In this post, we’ll explore how businesses can harness the power of Generative AI (GenAI) to automate insights into performance, saving time, reducing complexity, and driving smarter decision-making.
AI Is Complex, But Using It Doesn't Have to Be
Artificial Intelligence is one of the most complex technologies humanity has ever developed. Companies like OpenAI (ChatGPT), Anthropic (Claude), and Google (Gemini) have invested billions of dollars and years of effort to create the tools we have today.
The great news? You don’t need billions of dollars or a PhD to use AI. In fact, the most revolutionary aspect of today’s AI boom is how these tools make advanced technologies accessible to everyone—allowing even us mere mortals to use AI to improve workflows and drive better outcomes.
GenAI: Turning Data Into Actionable Insights
One area where GenAI is crushing is in automating insights from data—especially for marketing teams. Businesses now generate and collect data at unprecedented scales. The challenge isn’t gathering this data—it’s making sense of it quickly and effectively.
That’s where GenAI steps in. By combining data storage, processing, and analysis with the power of large language models (LLMs), organizations can bridge the gap between raw data and actionable insights.
A Common Framework for GenAI Insights
So, how does this work? Here’s a common framework we’ve used to help businesses automate performance insights.
How it Works
Define the Business Questions Before building anything, you need a clear understanding of what matters to your business. Start by identifying the questions that you need answered. Knowing what you need to answer ensures your pipeline is focused on actionable insights. For example:
How are key metrics performing over time?
What trends or anomalies should we pay attention to?
Start with Data Not surprisingly, everything you build begins with the data you have. Work backwards from the questions you need answered to identify the sources of data you need. Whether it’s website traffic, campaign metrics, or customer engagement data, you need a reliable way to store and access it. This raw data fuels the rest of the pipeline.
Preprocess the Data Raw data rarely comes ready for analysis. The next step involves preprocessing the data into an analysis-ready format. For instance, in one example, we used Databricks as the environment to write a Python script that programmatically queried the raw data and applied pre-processing. This included:
Aggregating data by date
Calculating metric totals across different breakdowns
Measuring how these metrics changed over time
The resulting outputs provided summarized data points ready to be analyzed further.
Leverage an LLM for Analysis The preprocessed data is then passed to an LLM, such as OpenAI's GPT (using their API). Here, prompt engineering plays a critical role. By structuring prompts to incorporate specific data points, the LLM can analyze the summarized data and generate key insights, such as identifying trends, detecting anomalies, or offering performance highlights.
It's worth taking a pause here. You've most likely heard of prompts or prompt engineering when it comes to LLMs or GenAI, but what does that actually look like? Here is a basic example of a prompt template. The items surrounded by curly braces are placeholders and would be replaced with the pre-processed data.
prompt_template = """
You are a marketing analytics assistant. Analyze the following campaign performance data and provide actionable insights to improve future campaigns:
Here are the trends over time, please advise on how the campaign has been trending over time and what factors are leading to this performance:
{trends_over_time_dataset}
Here is performance by marketing channel. Please inform us which channels have been performing well and which have not been:
{marketing_channel_performance}
Here is the performance of each of our ads over the last 14 days compared to the 30 days prior. Are there any ads that have been performing exceptionally well?
{advertising performance}
Your analysis will be used by the team to monitor performance and spot ways in which they can improve performance. Please make it concise and include a funny joke at the end.
Please provide your analysis in the form of an HTML email body. Please follow these brand guidelines:
{brand_guidelines}
"""Deliver Actionable Insights (Programmatically) The final step involves programmatically delivering these insights to stakeholders. In the above example, we’ve automated this process by formatting the LLM’s response into an HTML email body, but the delivery can be in whatever way works best for your team. Imagine receiving a daily or weekly performance summary directly in your inbox, complete with trend analysis and key takeaways, all without lifting a finger.
What Makes This Powerful?
This pipeline is designed to be:
Scalable: Once set up, the process runs automatically, saving hours of manual work.
Actionable: The insights generated are ready to use, tailored to your business needs.
Accessible: No data science expertise required to make sense of the outputs.
Ready to put GenAI to work for your business? Let’s talk!
Comentários