Issue 6
Welcome to TRL Cribs: AI Architecture Edition.
Today we’re touring the 3-story system powering every returns decision teams make with Appriss.
You know the penthouse (think dashboards, alerts, and risk scores).
Vishal Patel, Appriss’s Chief Product Officer, has spent a decade building the floors underneath the penthouse that most teams never see. His point? You can’t get full value working only on Layer 3 when Layers 1 and 2 are doing the heavy lifting below — especially when customers move across channels.
But first, your industry brain teaser of the week:
What percentage of shoppers are open to tech-driven return eligibility checks?
Scroll to the bottom for the answer.
What's in stock
Here’s what we have in store for you this week:
- The Rundown: What’s running underneath every risk score your team reviews
- Worth Your Time: A pasta scam, retail AI rollouts, and how employees use agents
- What We’re Up To: We’ve been on the road and the conversation keeps finding us.
The Rundown
So many teams inherit a dashboard and start working. They see risk scores, review alerts, adjust thresholds. What they’re missing is visibility into how those outputs get produced in the first place and how the system connects customer behavior across every channel..
Vishal breaks down the 3-layer architecture that powers Appriss’ AI-driven returns reduction:
Ground floor: the foundation.
Layer 1 is data intelligence – it handles identity resolution across cards, emails, phone numbers, and channels. A customer buying online and returning in-store at 3 locations isn’t showing up as 3 separate low-risk people. Layer 1 already connected them.
Middle floor: the structural support.
Layer 2 is decision intelligence – machine learning models score that unified customer data. They approve, warn, or decline in under one second with 99.99% accuracy at the point of return.
What they catch varies by vertical:
- apparel sees repeat event-purchase-and-return behavior
- sporting goods catches seasonal gear trials
- beauty surfaces gift-with-purchase exploitation
Top floor: the penthouse.
Layer 3 is workflow intelligence – for Appriss customers, this is Sidekick, where your team actually works. It queries alerts, assigns investigations, and turns manual prep nearly automatic. Your team still makes the call. Sidekick just surfaces what matters.
Working only on the penthouse while the foundation cracks means doing more manual work while fraud patterns move faster than anyone can keep up.
You don’t buy a new jacuzzi when your basement is leaking.
To understand how our stack works top-to-bottom, read here.
Worth your time
We know time is money, so we won’t waste yours
- A California man swapped $34,000 worth of Lego sets with dry pasta before returning them for refunds. (The Guardian)
- The Vitamin Shoppe and Tecovas have rolled out in-store AI tools, and employees are the heaviest users. (Retail Brew)
- Ace Hardware launched “Hey ARMA,” an AI assistant now active in more than 2,300 stores. (Chain Store Age)
What we're up to
Women in Retail was stellar. My session was a 15-minute personal brand exercise to help individuals stay grounded and lead with confidence when the ground keeps shifting (hi AI). Multiple attendees said they were taking it straight back to their teams.
Takeback Talks Session 2 continued a thread the first session surfaced—something all retailers should be concerned about: almost no one in the room had a unified view of loss across their org, and yet every team thinks they’re solving it. That’s exactly the tension we kept pulling on.
We were also on stage at NRF Protect on June 9 with Tim Murfin from Winn-Dixie, on turning AP data into a strategic asset—cross-functional buy-in, capital allocation, better decisions under pressure. Grocery is the ultimate proving ground for this. Razor-thin margins don’t forgive a siloed approach.
Want to know where we’ll be next? See our upcoming events.
