Omni Talk Retail: "Computer Vision Deployment Disasters: Your Playbook On How To Avoid Them"

Date: November 6, 2025
Hosts: Anne Mezzenga & Chris Walton
Guests: Joe Serrano (Global Managing Partner, Retail & CPG, HTech), Dan Horton (VP Engineering & Delivery, HTech)

Episode Overview

This episode dives deep into the realities of deploying computer vision in retail environments, separating the hype from practical application. Drawing on first-hand expertise, Joe Serrano and Dan Horton from HTech unpack common pitfalls, offer candid war stories, and reveal a playbook for retailers trying to integrate computer vision into their stores. The discussion ranges from technical integration and edge computing to privacy, change management, and building sustainable ROI.

Key Discussion Points & Insights

1. The Landscape: Real vs. Hype in Computer Vision

• Computer Vision is everywhere: It's the backbone of many current retail innovations, from inventory control to customer analytics ([00:18], [09:54]).
• Deployment is hard: Even Amazon has struggled; most demos oversell accuracy and underestimate store chaos ([00:18], [10:47]).

2. Meet the Experts ([02:49], [04:06])

Joe Serrano: Started as a retail tech founder, innovated at Best Buy and Macy’s, now at HTech to build holistic, harmonized store systems.
Dan Horton: 30+ years modernizing retail systems; experienced in architecture, consulting, and business ownership. Brings a panoramic view of physical store tech.

Quote:
"I've probably touched pretty much every system that's in a retail environment...built one from scratch, worked off one, or tried to repair one. It gives a really unique perspective." — Dan Horton [04:06]

3. Where Computer Vision Adds Value Today ([06:46], [09:06], [26:36], [28:51])

• Focus on in-store first: HTech predominantly works on in-store solutions, but integration with e-commerce is looming.
• Big value buckets:
  • Inventory visibility & on-shelf availability
    "You gotta have things available that your customers want. Inventory visibility is a great place to start." — Joe Serrano [26:41]
  • Shrink control (loss prevention) and operational efficiency
  • Warehouse automation and robotics for fulfillment
    "We've had some success operationally with warehouse robotics...saving one client $6 million a year." — Joe Serrano [31:18]
• Personalization and retail media: On the horizon, but tread carefully due to “creepiness factor.”

4. The Demo Trap: Gaps Between Promise and Reality

[10:47 – 13:39]

• Demos ≠ Reality:
  "Everything in a demo is going to look like it's 100% perfect...if it says it's 90 or 100% accurate, in the store it's probably going to be 60." — Dan Horton [10:47]
• Questions every retailer should ask vendors:
  • Has it run in multiple, real stores, not just a controlled lab?
  • How does it handle messy, real-world data?
  • Can it work with my current hardware, or does it need expensive rip-and-replace?
  • Can legacy or analog cameras be leveraged?
  • How easy is it to update data sets to improve accuracy?

Memorable moment:
"A lot of retailers can leverage 60-80% of their existing environments with minor augmentation." — Dan Horton [13:11]

5. Customer Impact & Trust ([14:50 – 16:31], [22:32 – 25:39])

• Don’t pilot on your customers: Inaccurate solutions must be ring-fenced until robust.
• Real-world missteps:
  • Amazon’s smart cart failed on convenience because you couldn’t take it to the car ([14:50]).
  • Privacy issues—constant monitoring drives discomfort for shoppers and employees alike.

Quote:
"There's a misplacement of the technical value we're adding to the customer satisfaction we're hoping to give… That's a loss of trust." — Dan Horton [15:08]

• Privacy is not just compliance (GDPR, CCPA), it's customer comfort:
  • Signs are not enough.
  • Overreach erodes trust; opt-in and transparency are key.

Quote:
"If you start losing customer trust… no matter how great the solution is, it's not actually going to increase revenue." — Dan Horton [25:18]

6. Operational Readiness & Playbook

[16:31 – 21:15]

• “Perceptual AI Readiness”: Are your infrastructure, networks, and data platforms ready?
• Change management crucial:
  • Cross-team buy-in is vital—align operations, store ops, IT, and leadership.
  • "You have to have the discipline to scale down on that one use case... it's really hard to do that." — Joe Serrano [19:14]
  • Don't let silos dictate tech adoption speeds.
• Most pilots stall: Only 25% of AI pilots scale beyond proof-of-concept [20:55].
• Why point solutions thrive: They bypass legacy integration headaches but can lead to fragmented, hard-to-scale systems.

7. Avoiding Deployment Disasters ([20:55 – 25:39], [32:46 – 36:44])

• Perform an honest, thorough site assessment:
  • Evaluate existing cams, networks, sensors, bandwidth, and overall infrastructure.
  • Legacy is normal in retail; work with what you have where possible.

Quote:
"Most large retailers are never greenfield... so you have to assess what's available and how old is it." — Dan Horton [33:07]

• Augmentation over replacement:

  • 60–70% of hardware can often be upgraded, not replaced [34:23].

• Build vs. buy is case by case:

  • If it's truly end-of-life, replace.
  • If viable, augment and integrate.
  • Use edge computing (processing near the device) to solve network issues and unlock more real-time capabilities.

8. Strategic Recommendations & The Path Forward

Start with needs, not tech:

• Focus on operational “imperfection-tolerant” use cases (like out-of-stock detection), which provide ROI and don’t require 100% accuracy out of the gate.

Quote:
"Those tolerate imperfections so they allow us to provide faster ROI, because real-time is not necessary day one and perfection is not necessary day one." — Dan Horton [29:37]

• Prioritize low-risk, high-reward pilots: inventory, shrink, warehouse automation.
• Don't leap into consumer-facing applications until back-end basics are solid.
• Move from point solutions to unified architectures ("single pane of glass").
• Prioritize privacy, transparency, and opt-in to maintain customer trust.

9. Timestamps for Key Segments

• Introductions/Context: [00:08 – 06:14]
• Guest backgrounds: [02:49 – 06:14]
• State of computer vision in retail: [06:46 – 09:54]
• Demo expectations vs. reality: [10:47 – 13:39]
• Customer trust & privacy: [14:50 – 16:31], [22:32 – 25:39]
• Operational/AI readiness & organizational alignment: [16:31 – 21:15]
• Common pitfalls & avoiding failure: [20:55 – 22:32], [32:46 – 36:44]
• Build vs. Buy Considerations: [32:46 – 36:44]
• Strategic advice—what to tackle first: [26:36 – 30:37]

Memorable Moments & Quotes

• "Everything in a demo is going to look 100% perfect... if it says it's 100% accurate, it's probably going to be 60 in-store." — Dan Horton [10:47]
• "Privacy is not just compliance. You lose trust, you lose revenue." — Dan Horton [25:18]
• "Perfection is not necessary day one. Real-time is not necessary day one." — Dan Horton [29:37]
• "Computer vision is the eyes of AI." — Dan Horton [34:22]

Conclusion

This candid, jargon-free episode delivers a practical playbook for retailers evaluating or struggling with computer vision deployments. The real-world advice: start with operational problems, cautiously pilot with current infrastructure, don’t over-promise accuracy, avoid customer-facing tech until confidence is high, invest heavily in privacy and transparency, and focus on cohesive, cross-team strategies to build trust and sustainable ROI. If you’re wrestling with the “demo gap,” hardware headaches, or customer trust issues in your computer vision journey, this episode is an invaluable resource.

Further Learning

• Connect with Dan Horton and Joe Serrano:

  • Visit htec.com
  • Joe: joe.serrano@htecgroup.com
  • Dan: dan.horton@htecgroup.com

• Upcoming Webinar:

  • “Smart Vision in Action: Turning Retail Cameras into Profit Engines” — deeper dive into making existing camera investments pay off (January 2026).