Derek Crager has built his career solving problems others deemed impossible. After creating Amazon's highest-rated employee training program in company history, achieving a 4.92/5 learner satisfaction score, 92 NPS, and saving the company over $150 million annually, he turned his attention to a universal challenge: What happens when your best employees retire and take decades of expertise with them?
His answer is Practical AI and its flagship product, Pocket Mentor, is a voice-activated AI system that captures tribal knowledge and makes it available 24/7 through a simple phone call. There’s no app required, no complex integration needed – just expert guidance when and where workers need it most.
I sat down with Derek to discuss his journey from Amazon's training rooms to pioneering voice-first AI solutions for industrial environments.
Carl Ford: Your company is built on lessons learned at Amazon. What were the key insights that shaped Practical AI?
Derek Crager: Amazon taught us that operational excellence is a design choice. If a process isn't instrumented, standardized, and instantly accessible at the moment of work, it will fail under load. We borrowed Amazon's obsession with SOP rigor, escalation culture – think 'andon cord' mindset – and zero-friction UX, then turned it into a hands-free, eyes-free mentor you can reach by phone, anytime.
At Amazon’s scale, those principles reduce errors, accelerate ramp, and protect customer trust. Pocket Mentor applies the same discipline to every frontline business. The breakthrough came when I realized we were fighting the same battle everywhere: Critical knowledge locked in people's heads, walking out the door when they retire or leave.
CF: You’ve identified two major pain points: Gathering up-to-date materials and delivering information precisely when needed. What other challenges do you address?
Derek Crager
DC: You nailed the big two, but I'd add three more that quietly crush ROI. First, tribal knowledge bottlenecks, where the best people become a queue and everyone else waits. Second, the forgetting curve: Humans forget up to 50% of what they learn within 20 minutes. We deliver guidance in-flow, not after the fact. Third, talent risk: Tturnover, retirements, and acquisitions erase know-how. We capture it once and make it available 24/7, so the organization stops bleeding expertise.
The ripple effects are massive. I've seen entire operations stall because only one person knew how to reset a legacy VFD drive. When that expert wasn't available, six hours of production stopped. After implementing Pocket Mentor, any technician could complete that reset in under three minutes.
CF: Managing and updating knowledge seems complex. How do you ensure accuracy and freshness?
DC: We harvest conversationally; we sit with your best people and capture SOPs, decision heuristics, exceptions, and “what do you do when...” edge cases. Updates follow human-in-the-loop QA with version control. Your source of knowledge lives in a secure environment, and a human approves changes before they go live.
We can monitor conversations to suggest flow improvements, but people stay in charge of final updates. We always stage, test with a small group, then release safely. The key is making updates as simple as having a conversation with your expert – because that's exactly what it is.
CF: How do you ensure information accuracy in critical moments?
DC: Three safeguards: First, governed content: SME-approved, version-controlled flows with no rogue advice. Second, context prompts: We ask clarifying questions, so the guidance fits the exact scenario. Third, proximity to work: Because it's hands-free, eyes-free by phone, teams can follow steps as they act, which beats memory every time.
The result is 30x impact versus traditional training and far fewer "I thought I remembered..." errors. When you're troubleshooting a $250,000-per-hour production line, accuracy isn't optional.
CF: What about companies supporting multiple product generations?
DC: Users don't hunt; Pocket Mentor does the sorting. We start by asking a couple of disambiguating questions, versions, symptoms, environments, and route to the correct flow. If needed, we'll confirm with a quick follow-up, like, "Is the serial number in the X range?" The point is zero scavenger hunt and no menus while someone is trying to fix a problem in real time.
CF: Do you integrate with digital twins or IoT signals to reduce guesswork?
DC: Yes, and it’s available today for enterprise integrations. We can ingest telemetry like error codes, temperature, vibration to pre-fill context and jump straight to the right solution branch. Instead of generic troubleshooting, we say, "Given Code 47 and last service at 200 hours, here are the three most likely fixes; let me walk you through option one."
It shortens triage, reduces speculation, and speeds Mean Time To Repair. The combination of contextual AI and expert knowledge is incredibly powerful.
CF: You're committed to voice-first design. When do visual elements become necessary?
DC: We're voice-first by design because that's how we stay hands-free, eyes-free, and useful in the moment. When a visual is necessary (e.g., orientation, wiring, part identification, etc.) we reference existing posted or printed visuals or company-hosted resources already in the environment. The assistant tells the user exactly what to look for and where.
We keep the interaction voice-led without forcing new apps or logins. Voice is stable; it doesn't go through functional obsolescence like interfaces do. A phone call works the same way today as it did 20 years ago.
CF: How do you balance limited industrial vocabularies with general language understanding?
DC: Great models fail on bad processes. Our edge is how we capture, govern, and deliver knowledge not just the model. We build SME-approved, versioned flows that map real decision trees and exceptions. Then we deliver them at the precise moment of work, so employees don't have to recall training; they can ask and do.
The combination yields up to 80% faster onboarding, approximately 30% less downtime and rework for manufacturers, 53% lower early attrition, and 30x impact versus traditional training – because the right step shows up exactly when needed.
CF: Who benefits most from this approach?
DC: Any enterprise where execution quality determines margin. That includes manufacturing and industrials, field service and logistics, multi-location services, healthcare devices, and M&A/PE portfolios that must stabilize acquired businesses fast. Leaders who care about time-to-competence, safety, uptime, first-pass yield, and retention get the biggest lift.
We're seeing tremendous adoption in companies managing remote field teams, like technicians who work alone without immediate access to senior experts. Pocket Mentor becomes their "expert in their pocket," available 24/7 via phone call.
CF: What's your approach to market expansion?
DC: We sell to the enterprise directly and through partners who own mission-critical workflows. Our key channels include OEMs and systems integrators who embed Pocket Mentor into service and maintenance programs, L&D and operational excellence providers who turn training into live, in-flow guidance, and M&A advisors and PE operators who capture tribal knowledge pre-sale and de-risk post-close.
The message is simple: Capture once, access forever. It's not an app; there's no Wi-Fi or passwords to fight – just tap and say, "Talk me through it," and we will.
CF: As our conversation concludes, Derek's vision becomes clear: Practical AI isn't just about technology, it's about preserving human expertise and making it universally accessible. In a world where skilled labor shortages meet accelerating technological change, solutions like Pocket Mentor represent a bridge between what we know and what we need to learn.
DC: If you've ever wished you could multiply your best employee across every shift, or keep a founder's brain available 24/7, that's Pocket Mentor. The future of work isn't about replacing human expertise; it's about amplifying it.
Edited by
Erik Linask
