insights from the field, allowing them to respond faster to retail conditions and market shifts and to more actionable analytics at HQ.
Reason #4: The Vendor Can’t Scale
The growing maturity of Image Recognition in Consumer Goods technology has lowered the barrier to entry. As a result, new startups offering shelf Image Recognition solutions emerge frequently. While creating an MVP is easy thanks to the maturity of technologies like machine learning and computer vision, scaling it to thousands of stores and millions of images, with weekly SKU updates, is a completely different challenge.
In practice, sometimes we see companies selecting vendors without properly assessing their ability to scale or the adaptability of their recognition algorithms to complex FMCG environments. Common issues include:
Lack of experience handling large data volumes or complex assortments
No offline mode – problematic in stores with unstable connectivity
No automatic model updates or retraining workflow
Or, simply – the model doesn’t work as expected
At SSBS, we fully control the entire process, from annotation and training to model support and customer service. This is the only way to ensure consistent AI product recognition quality across different markets and regions.
Reason #5: Internal Resistance to Change
One of the most underestimated obstacles is the human factor. Image Recognition in consumer packaged goods introduces objectivity through Artificial Intelligence, reducing human bias in execution data. This can be perceived as a threat, exposing gaps in execution that were previously overlooked or manually adjusted.
In some cases, field teams may distrust the system or feel that insights don’t reflect their effort. In other teams, fear of the change may affect bonuses or lead to micromanagement.
That’s why clear communication is crucial. Teams must understand that IR doesn’t punish – it helps. It eliminates unnecessary bureaucracy, provides accurate performance data, and increases team satisfaction through transparency.
Posts: 1
1 post
• Page 1 of 1
