Sales Automation
Impel's product and engineering teams were shipping new automotive AI capabilities faster than 150-plus customer-facing reps could learn them, while monthly business reviews and strategic projects still consumed hours of manual prep.
Reported impact
Impel cut new-product ramp time from three to six months to about one month, reduced dealer business-review prep by roughly 85 percent, increased support-case throughput by 20 percent, and built its flagship roleplay agent in about three weeks while scaling Claude to more than 400 internal users.
Impel uses Claude as the reasoning layer behind a portfolio of internal agents selected by business impact. For the highest-priority training problem, Claude analyzed more than 150,000 Gong calls plus Salesforce data to create a dealer roleplay agent that simulates hard conversations, speaks through speech-to-text and text-to-speech, and scores objection handling, rapport, product knowledge, and next steps before DMing scorecards in Slack. The company routes reasoning-heavy design work to Opus 4.5 and Fable 5.1 across MCP-connected workflows, then tests lighter Sonnet models as executor agents once a flow is proven. Beyond training, Snowflake-backed briefs reduce monthly business-review prep, and 400-plus Claude Enterprise seats extend the same agentic patterns into support, migrations, and knowledge work. This lets Impel treat agent building as an investment portfolio rather than a one-off experiment.
Recorded Gong calls, Salesforce records, and Snowflake business data feed Claude-powered agent workflows; Fable or Opus reason over the hard parts, Sonnet executes stabilized tasks, and outputs such as roleplay scorecards or review briefs are delivered back to teams in Slack and other internal workflows.
Cut new-product ramp time from 3 to 6 months to an average of 1 month of roleplay and certification
Once a flow is proven, we test lighter models like Sonnet and routinely find they deliver sufficient results as executor agents.
Reduced monthly business review prep from 3 to 4 hours to about 30 minutes with Claude-drafted briefs from Snowflake
We use frontier models like Opus 5 and Fable 5.1 where stronger reasoning matters most
digging through more than 150 thousand Gong calls and our entire Salesforce instance
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Google Cloud Sample Jun 4, 2026 Renault Group Renault needed a consistent, reliable, and personalized way to guide online vehicle shoppers from discovery to servicing, especially for electric and hybrid questions, without forcing visitors to manually search across… 60% of the questions relate to vehicles and their configuration first. Financing is the third topic of interest. Gemini 2.5 Flash · Renault brand configurator · Renault brand website 
Microsoft Sample Jan 28, 2025 Tekion To streamline vehicle purchasing for consumers and dealerships, Tekion aimed to create a unified, cloud-native platform that uses generative AI. One Tekion customer reports up to $5,000 savings a month. Azure · Azure Cosmos DB · Azure Kubernetes Service Microsoft Sample Nov 11, 2024 JATO Dynamics To further support customers, JATO Dynamics wanted to create a reporting tool to offer car dealerships insights into their own performance compared to the market, including metrics like sales price, discount rates, and… By automating content generation, Crow says, dealerships are saving approximately 32 hours each month, allowing… Azure AI Search · Azure App Service · Azure Data Factory 
