Overview
Service organizations have spent the last two years adding AI to warranty adjudication, field service and parts. The models improved. The decisions did not always follow, because every new signal added one more thing for a person to weigh. For a warranty adjudicator, more data did not mean better decisions. It meant more decision points on the same claim, and slower calls on the ones that actually mattered.
On Service Council’s inService™ Podcast, Roshan Pinto, Head of Manufacturing at Tavant, joins host John Carroll to unpack adaptive guidance: technology that shapes what it surfaces around the person doing the work, instead of serving everyone the same answer. Roshan walks through what happened when customers running at 90 to 95 percent claim automation deliberately pulled back, why claim scoring rather than more data raised that ceiling again, and what dealers actually asked for when Tavant set out to automate claim submission. The through line is that confidence, not accuracy, decides how far automation goes.
On-Demand Podcast
Speakers
Roshan Pinto
John Carroll
Key Takeaways
More AI, More Decisions
Adding AI to adjudication did not simplify the job first, it multiplied it. Hear why the volume of information is the wrong measure of progress, and what an adjudicator actually needs at the point of decision.
Why Teams Dialed Automation Back
Organizations running at 90 to 95 percent automation reduced it on purpose, because their SMEs stopped trusting what the agents were approving. Confidence sets the ceiling, not accuracy.
Scoring Beats Volume
What raised the ceiling again was a quality and risk score on every claim, plus clustering of similar claims, so adjudicators review only what genuinely needs review and manage the rest by exception.
Guidance Adapts to the Persona
Dealers did not ask for claim submission to be fully automated. They asked for policy, job codes and past claims in one conversational ask. Guidance that adapts to the person is the differentiator.
Confidence Sets the Ceiling
Trust, not model accuracy, decides how far automation goes
Score, Then Automate
Claim quality and risk scoring turns manual review into exception management
Built for the Persona
Adjudicators, technicians and dealers need different guidance from the same data
Also Covered in This Conversation
- Serialized assets and components within components, and what agentic AI changes at the subcomponent level
- Why dealer inventory visibility is a governance problem before it is an AI problem
- Modernizing legacy warranty systems in months through reverse specification, without a rip-and-replace
- How the buying question shifted from “do you have everything we need” to “can you adapt fast enough to what we need next month”
Audience
Warranty and Aftermarket Leaders
VPs of Warranty, Aftermarket Directors and Service Operations leaders accountable for claim cycle time, leakage and adjudication capacity
Field Service and Dealer Network Leaders
Leaders responsible for technician productivity, dealer experience and first visit outcomes across a distributed network
Digital and AI Executives
CIOs, CTOs and Heads of Data assessing where agentic AI belongs alongside an existing warranty, DMS and ERP landscape
FAQs
Roshan Pinto, Head of Manufacturing at Tavant, joins John Carroll on Service Council’s inService™ Podcast to discuss adaptive guidance: AI that adapts what it surfaces to the person using it. The conversation covers warranty adjudication, dealer claim submission, field service and why confidence rather than accuracy sets the ceiling on automation.
Adaptive guidance is an operating model where technology decides what guidance a person needs based on their role, their experience and the situation in front of them, then delivers it in the right context. A warranty adjudicator, a dealer claim submitter and a field technician each get different guidance from the same underlying data.
Because their subject matter experts stopped trusting what the agents were approving. Teams running at 90 to 95 percent automation scaled back on purpose once the models revealed that claims were being approved which should not have been. Confidence in the output, not the accuracy of the model alone, determined how far automation could go.
Every claim receives a quality and risk score at intake, and similar claims are clustered so the adjudicator can see how comparable cases were handled. High scoring claims pass through automatically. Lower scoring claims arrive with the risk already identified, so adjudicators review only what genuinely needs judgment and manage the rest by exception.
They did not want every claim submission automated. They asked for policy information, job codes and clustered historical claims available through a single conversational ask, rather than searching several disconnected systems before submitting.
No. Connecting models and agents into existing warranty, dealer management and ERP systems is considerably easier than it was even a year ago, and legacy systems can be modernized in months through reverse specification, without a rip-and-replace.
The conversation runs 38 minutes. It was broadcast live on LinkedIn on August 13, 2026 on the inService™ Podcast, produced by Service Council™.
The full recording is available on this page, courtesy of Service Council™. It is also available on Service Council’s own channels and wherever you listen to podcasts.