Hello. I’m Eric Westerkamp, CEO at CaliberMind. Today, we’re discussing how to use the power of generative AI to understand the buyer journey. I’ll walk you through how to understand your siloed data, extract and transform it, and use your data to understand the buyer journey through your marketing funnel.
The Data Deluge in B2B Marketing
First, let’s acknowledge the elephant in the room: data overload. In a typical B2B marketing organization, we’re swimming in data from all corners:
- CRM systems like Salesforce
- Marketing automation platforms like HubSpot and Marketo
- Customer interactions, including events, emails, and sales conversations
- Ad Sources like LinkedIn, Google, Bing, and Facebook
- Intent Data from Bombora, G2, and others
Buried inside all of this data are answers to your specific questions. Unfortunately, even if we pull it all into a data warehouse like Snowflake, BigQuery, or Redshift, the data often remains in its original, disconnected format. How do you link that data together? How do you do anything with those different silos to see different patterns and gain the insights that you’re looking for?
Understanding your buyer journey means answering questions like: What is the buying group within this company that I’m targeting? Who are these different people, and what are their roles? How are they engaged? What are the key marketing activities? What’s working and what’s not working? At what points in time do different individuals or roles get engaged, and what did they do?
Breaking Down the Jargon: Key Definitions
Before we get too deep, let’s define some terms:
- Generative AI: Advanced algorithms/models that predict what comes next in a sequence, trained on broad data sets. These models do a good job of understanding language and information and summarizing data to be helpful to a marketer.
- Data Warehouses: Centralized repositories where we store data in a structured way.
- Retrieval Augmented Generation (RAG): A method where unstructured data is placed into a vector database to extract context, which is then fed into generative AI.
- Data Augmented Generation (DAG): Similar to RAG, DAG examines direct databases to extract data that it feeds into generative AI with prompts.
- Extract, Load, and Transform (ELT): The process of extracting data from sources, loading it into a warehouse, and then transforming it into data you can use.
The Roadmap: From Siloed Data to Actionable Insights
So, how do we go from fragmented data to a cohesive, actionable buyer journey? Here’s the step-by-step guide:
1. Extract and Load Your Data
First, we need to consolidate all our data into one place. Tools like Fivetran or Google’s data transfer services connect to your various platforms—like Salesforce and Marketo—and pull the data into your warehouse on a set schedule.
Now, you’ve got a treasure trove of raw data sitting in hundreds and hundreds of tables. While it may be interesting, it doesn’t have intrinsic value yet.
2. Transform the Data
We can create data pipelines to stitch the data together using SQL, Python, Spark, or other technologies. To create a buyer journey, we need to understand all marketing touches related to accounts, contacts, and, potentially, leads within one organization across all platforms.
We may need to handle a combination of known data (this person filled out a form on our website) and unknown data (intent data showing that a company is interested in solving a problem). This transformation results in consolidated tables where data is interconnected, giving it context and meaning.
3. Build Analytical Models
With our data stitched together, we can run different models that add value to the data based on what’s essential to our organization. For example, we can use multi-touch attribution, which takes all of these touchpoints and gives them weight and value so we can understand which campaigns are seeing the most engagement. We can see which channels are driving what behavior and at what stage of the funnel.
4. Leverage Generative AI
We can create and view a buyer journey when we combine our data about our touchpoints with a prompt for a generative AI model, like Google’s Gemini or OpenAI’s ChatGPT. What questions do we want to answer?
Maybe we want to understand the difference between recently won accounts and lost ones, or the personas for different roles in the organization. Each question may require a different set of data and a different prompt for the LLM to analyze and pull it all together.
An important caveat: remember that when dealing with large language models, you must ensure that you preserve the integrity and security of the data you’re using. To do this, embed the LLM directly in your project so the data never leaves your systems. This way, you don’t have to worry about your query being used for further training.
Our step-by-step results
Throughout this video, we’ve gone from looking at data siloed across different data sets to consolidating it into a single system and stitching it together, then transforming it into a format that a large language model can interpret and understand.
After running it through generative AI with the right prompts, we have an analysis that marketing teams can use to make better decisions when targeting prospects and refining their marketing campaigns.
For the best results, pay attention to these gotchas
Generative AI is becoming one of our best tools for data analysis. But the process isn’t without its hurdles. Look out for these gotchas:
- AI sensitivity: Generative AI models are sensitive to the input format and instructions. You need to test not just the prompts that you’re building but the data that you’re feeding it. Make sure that you’re structuring the data in a way that the LLM can actually readily use and interpret.
- Context window limits: If you start to exceed that context window, you’ll start to lose data, and these requests will fail. For example, I ran this on 20 accounts, but if I tried to run it on 300 accounts, I would exceed that context window pretty quickly because of the number of marketing events. The output can vary from one call to the next.
- Costs: Processing large amounts of data with AI models can add up. If you’re going to run LLMs, you need to understand when to run them, how and how often you’re running them, and at what frequency you plan to run them.
- Rapid Changes: The AI landscape evolves quickly. If you’re building AI into your product, make sure you build it in a way that allows you to absorb and take advantage of model improvements.
Final thoughts
I hope this video was helpful and gives you at least a high-level view of how you would go from raw data across hundreds of different marketing systems, all the way to consolidated buyer journeys leveraging the power of AI and large language models.
