The Hidden Story in Every Service Visit
It’s a busy Tuesday morning at a dealership. A customer pulls in for what should be a simple oil change. The technician performs the inspection, then notices something more urgent - the brake pads are worn down to just 3mm. He types the note into the system and moves on to the next job.
But here’s the catch: that note, a vital piece of information, never reaches the customer in a meaningful way. The driver leaves without a second thought, unaware of the risk. Weeks later, when the brakes fail, frustration follows. Instead of building trust, the dealership has missed an opportunity - and potentially lost a customer.
This is not an isolated incident. Every vehicle service generates a treasure trove of observations: uneven tire wear, a weak battery, worn belts. Yet too often, these insights sit buried in databases, unused. The result? Lost revenue, compromised safety, and weakened customer loyalty.
Why Important Insights Get Lost
Service advisors juggle dozens of customers each day. Technician notes vary in tone, style, and detail. One technician might write “pads look thin,” another “needs new brakes.” With limited time, important details slip through the cracks.
The cost of these missed opportunities is high:
- Lost revenue: Customers leave without addressing problems that could have been solved on the spot.
- Customer dissatisfaction: When issues surface later, blame often falls on the dealership.
- Safety risks: Overlooked problems put drivers at risk and erode trust.
The problem isn’t the lack of data - it’s the inability to surface and act on it at the right time.
A Data-Driven Solution: Turning Notes Into Action
What if those scattered technician notes could be transformed into clear, actionable customer communication? That’s where data and AI make the difference.
Using Databricks, service centers can unlock the value hidden in unstructured notes and turn them into upsell opportunities. The process works in four stages:
- Data Extraction
Service notes are collected from dealership management systems and funneled into a secure, centralized analytics environment.
- Understanding with LLMs
Large Language Models (LLMs) read technician notes as a human would, capturing nuance and context. They recognize the difference between “brakes replaced” and “brakes need replacement,” normalize writing styles across technicians, and ensure no insight is overlooked.
- Opportunity Mapping
Identified issues are mapped to upsell categories such as brakes, tires, or batteries. Severity scoring helps prioritize urgent recommendations.
- Personalized Outreach
Instead of sending generic “Your car may need service” messages, customers receive personalized communication:
- “During your last visit on March 5, our technicians noted your rear tires were near minimum tread. Driving with worn tires increases stopping distance in rain - we recommend scheduling a replacement this week.”
- Tone, urgency, and channel are adapted to the audience. A young driver receives a friendly reminder; a fleet manager, an efficiency-focused update.
Under the Hood: The Technical Solution
Behind the scenes, prompt engineering and Databricks’ ai_query function enable this transformation. Prompts are designed to extract the right signals, and foundation models like Llama 4 deliver structured outputs ready for downstream analytics.
From there, outreach templates generate customer-ready messages across channels - email, SMS, push notifications - tailored with the right tone, urgency, and vehicle details. What once was a scattered note becomes a polished, actionable recommendation at scale.
Complete code is available at repo.
Step 1: Data Collection
- Source: Service notes stored in repair orders, CRMs, or dealership management systems.
- Transformation: Extracted via APIs or batch processing and fed into a secure data pipeline.
Step 2: Analytics Layer with LLMs
This is where the unstructured service notes are transformed into structured insights using LLMs and prompt design.
- Prompt Writing: We craft prompts that guide the model to extract the right signals from technician notes.
- Example prompt:
“Read the following technician note and identify any potential service recommendations for brake pads). Brake pads < 3mm → “High” urgency. Classify if required replacement is urgent, neat future or no replacement required. Return the result in JSON format with fields: {brake_replacement_recommendation, ‘brake_replacement_reasoning’}.”
- Databricks ai_query: We use the ai_query function to call foundation models directly within SQL or Python workflows in Databricks.
SELECT id,
ai_query('databricks-llama-4-maverick',
'Extract service recommendations from this note:
{{note}}', note) AS llm_output
FROM service_notes;
- Foundation Models: Models like Llama 4 (available through Databricks Model Serving) can be selected based on accuracy, cost, and latency requirements.
This approach ensures that LLMs not only parse text, but also return structured outputs ready for downstream analytics.
While ai_query() gives you the flexibility to invoke any model (foundation, fine-tuned, or custom) via a SQL-friendly interface, Databricks provides many other AI task-specific functions, which provide ready-to-use wrappers targeted at common use cases such as sentiment analysis, summarization, translation and document parsing.
Step 3: Personalization & Delivery
Finally, the structured insights power personalized outreach:
- We use prompt templates to generate different communication styles (email, SMS, push notifications) from the structured LLM outputs.
from databricks import ai_query
prompt = f"""
Write a friendly customer email based on this service issue:
Customer: {customer_name}
Vehicle: {vehicle_detail}
brake_replacement_recommendation: {recommendation}
brake_replacement_reasoning: {reasoning}
"""
result = ai_query("databricks-llama-4-maverick", prompt)
- This ensures tone, urgency, and vehicle-specific details are woven seamlessly into the customer message.
Business Value for Service Centers
Pilots of this approach are already showing strong results:
- 15–20% increase in upsell conversions through timely, relevant recommendations. Customers are more likely to say “yes” when recommendations are timely and personalized.
- Stronger customer trust and loyalty built on proactive communication.
- Greater efficiency for service advisors, who spend less time parsing notes and more time engaging customers.
This is more than a sales tactic. It’s a win-win: customers stay safer on the road, while dealerships grow their service revenue.
What’s Next: Beyond the Email
The potential doesn’t stop here. With the same framework, dealerships can move toward:
- Predictive maintenance that forecasts issues based on mileage and driving patterns.
- Omni-channel engagement, from SMS to in-car infotainment reminders.
- Dynamic offers, combining upsells with discounts or rewards.
- Real-time AI prompts, helping advisors recommend services during live conversations.
The more data is unlocked, the more opportunities open up - for revenue, safety, and customer care.
Conclusion
Every technician note tells a story. With the right tools, those stories can be translated into timely conversations that keep drivers safe and dealerships thriving. What was once overlooked text in a database can become the fuel for stronger relationships and measurable revenue growth.
