In our previous blog posts on the opportunities in Generative AI (GenAI) and AI challenges, Justin DeBrabant, Senior Vice President of Product at ActionIQ walked us through opportunities and challenges with AI and Generative AI in the context of customer data platforms (CDPs) and customer experience (CX). Now, we’ll explore generative AI use cases, with one specific example.
Opening Doors for Generative AI Use Cases
Progress in generative AI is opening the door for a wide range of genAI use cases. Among them, we see a new revolution in the interface between business users and the customer data.
To gain a better understanding of this, let’s walk through the history of getting access to audiences within enterprise organizations, and explore the anticipated changes we can expect to see in the future.
The Infamous (Jira) Ticket
The ticketing approach was in play prior to the emergence of marketing technology and continues to hold a strong position in many organizations.
In this scenario, marketers do not have direct access to the customer data. To obtain access to an audience list, they must submit a request to either the IT department or a commonly present team in enterprise organizations known as marketing operations.
Mastering SQL is the only way to access data. The process unfolds as follows:
- Marketing wants a new audience
- Marketing creates a ticket with the audience definition: I would like an audience list with all the customers who made a purchase last year but did not make a purchase this year
- IT/Marketing Operations picks up the ticket from the queue, interprets the request, maps it against the customer data, writes down the SQL required, extracts the result of the query and sends it back to the marketing department as a list
The ticket interface may be convenient for marketers, but the turnaround time required to access data is a significant burden that hinders an organization’s ability to innovate or respond to the agility requirements of today’s world.
Additionally, there is a “trust” risk associated with relying on the results of such requests. After all, who guarantees the definition of a “customer” when asking for “all customers who…”? Is a “customer” someone who made a purchase at any time in the past? Someone who made a purchase in the past six months? Someone who didn’t return a product or canceled their membership? Definitions may not be consistent between organizations or use cases, which introduces the risk of “SQL writers” making incorrect assumptions.
The No-Code Interface (DIY)
The next wave of marketing technology aims to eliminate the bottleneck from other teams or, in cases where the marketing operations teams are in place: reduce effort, accelerate turnaround time and enhance innovation opportunities.
The user interface is no longer requiring SQL expertise. Instead:
- IT/Marketing Operations grants governed data access to business users such as Marketing
- Marketing uses a no-code interface to create audiences themselves, gain insights and activate audiences on any channel.
This approach has undeniably yielded positive results for organizations that have adopted it. However, I’ve observed friction in places moving from a ticket-based approach to the “do it yourself” model. Not every marketing team has embraced the change involving an increased effort on their side.
The translation challenge — e.g. definition terms such as “customer” — has shifted, and the person in charge of creating an audience needs to understand how to map his/her definition into the data structure set up by IT/data teams.
The question which remained opened: can we access the benefits of both previous approaches without encountering their drawbacks?
AI Use Cases: The Next Gen(erative) Audience Platform
Ask in your language, get the audience immediately!
The progress made in generative AI is opening the door for a new approach to audience segmentation within organizations:
- You can now request audiences in the same way you would submit a ticket to your IT/Marketing Ops team.
- The platform translates the request into SQL, which can be executed to retrieve the desired audience.
Imagine being able to type in your own words the audience you want to access, instead of having to drag-and-drop elements. For example:
Customers who live in NY, have savings accounts opened for more than 2 years and accept to receive marketing emails.
Not only will business teams be relieved from understanding the data structure, but they will also have their audience created in a matter of seconds!
Generative AI for Audience Segmentation, Myth or Reality?
Is it too good to be true? The short answer is: yes and no. We have undoubtedly come closer to a revolution in how business teams perform audience segmentation.
However, we are still in the early stages and not ready to have this technology deployed and in the hands of business users today. They are several barriers that need to be overcome along the way, including:
- Trust: Can the result be trusted? Large Language Models (LLMs) are known for their hallucinations. Therefore, users may have doubts about the accuracy of results obtained through natural language requests. Defining each term still poses a challenge in this context.
- Cost: While most vendors have been absorbing up the costs of testing, feeding and improving their models, AI models are extremely expensive to run. Organizations and vendors will have to find a balance between the value derived and associated costs.
- Data: Promising results have been achieved with clean data and well-structured schemas. However, the reality and complexity of organizational data and data structures can potentially undermine these results today.
There is no doubt that AI and generative AI use cases like this will be part of our future. The question is more about “when” than “if.” And when I say “when,” I am referring to the emergence of new innovative usage, as AI is already present in today’s world, supporting CX activities through models that help identify purchase propensity, predict churn and more.
That being said, ChatGPT and similar alternatives are not going to replace humans. They are going to serve as a copilot for business users.
In our upcoming blog post, we will double-click into this notion and share the results of our own experiments with generative AI and models.