# How We Built a Scalable, Intelligent Engine for Sales Insights

### The Seller's Dilemma

In today's fast-paced, data-rich environment, sellers face significant challenges:

* Tracking accounts and opportunities involves tedious, time-consuming research.
* Critical updates from fragmented sources are easily overlooked or outdated.
* Personalizing outreach is challenging without reliable, unified data.

### How We Solve It

At **Rox**, we're tackling these challenges by delivering automated, actionable, real-time insights. By integrating public and private data into a single view, we empower sellers to act on context-rich, personalized information precisely when they need it—driving better engagement and results.

## **The Problem**

### **Fragmented Information Ecosystem**

Sellers rely on multiple data sources—news articles, financial reports, CRMs, and internal systems. However, these systems rarely integrate, forcing sales teams to manually piece together information. This inefficient process wastes time and is prone to errors.

### **Lack of Timeliness and Relevance**

Sellers need fresh, specific, and directly actionable insights. Traditional systems often deliver stale or generic data, leading to missed opportunities and ineffective outreach.

### **Personalization Challenges**

Generic outreach based on outdated or incomplete data fails to resonate with prospects. Without timely and detailed insights, sales teams struggle to identify key stakeholders and craft messaging that drives meaningful engagement.

## **Our Solution: Building the Real-Time Insights Engine**

### **Integrating Diverse Data Streams**

We aggregate data from two primary sources:

* **Public data**, such as news, job postings, and financial reports.
* **Private data**, securely pulled from CRMs, internal systems, and data warehouses.

This unified approach gives sellers a comprehensive, accurate view of their accounts. For instance, by matching CRM metrics with hiring trends from job postings, we highlight accounts primed for growth or investment.

### **Real-Time Data Processing**

Real-time data processing is crucial for delivering timely insights. Our event-driven architecture efficiently manages incoming data from multiple sources simultaneously. When a new report is published or CRM data updates, our system triggers a processing pipeline to clean, de-duplicate, and prepare the data for immediate use.

### **Generative AI based Scoring**

We apply GenAI to evaluate the **specificity**, **relevance**, **actionability**, and **timeliness** of every piece of information. Unlike traditional scoring models, GenAI excels at understanding unstructured data like news articles or social media trends. It tailors its analysis to align with business goals and user profiles.

This ensures sales teams receive only the most actionable insights, reducing information overload.

### **Contextualizing Insights**

We enrich insights with:

* **Summaries**: Concise TL;DRs for quick understanding.
* **Stakeholder recommendations**: Identifying key decision-makers.
* **Personalized outreach support**: GenAI drafts tailored email templates, saving sellers time and effort while improving message relevance.

### **Multi-Channel Delivery System**

Insights are delivered where sellers need them:

* Daily or real-time notifications via **email**, **Slack**, or our **Mac app**.
* A reliable delivery pipeline ensures insights arrive promptly, formatted for easy consumption.

## **Technical Challenges We're Tackling**

### **Scalability**

One of our most significant challenges is scalability. We process large volumes of data in real-time, making it essential to manage high concurrency and distributed data processing effectively.

To address this, we've implemented:

* **Event-driven frameworks**: Incoming data updates trigger workflows that distribute tasks across containers (via AWS ECS).
* **Task queuing**: SQS ensures efficient handling of high concurrency during peak loads.
* **Error handling and retries**: Failures are gracefully managed to maintain system reliability.
* **User Preference Embedding Similarity Search**: Using Turbopuffer to filter candidate insights by calculating cosine similarity between artifact and user preference embeddings.

For example, when processing public news streams, our architecture can ingest and analyze thousands of updates per second without impacting performance.

### **Accuracy and Precision of the Scoring Model**

Another major challenge is improving the **accuracy** and **precision** of our scoring model. Since we deal with large amounts of **noisy** and **incomplete data**, filtering out irrelevant information and focusing on what matters is essential. To achieve this, we layer in LLMs alongside traditional scoring techniques.

### **Evaluating GenAI Output**

Using GenAI to automatically create account research reports, summaries of aggregated sources, and personalized email drafts presents its own set of challenges. It's crucial to ensure that the AI-generated content is contextually appropriate and relevant to the specific customer or opportunity. We consistently evaluate our AI models' output to strike the right balance between automation and personalization.

## **Why This Matters for Sellers**

Our system transforms how sellers work by:

* **Saving time**: Automating research frees sellers to focus on strategic efforts.
* **Driving better results**: Hyper-relevant insights help prioritize high-value accounts.
* **Strengthening relationships**: Personalized, timely outreach fosters trust and improves engagement.

## **Why This Matters for Engineers**

### **The Opportunity to Build**

Engineers at Rox face challenges at the forefront of data, AI, and distributed systems:

• Scaling real-time data pipelines to process millions of updates daily.

• Balancing automation and human-centered design with GenAI.

• Building robust systems for actionable insights delivery.

### **The Impact**

Every line of code contributes directly to solving real-world problems:

* Reducing research time by hours per week for sales teams.
* Increasing response rates and sales outcomes through better targeting.

If you're passionate about solving complex engineering problems and driving tangible business outcomes, this is your chance to make a difference.

## The Beginning

At Rox, we're building the future of sales intelligence. By combining cutting-edge technology with practical engineering, we're empowering sellers to work smarter and achieve better results.

If you're excited about tackling these challenges and shaping the next wave of intelligent sales tools, **join us** on this journey.