Turning Fragmented Customer Communications into Continuous OEM Relationship Intelligence
Overview
A leading multinational engineering and technology company managed long-term relationships with more than thirty automotive OEMs, each a strategic commercial partnership tied directly to revenue, positioning, and future programs. The intelligence needed to manage those relationships well already existed, in emails, meeting discussions, CRM records, and the knowledge held by individual account managers, but it was fragmented and difficult to act on. Relationship risks and competitive threats were typically identified only after they had escalated.
Yarnit implemented an agentic Customer Intelligence Platform that continuously analyses customer communications and turns them into relationship insight, early-warning alerts, and strategic recommendations for sales and leadership teams. A pipeline of specialist agents structures and scores every interaction; an Ask Engine lets executives and account managers query the resulting intelligence in natural language and receive citation-backed answers. The platform was deployed inside the company's own infrastructure: it runs in the customer's Linux VM, and raw email and source data remain in SharePoint within the customer's environment — only model inference calls leave it.
The result was a shift from retrospective, manual relationship review to continuous, proactive customer intelligence across the OEM portfolio.
The Challenge
Strategic relationships managed without real-time intelligence
The company's OEM relationship management relied heavily on fragmented communication and manual account knowledge, making customer intelligence reactive rather than proactive. As the OEM portfolio grew, identifying relationship risks, competitive threats, and emerging opportunities became increasingly difficult.
Customer Intelligence Was Fragmented Across Systems
Relationship insights were scattered across emails, meeting notes, CRM records, and individual account managers' knowledge. Without a centralized intelligence layer, teams lacked a consistent view of customer health across the 30+ OEM portfolio.
Relationship Risks Were Identified Too Late
Early indicators such as shifting customer priorities, declining engagement, and competitive evaluations remained buried in day-to-day conversations. Risks typically surfaced only after they became delayed programs, formal escalations, or lost business opportunities.
Relationship Health Was Inconsistent and Difficult to Measure
Health assessments varied across regions, teams, and account managers, making it difficult to compare accounts or identify portfolio-wide trends. Leadership lacked standardized metrics to monitor customer sentiment and relationship performance.
Leadership Lacked Real-Time Strategic Visibility
Without continuous analysis of customer communications, executives had no live view of relationship trends, emerging risks, or strategic opportunities. The business needed a secure, standardized intelligence platform that could transform existing communication data into actionable insights without moving sensitive customer information outside its environment.
The Solution
Turning Customer Communication Data Into Continuous Relationship Intelligence
Yarnit built the platform as an agentic system with two parts: a continuous intelligence pipeline that converts raw communications into structured, scored relationship intelligence, and an Ask Engine that makes that intelligence directly queryable by business users. Both run inside the company's infrastructure.
The continuous intelligence pipeline is a sequence of specialist agents:
The Labeling Agent ingests emails and threads from SharePoint, then labels and structures them, identifying who is communicating, the topics in play, and shifts in engagement, and produces structured, labelled threads as JSON. This converts unstructured correspondence into a consistent, machine-readable form.
The Analysis Agent takes the labelled threads and evaluates the relationship: it produces health scores with confidence levels, the top drivers behind each score, and the key words and phrases that justify it. The output is processed, evidence-backed relationship data rather than an opaque rating.
The Strategic Recommendation Agent takes that processed data and translates it into OEM-level strategic actions, the concrete steps an account team should take in response to the signals the analysis surfaced.
The Ask Engine makes the resulting intelligence directly accessible. It is itself agentic: an Orchestrator Agent coordinates each query, an Intent Agent interprets the natural-language question, and a Retrieval Agent retrieves the relevant evidence. A user can ask questions such as which accounts show churn risk, which OEMs are showing competitive-evaluation signals, or how a specific customer's relationship is trending, and receive citation-backed answers with confidence scoring, delivered alongside executive dashboards and account-manager workbenches.
Deployment and data residency were central to the design. The platform runs in the company's own Linux VM, deployed as containerised services. Raw email and source data remain in SharePoint within the company's environment; the structured intelligence the agents produce is held in a vector store and an application database inside the same VM. The only data that leaves the environment is the inference call to the language model. For a use case built entirely on sensitive OEM correspondence, this meant the company gained a continuous intelligence capability without surrendering control of the underlying communications.
The Results
Transforming OEM Relationship Management Into a Proactive Intelligence Operation
The platform changed how OEM relationships were monitored, analysed, and managed across the portfolio.
Proactive Risk Detection
The platform identifies relationship risks, competitive threats, and dissatisfaction signals early, enabling teams to intervene before issues escalate.
Unified, Real-Time Customer Intelligence
All OEM communication is consolidated into a single intelligence layer, giving sales, account managers, and leadership live visibility into relationship health and trends.
AI-Powered Strategic Decision Support
Natural-language querying, citation-backed insights, and standardized relationship scoring enable faster, data-driven decisions across the entire 30+ OEM portfolio.
More broadly, the platform operationalised years of accumulated communication history. Rather than manually reviewing emails and fragmented account updates, teams now access relationship insight informed by historical engagement patterns, stakeholder behaviour, and evolving customer priorities, continuously, and within the company's own environment.
“Our account teams held a great deal of knowledge about our OEM relationships, but it lived in inboxes and individual memory, and we usually learned about a problem once it had already escalated. The platform changed that. Customer communications are now continuously structured, scored, and explained, and our leaders can ask a direct question, which accounts are at risk, where are we seeing competitive evaluation, and get a citation-backed answer in seconds. What made it viable for us was the deployment model: it runs inside our own environment and our email data never leaves SharePoint. We moved from reviewing relationships in hindsight to managing them in real time, consistently, across the entire portfolio.”
Vice President,
Global Key Accounts
Behind the Scenes
How Yarnit Built a Real-Time Relationship Intelligence Engine
The platform is built as a multi-agent system deployed in the company's Linux VM as containerised services. Raw emails and source data remain in SharePoint within the company's environment; the Labeling, Analysis, and Strategic Recommendation agents produce structured intelligence that is held in a vector store (Chroma) and an application database (MongoDB) inside the VM. The Ask Engine (Orchestrator, Intent, and Retrieval agents) serves citation-backed answers to dashboards and account-manager workbenches. Language-model inference is provided by OpenAI; it is the only data that leaves the customer environment.
.png)
.png)
Your AI Marketing Team
