TITANIUM COMPUTING
(512) 623-9199
Free consultation
CASE STUDIES · RUSH AUTOMOTIVE
How Rush Automotive Coaches Every Phone Call Without Listening to a Single Recording
Titanium Computing built CallScrub: an AI call-intelligence platform that reviews every call so management doesn't have to, with Anthropic's Claude as the analysis engine — grading, sentiment, and coaching for every conversation.
The Problem
Every service call that comes into an automotive shop is a revenue moment. Did the advisor capture the VIN and the make and model? Did they offer to book the appointment, and book it this week instead of next? Did a frustrated caller leave the call calmer than they started, or angrier?
Management at Rush Automotive knew these moments were being won and lost on the phone every day. What they didn't have was any practical way to see it. The traditional options were all bad: listen to call recordings for hours (nobody does, past the first week), spot-check a random handful (and miss the patterns), or wait for problems to surface as lost customers and bad reviews (by which point the coaching moment is months gone).
Call quality was the most important thing nobody had time to measure.
The Solution
Titanium Computing built CallScrub with Claude as the analysis engine. For each call, Claude produces a complete quality picture:
Grading against your standards
Rush's management defines the call flows that matter, and Claude scores every call against them: Did we get the VIN, make, and model? Did we attempt to upsell an appointment? Did we get them scheduled this week? The checklist is the shop's own playbook, not a generic rubric — and management can add training data and new required flows as standards evolve.
Sentiment across the call arc
Claude tracks the caller's mood at the beginning, middle, and end of every conversation — so a call that started hostile and ended booked shows up as the win it was, and a call that started friendly and ended cold gets flagged for review.
Summaries, lessons, action plans
Every call gets a concise summary, intent classification, and a constructive action plan: what the team member did well, what could improve, and specifically how. The output is coaching material, not surveillance — written to make the next call better.
Threshold alerts, not constant monitoring
Management defines the signals that warrant attention. When a threshold is crossed, they get an email, review the specific call, and can set up a one-on-one — sending the team member the call details alongside suggested call flows and corrections. Audit by exception, not by exhaustion.
The Architecture
Titanium Computing builds AI systems as engineered pipelines, not chat wrappers. CallScrub runs Claude inside a structured evaluation harness: every call is scored against the client's explicitly defined flow requirements rather than the model's general opinions, sentiment is assessed at fixed points in the call arc for comparability across calls and staff, and grading criteria are versioned as management updates them — so scores stay consistent and defensible over time. Alerting is deterministic: thresholds are business rules, not model judgment calls.
The Results
Every call reviewed and graded — up from the handful a manager could spot-check by hand.
Management's call-review time reduced to minutes of exception review instead of hours of listening.
Missed-upsell and missed-booking opportunities surfaced with the specific call attached.
Coaching one-on-ones now backed by specific call evidence instead of recollection.
CallScrub also runs in production at Exacta Bookkeeping, where the same engine grades client calls against bookkeeping-specific service standards — proof the flow-definition model adapts across industries without rebuilding the product. Exacta is a repeat Titanium client, also running our Claude-powered AmpCortex communications platform.
Why Claude
Call grading only works if the grades are trustworthy. A QA system that hallucinates a missed VIN capture, or scores identical calls differently on different days, gets ignored by week two — or worse, poisons the coaching relationship between managers and staff. Titanium selected Claude for consistent, instruction-faithful evaluation against explicit criteria, nuanced conversational understanding (a caller's mood shift matters as much as their words), and the reliability to generate feedback a manager can hand directly to an employee. When the output of an AI system becomes part of someone's performance conversation, it has to be right — and constructive.
ABOUT TITANIUM COMPUTING
Titanium Computing is an Austin-based managed services and AI development firm specializing in cybersecurity and production AI systems. We design, build, and operate Claude-powered platforms for clients across legal, automotive, communications, and business services. CallScrub is available now at callscrub.io.