Is AI Really Causing Job Loss?

It has been nearly five years since I was inadvertently introduced to the artificial intelligence space during my time at Deloitte Digital. I was initially hired as a Senior User Experience Designer with UI capabilities to help convert more than 17 applications from traditional RF handheld scan guns to the Zebra TD52 digital mobile device. While working on the Kroger account, my project manager recognized my potential and invited me to join another initiative called Store of the Future.

If you’re familiar with Amazon’s frictionless shopping experience in Manhattan, often branded as Amazon Go, you already have a sense of what we were exploring. Here is a SNL Skit that actually does a pretty good job showing you the tech while getting a good laugh...

Amazon Go SNL Video

Amazon Go stores use a combination of computer vision, sensor fusion, deep learning, and cashier-less checkout technology to allow customers to walk in, pick up items, and leave without stopping at a register. Cameras and sensors track items selected, and purchases are automatically charged to the customer’s account. This experience removes traditional checkout friction and redefines how retail operates. Whole Foods, owned by Amazon, later replicated similar “Just Walk Out” technology in select locations, which prompted us to conduct discovery research on whether Kroger could pursue a comparable model.

At the time, the technology felt futuristic and prohibitively expensive. Implementing such systems across flagship stores would have required investments in the billions, making the return on investment unrealistic. However, if we revisit that same concept today using Generative AI, Retrieval-Augmented Generation (RAG), and AI agents, the cost drops dramatically. What once required custom-built infrastructure and massive hardware investments can now be achieved with cloud-based AI systems at a fraction of the cost, often in the tens of millions rather than billions.

A Brief History of Computer Vision

Computer vision is not new. Its roots date back to the 1960s, when early researchers attempted to teach computers to interpret visual data. Over the decades, advances in machine learning, neural networks, and GPU computing transformed computer vision into a powerful tool. It has long been used in military surveillance, airport security, facial recognition systems, medical imaging, and industrial automation. What has changed is not its existence, but its accessibility, accuracy, and scale.

Today, computer vision operates far beyond passive video recording. In retail environments, it can identify behaviors historically monitored by human security guards, shoplifting patterns, concealed weapons, and suspicious movements but, at a scale humans simply cannot match. Instead of multiple guards monitoring screens, AI systems can flag incidents in real time, send images or video clips to a single guard, and allow that guard to investigate. Importantly, AI should never be treated as 100% accurate. Human oversight remains essential. Guards review footage, validate the evidence, and make final decisions. AI accelerates detection; humans retain judgment.

📉 Retail Shrink: The Cost to Retailers

Lets talk about some real numbers. Retail shrink (loss due to theft, error, fraud) is a major cost driver.

According to the National Retail Federation’s 2023 National Retail Security Survey:

• The industry average shrink rate is around 1.6% of sales

• Total U.S. retail shrink in 2022 was $112.1 billion

• Of that shrink, ~27% is due to operational errors, process failures, and mis-scans (not just shoplifting)(Source: NRF Shrink Survey 2023 ABSCO / NRF)

To fuel the argument lets work in a scenario, Lets say Macy’s reported $500M annual sales, according to the stats from above:

• Typical shrink (1.6%) = $8M loss per year

• Operational errors portion (27%) = $2.16M

• Theft portion (remainder) = $5.84M

  
  

So even without intentional theft, mistakes cost millions.

👁️‍🗨 Computer Vision + AI: How Retailers Are Addressing Shrink

Many retailers are now deploying AI-powered computer vision systems (both behind the scenes and integrated into self-checkout or loss prevention analytics) to:

• Detect suspicious activity in real time

• Flag shoplifting events automatically

• Trigger alerts to security personnel only when evidence is clear

• Reduce false positives (less need for manual monitoring)

  
  

Companies like Grabango (vision-based analytics) and Everseen (AI loss prevention) have published data showing shrink reduction after adopting computer vision technology.

📊 Shrink Impact Comparison

From Grabango’s self-checkout study:

• Self-checkout lanes without computer vision: ~3.5% shrink

• Traditional cashier lanes: ~0.21% shrink(Source: Grabango self-checkout analytics)

  
  

For $200M in self-checkout revenue:

• At 3.5% shrink = $7M lost

• At 0.21% shrink = $420,000

• Difference = ~$6.58M loss attributable to weak controls

  
  

Computer vision systems can reduce effective shrink closer to cashier lane rates or better as security improves.

🧠 Real Case: Everseen’s ROI

Everseen’s retail shrink reports show:

• Shrink from self-checkout fraud doubled from one year to the next

• But when AI systems are deployed, retailers recover significant losses

• Reported ROI payback periods are as short as 6–12 months in modeled scenarios(Source: Everseen Retail Threat Curve & Forrester TEI on AI Loss prevention)

Illustrative retailer example (100 stores)

• Each store loses ~$102k/year to self-checkout loss

• Total chain loss = $10.2M/year

• If AI reduces loss by even 40% → savings of $4.08M/year

• Deployment + maintenance often pay back within 1–2 years

  
  

🧠 Why This Matters

🔹 Shrink IS expensive, causing Millions of dollars leak annually, even before considering theft and error together.

🔹 Computer vision doesn’t just detect, it prevents and AI can spot events and patterns humans miss and focus attention where it matters.

🔹 ROI is measurable, unlike many speculative tech investments, AI loss prevention has real financial metrics:

• Shrink reduction %

• False positive reduction

• Security response efficiency

These translate directly into dollars saved.

Here’s another example with stats showing how scanning errors / checkout shrink can cost retailers millions, and why Visual AI / computer-vision checkout tools can pay back quickly.“Just honest mistakes” at checkout cost millions

“Just honest mistakes” at checkout cost millions…

The National Retail Federation reports an average shrink rate of 1.6% of sales (FY 2022), totaling $112.1B in losses. In that same report, retailers attribute 27% of shrink to process/control failures and errors (which includes operational mistakes like scanning, ringing, mis-keyed produce PLUs, etc.).

If a retailer does $1B in annual sales:

• Shrink at 1.6% = $16,000,000

• Errors/process failures at 27% of shrink = $4,320,000/year tied to mistakes (not necessarily theft)

  
  

That’s the “small errors” problem at scale, millions, annually.

Self-checkout error rates can be dramatically higher

A large computer-vision analysis by Grabango found:

• 3.5% shrink rate at self-checkout vs 0.21% at staffed cashier lanes

• 6.7% of self-checkout transactions had partial shrink vs 0.32% with cashiers

If $200M of sales flow through self-checkout:

• 3.5% shrink = $7,000,000

• 0.21% shrink = $420,000

• Difference = $6,580,000/year (potentially addressable via better controls + visual AI)

  
  

ROI: why AI can pay back in years (or faster)

A Everseen report notes cart-based loss at self-checkout and estimates that for an average grocery store with 12 self-checkout lanes, this loss exceeds $102,000 per year. Separately, Forrester Consulting’s Total Economic Impact study on Everseen’s Evercheck reports a 6-month payback, with modeled benefits over three years (composite org).

Simple chain math: 100 stores × $102k ≈ $10.2M/year leakage. Even partial reduction can justify AI + operations costs quickly especially when payback periods reported are measured in months, not years.

🧠 Human + AI: A Powerful Combo

Computer vision doesn’t replace human judgment, it amplifies it. The system flags events, and trained retail security or loss prevention professionals make final calls. Where once two guards might monitor dozens of cameras and miss events, AI narrows attention to likely threats, reducing labor hours and increasing capture accuracy. So, Yes, AI (specifically computer vision) can reduce millions in shrink every year while providing less jobs in those existing fields like security and loss prevention professionals. And the reduction payoff can be realized within years, often within months, making the ROI real and tangible.

How AI in Retail Redistributes — Not Eliminates — Labor

The dominant narrative around artificial intelligence in retail is simple: automation replaces cashiers and reduces headcount.

But the reality is far more nuanced. AI does not eliminate labor. It redistributes it.

As retailers adopt computer vision, predictive analytics, and cashier-less checkout methodologies, the labor model shifts from repetitive manual tasks to higher-skill technical and operational roles. The question is not whether jobs disappear it’s how they transform.

New Job Categories Emerging from AI Adoption in Retail

1. AI System Operations

When AI systems are deployed in stores, they require oversight, calibration, and human verification. New operational roles made include:

• Computer vision monitoring analysts

• AI exception review specialists

• Incident verification teams

• Model training auditors

AI flags anomalies. Humans validate and intervene. This “human-in-the-loop” layer is critical for accountability and system accuracy.

2. Data & Engineering Roles

Behind every AI deployment sits a growing technical backbone:

• Machine learning engineers

• Data labeling and annotation teams

• Model validation analysts

• AI performance QA specialists

• Edge computing technicians

Cashier-less systems and shrink detection platforms require continuous model improvement, retraining, and performance testing. These are ongoing roles, not one-time implementations.

3. Infrastructure Expansion

Retail AI relies heavily on physical and digital infrastructure:

• Sensor installation technicians

• Hardware maintenance engineers

• Network reliability specialists

Cameras, weight sensors, shelf trackers, and compute nodes must be installed, maintained, and upgraded. AI does not remove infrastructure labor, it increases it.

4. Retail Technology Strategy Roles

As AI becomes embedded in operations, retailers are building strategic functions around it:

• AI implementation managers

• Retail innovation leads

• Loss prevention AI analysts

• Ethics & governance compliance officers

AI governance is becoming its own discipline. Decision boundaries, compliance, and risk management require leadership and oversight.

5. UX & Customer Experience Roles

Cashier-less checkout is not purely technical, it is experiential. New experience-driven roles include:

• Checkout experience designers

• Behavioral friction analysts

• Human-in-the-loop workflow designers

  
  

Retailers must design systems that feel intuitive, transparent, and trustworthy. This requires UX leadership, particularly in high-risk domains like payment and loss prevention.

The ROI → Job Expansion Cycle

Consider the economics. If AI reduces shrink by just 0.5% in a retailer generating $10 billion in annual revenue, that represents $50 million recovered annually.

Recovered capital does not sit idle. It is reinvested into:

• New store openings

• Supply chain optimization

• Ecommerce expansion

• Product development

• Digital transformation initiatives

Growth fuels hiring. Margin protection funds innovation. Innovation creates new job categories.

A Necessary Reality Check

Yes, traditional cashier roles may decline in certain environments. But context matters. Retail staffing across many regions is already constrained. Many stores struggle to hire consistent front-end labor. AI systems frequently supplement understaffed operations rather than replace surplus employees.

Additionally, AI expansion drives growth in:

• Floor associates focused on customer engagement

• Pick-and-pack roles for omnichannel fulfillment

• Store technology operators and digital support teams

The shift is not elimination. It is reallocation.

Automation vs. Elimination: A Historical Pattern

History provides perspective. ATMs did not eliminate bank tellers, they redefined their responsibilities. Self-checkout did not remove store associates, it shifted them to oversight and customer support roles. Ecommerce did not eliminate retail, it expanded logistics, warehousing, and fulfillment careers.

AI in retail follows the same pattern. It reduces repetitive scanning. It reduces manual monitoring. It increases technical oversight.

Manual operations become technical operations.

The Core Insight

AI in retail:

• Protects margins through shrink reduction

• Improves operational efficiency

• Recovers millions in losses

• Funds innovation and expansion

• Creates higher-skilled technical and governance roles

  
  

The workforce does not disappear. It evolves.

And the retailers who understand this evolution, who invest not only in AI systems but in the human roles that support them, will be the ones who scale sustainably.

Job Loss or Job Transformation — whats the reality?

However, many people stick to fear, causing unreliable facts, leading to many people being understandably upset, believing AI is taking their jobs. But this perspective overlooks the broader reality: AI is also creating jobs. Roles are shifting, not disappearing. If a cashier role is reduced due to self-checkout systems, that same individual possesses invaluable domain knowledge, customer behavior, sentiment, edge cases, and operational friction points. These insights are critical during the AI discovery phase. By contributing to research, testing, and validation, former cashiers can transition into roles such as AI trainers, UX researchers, QA analysts, or operational consultants. Often, foundational skills, such as proficiency in Word, Excel, or PowerPoint are enough to translate real-world experience into structured research contributions.

The biggest challenge is not skill, but adaptability. Studies show that Millennials (Generation Y) were among the first to adapt rapidly to digital change. Baby Boomers are still adjusting to technologies like self-checkout and automation. Gen X has adapted, often with frustration but, continues to deliver results. Younger generations have grown up immersed in constant technological evolution.

On the second hand a college of mine did ask: “If an agent compresses time on some tasks, but there are incumbent humans doing those tasks with an added scope of judgement, that area isn’t a huge revenue loss or opportunity (it’s stably doing well) and adding agents adds considerable cost, is it ethical to replace the humans with agents?” 

My belief is if a workflow is already stable, revenue-neutral, and heavily dependent on human judgment, replacing people with agents purely for compression or marginal efficiency gains is not only unnecessary, but ethically questionable. In those cases, AI should support humans, not displace them.

Where I tend to draw the distinction is between replacement vs. redistribution of human effort. In my experience, the strongest use cases for agents are not where humans are already adding high-value judgment, but where people are forced to spend disproportionate time on repetitive, administrative, or mechanically intensive tasks around that judgment. When agents remove that friction, humans are freed to do more of the work that actually creates value, strategic thinking, oversight, empathy, and decision-making.

I also agree that deploying agents where the revenue impact is neutral or marginal doesn’t justify the cost, financially or morally. These AI companies need to be honest and ask are they really revolutionizing something or is the ROI just going to be minimal or cancel its self out with the amount of investment.

That’s where a lot of AI adoption goes wrong: optimization for novelty rather than necessity. The ethical line, to me, is crossed when organizations pursue automation simply because they can, not because it improves outcomes for users, employees, or society. Using AI in areas that enable nonlinear growth, improve safety, or expand access, while keeping humans in the loop, feels like the more responsible path forward. Ethics isn’t about resisting change; it’s about choosing where and why we apply it.

Boomers for years have been complaining of self check out lanes and stations. They still need to write out that check, “I need a live person to greet me”, “Who is going to help me bag my items?”, these frustrations are driven out of long accustom behaviors and fears of change. Ever heard of the phrase if its not broken don’t fix it? Boomers say it constantly, not aware that these AI integrations (mostly) are because of ROI. When research is drawn and conclusions support that need for change, its is not always brought into the surface to all participants. 

Conclusion

Yes, AI is reducing certain manual and entry-level roles. But it is simultaneously creating higher-skilled, better-paying, and more specialized professional jobs. The narrative should not be framed as AI versus workers, but as AI transforming work. Those willing to adapt, learn, and apply their domain knowledge in new ways will find opportunities in this evolving landscape. AI is not the end of work, it is the next chapter in how work is defined.

If you found this article interesting and would like to learn more, feel free to contact me through my website at www.andreacassar.com or connect with me on LinkedIn:

https://www.linkedin.com/in/andreacassar/