How we analyze our own CRM

Introduction

During intense periods of product development, it’s easy to lose track of customer relationships. We experienced this firsthand after emerging from a focused building phase, only to realize our CRM had become a neglected mess. When we finally reached a point where re-engagement made sense, we were faced with a critical question: Who should we speak to next?

Understanding Our Customers: The Foundation of Meaningful Engagement

Typically, the "next" person to contact is someone with whom we have a strong personal relationship or whose case is fresh in our minds—for whatever reason. While this instinctive approach can be comforting, it isn’t always the most effective when aiming to build a successful company.

As we wrapped up significant work on our data enrichment capabilities, we knew we’d had dozens of conversations with customers about related topics—but we couldn’t remember them all. Hence, relying on memory alone wasn’t sufficient to prioritize the right conversations either.

This lack of clarity was frustrating, especially when we needed to focus on partners who were most likely to take action based on their pain points. To move forward, we needed to answer several key questions:

Who have we discussed data enrichment with?
In what context was it mentioned?
Do they struggle with an insufficient database of companies, or does enrichment occur when new companies are added?
What data sources are they using? Can we access these sources?
Is data enrichment a priority for them, or just a nice-to-have?

Given enough time, we could manually sift through our CRM to find these answers—inputting data into spreadsheets and running analyses—but this approach was neither efficient nor scalable.

Instead, we decided to build an AI-based solution (and making it part of our own platform, knowing it would benefit not just us but others facing the same challenges). By integrating our CRM (Hubspot) with our software, we were able to answer these questions in one sitting.

Here is the simplified prompt we used to analyze if past interactions are related to the topic we are investigating:

Extract why {topic} is relevant for a recipient and summarize their exact problem. 
Pay specific attention to the details of the topic and the problem and avoid generalizing.
They always have specific reasons and pain points. 
Stay true to the conversation's tone.

Note: I left the variables in place to give you a better sense of what information is pivotal and which parts are there to instruct the model.

The above can be thought of as “pre-processing” the raw (unstructured) information. We then refined it further with this prompt to hone in on the major pain point of a given person:

You are an assistant that helps identify the most relevant information from past communication with a company. Your task is to extract the reasons why the topic {topic} is relevant for a company and summarize their concrete problem and why they got in touch with us.

You get a piece of conversation that most likely has the information you need. You can use the full conversation to get more context.

Last identified topic: {company["properties.research_reasons"]}

Here is the piece that most likely has the information regarding the need for {topic}: {reference_engagement}

Here is the full conversation with the company where you can find the bigger problem they want to solve and why they got in touch with us regarding {topic}:
{company_communication}

Preparing the Connection: Ensuring Relevance and Accuracy

With this wealth of information at our fingertips, the next challenge was transforming it into actionable insights. We needed a way to connect with our contacts and ensure that every communication was relevant and precise.

The process was simple yet effective: we used AI to generate draft emails based on the data we had collected. However, we didn’t stop there. Every generated email went through a personal review. This step allowed us to verify two critical points:

We had sufficient and relevant information about the contact on record.
Our model had correctly identified the key points of interest for that contact.

This review process was essential. While large language models (LLMs) are excellent at generating content, they often struggle to judge what truly matters in a given context without proper guidance. By combining AI with human oversight, we ensured our communications were on target.

Here’s a simplified version of the prompt we used to personalize our outreach:

Hi {first_name},

I hope you are doing well!

We've just published a new [case study](https://www.narratic.ai/blog/case-study-researching-37k-companies) and I was thinking about our last conversation about {major_pain_point}.

Similar to you, the project involved lead research in several steps—specifically for Amazon stores, but the principle could easily be transferred to {problem}.

Would something like this be of interest to you?

Best regards,
Arne

You might notice the template is fairly straightforward. We experimented with various levels of personalization in the past, but when you know you’re writing to the right person, it’s more about clearly communicating your point rather than finding the perfect adjectives to “make” someone reply—especially when you have an existing relationship.

Making the Connection: The Final Step

With our emails reviewed and refined, the final step was straightforward: making the connection. Armed with confidence in our data and messaging, we reached out to our contacts, ensuring that every interaction was meaningful and targeted.

In a similar spirit, we didn’t just blast out automated campaigns afterward. While our system made us more effective, we remained committed to building and maintaining strong human relationships. Automation was a tool to enhance our outreach, not replace genuine human connection.

The Promised Land: Efficiency Meets Effectiveness

By integrating our CRM with AI and maintaining a balance between automation and personal touch, we’ve significantly improved our customer engagement process. This approach has allowed us to focus on what truly matters: Building strong relationships and delivering value to our customers—without getting lost in the noise of a messy CRM.