Why I Don't Recommend Generative AI Customer Personas

Welcome to Part One of a two-part series. Below I share my serious concerns regarding the risks of adopting Generative AI tools like ChatGPT to create instant customer personas. In Part Two, I experiment with Gen-AI personas, sharing my observations and offering suggestions to mitigate the risk. Read on and stay tuned for the next installment!

My friend and mentor Daniel Roundy inspired a healthy dialogue with his recent LinkedIn post promoting a way to generate customer personas in three minutes.

This is Daniel:

According to Daniel, personas developed with input from real customers are great, but those culled from non-customer sources like ChatGPT are often good enough to get a project started. His post included a clever "chain of reason" prompt he created with a bit of code from Synaptic Labs to simulate an interactive persona creation workshop with ChatGPT. Kudos for creativity!

While I support using technology to automate a wide range of tasks and relish the time savings I'm personally realizing with AI, I strongly caution against Generative AI customer persona creation.

But First, What Exactly is a Customer Persona?

I assume, since you chose to read this article, that you're likely familiar with customer personas. However, for the benefit of my dad (Hi, Dad??) and anyone else unfamiliar with the term, a persona is a communication tool the organizations use to promote customer understanding and empathy. Created with aggregate data, a persona represents a key customer group. Each persona offers a unique snapshot of a particular customer type who either currently uses, or is likely to purchase or use, a company's products. It condenses the group's essential attributes into a digestible summary and commonly features insight into customer goals, desires, behavioral triggers, attitudes, challenges, and more. Personas help organizations create a shared understanding of important customer types, bringing them to life so that teams remember who they're designing for and what matters most to them.

Sample Customer Persona

Personas vary according to their intended uses and business context, but here's an example from Justinmind's collection of free persona templates to give you an idea:

Crafting Personas with ChatGPT

The persona development approach Daniel created doesn't require us to input any private company data into ChatGPT, potentially running afoul of data security requirements and data privacy regulations. Instead, it relies on ChatGPT's processing model and the public information on which it was trained. All a persona creator has to do is specify a goal, target audience, and up to six customer attributes of interest. Then ChatGPT proposes subject matter areas on which to focus its search, offering the opportunity to change them if desired--that's seriously cool!

A single persona-creator can answer a few questions, hit 'Generate,' and Voila! Why hold-up the Dev team while we chase customers to interview? ChatGPT can deliver that persona in minutes!

For the full discussion, here's the link to the original post: https://www.dhirubhai.net/feed/update/urn:li:activity:7112631668628185088/

The Case for ChatGPT Personas

Admittedly, I agree with Daniel's key point--personas created without primary customer research can be enough to get going on a project. He refers to these as "hypothesis-based" personas, meaning they're built with other people's assumptions about the customer. When developing personas, I start in a fairly similar way. I pool whatever internal customer knowledge is available, usually from individuals closest to the customer, seek-out existing product and operational data and any past customer research that's relevant. But next, I look for obvious gaps in our company knowledge and rank order internal beliefs that need to be validated with actual customers. I use our insider perspective as the foundation for creating a research plan and customer interview guide.

The Agile scrum master in me values an incremental and iterative approach to complex projects and asset development, starting with the minimum amount of work necessary to satisfy the immediate objective and fleshing-it-out over time with frequent infusions of customer input. So, getting started with a minimally viable persona as Daniel suggests appeals to me. By all appearances, Generative AI can be a great help for accomplishing that goal.

The Upside of AI Personas

As I see it, the following are some attractive benefits of Generative AI personas:

1. Expediency. Generative AI personas are lightning fast compared to building them from scratch. It can take weeks to months to create a high-quality research-based persona depending on the scope of a project and how easily we can get time with our chosen customers.
2. Vast Data Access. The data set on which ChatGPT is trained is massive, offering the potential for a greater diversity of insights than we might find within our own more limited data set. AI also has the potential to pick-up on patterns we humans would miss.
3. Economy. Near-instant personas consume minimal resources of time, human effort, and finances.

So, What's My Problem?

My problem is that promoting easy AI-based personas puts us at risk for committing CX malpractice. (Admittedly, that and many other assertions I make here are hypotheses that require testing). I believe individuals endorsing personas created in the absence of customer conversations underestimate the risks of shortcutting the process and are contributing to a widespread impression that ChatGPT personas are sufficient. For the record--Daniel Roundy never said that ChatGPT personas should replace research-based ones. But I've witnessed others in positions of influence endorsing exactly that...many times.

The Risks

Below are three of the risks of using Generative AI (and similar shortcuts) in lieu of direct research with customers.

1. You'll make costly mistakes. Strategy catastrophes frequently trace back to critical failures in understanding a brand's core customers, and the proof is all around us--New Coke, JCPenney, Bud Light. By reducing the role of customers in customer research, we are at extremely high risk for misconstruing information, basing decisions on notions that are completely wrong, and missing game-changing nuances, attitudes, and motivations known only to customers. Instead of putting our target customers' interests first, by leaning on ChatGPT, we inadvertently elevate the biases of its AI handlers. So, unless we're selling to the top 1% of data science geeks, I think it's safe to say that the people who developed and trained our Generative AI algorithms have vastly different needs, values, and motivations than our target audience. And let's not forget, when ChatGPT cannot find sufficient data related to our queries, it's been found to lie (return fabricated responses). What's fact? What's fiction? It's hard to tell. How confident can we be making business decisions based on suspect data? Bottom line, substitute AI for your customers' truth and risk getting it wrong.
2. Your personas will be generic, outdated, and offer minimal value. I've personally experienced the power of rich personas to spark meaningful conversation, open minds, and unify teams. I've also seen zillions of bad personas that offer little texture and no actionable insight. ChatGPT builds personas from information available on the internet (or that you enter, and I caution, please Do Not put any information into a Generative AI database that you wouldn't want shared with the world). The information ChatGPT returns may or may not be representative of our target customer or it may be too general to be useful. (I intend to elaborate on this in my upcoming Part Two article coming soon). And unless you've made the switch to ChatGPT's new beta version released just a few days ago, it's only trained on data up to 2021. Customer sentiment and expectations change rapidly, and the information you receive may be horribly outdated. I updated to the beta right away, but I know nothing about how recent web content is weighted and integrated with older data. Gen-AI engines are a big black box. However, if your organization is currently using shallow personas based on what anyone other than your actual customer thinks, Generative AI will help you produce cr*p personas faster.
3. You will struggle to understand your audience (and they'll know it). Shortcutting customer research limits an invaluable source of strategic insight, which in turn runs the risk of sacrificing marketing effectiveness, customer engagement, and brand strength. Moreover, genuine customer empathy is a powerful motivator that sparks employee engagement and unites teams around a shared vision. When we detach from customers, we risk dampening these potent forces and constraining our potential for success. In my experience, leaders believe they know their customers far better than they actually do. You've likely seen the Bain & Company statistic that eighty percent of organizations believe they deliver a superior customer experience, but only eight percent of their customers agree. We see that disconnect playing-out all the time. So, couple corporate ignorance and complacency with the allure of super efficiency, and vast amounts of customer research will be reduced to AI queries. It's already happening. Worse, reducing the role of the customer in insights gathering threatens our ability to gain and maintain support for all conducting customer and market research. I've been asked by a superior, "is talking to customers really a good use of your time?" As AI experts and UX/CX luminaries champion the Generative AI persona use case, direct customer research is being undervalued. We're practically begging executives focused on budget cutting to scale back investment in customer and market understanding, creating even greater distance between ourselves and the people we're most interested in attracting, engaging, and retaining.

In Conclusion

I get it--customer research is time-consuming and costly. Taking too long to get into action is also a serious risk. But I urge caution. Despite its undeniable benefits, I'm deeply concerned about the misapplication of Generative AI in customer research, a trend gaining traction across UX/CX, product management, and marketing circles. While opting for a cheap and speedy approach is tempting, embracing an inaccurate view of our customer ultimately carries a far steeper price.

How might we make this work? Believe me, I want to be quick just as much as anyone. In Part Two of this series, I'll be putting Daniel's interactive persona-creation workshop prompt to the test, reporting my observations and offering suggestions for mitigating the risks. I invite you to join me on my journey to explore the evolving landscape of AI in CX.

Now, Let the Experiments Begin!

?