AI Customer Personas: Five Ways to Avoid Costly Mistakes

Thank you for joining me for the final stretch of this three-part series that critically examines the rise of Generative AI usage for customer persona creation. The emergence of AI tools like ChatGPT is revolutionizing UX/CX, marketing, and product with AI aficionados and thought leaders widely promoting their rapid persona generation capabilities.

Note added April 2024: Since the time I first published the following article, my views on the subject have changed...or some would say, they haven't--I still strongly caution and do not endorse the popular practice of using Generative AI tools like ChatGPT, Google Geminin, etc., to produce customer personas. Rather than using tools that have been trained on public data sets, I recommend utilizing data-mining technology to extract detail from your own internal customer data for increased accuracy and data security. The recommendations I offer in this article are still applicable, particularly in areas where customer data is scanty or when relying on assumptive information such as from your frontline employees, both of which require additional verification. Hope this makes sense--please feel free to message me with questions.

A Quick Series Recap:

In part one titled, “Why I Don’t Recommend Generative AI Customer Personas,” I outlined pros and cons of Gen-AI personas, making the case that organizations should reject the "customer personas in minutes with Generative AI fad". Organizations who adopt AI generated personas risk making grievous errors and may undermine their own customer intelligence gathering capabilities. If you missed it, you can catch-up on the discussion here: https://www.dhirubhai.net/pulse/why-i-dont-recommend-generative-ai-customer-personas-micheleigh/.

In the second segment, “What I've Observed Experimenting with Generative AI Customer Personas,” I pressure tested my position, comparing Gen-AI personas with those developed with direct customer research. Based on my findings, I suggested trials with a combined Gen-AI/traditional customer research approach to assess Gen AI’s ability to balance speed with depth. You’ll find a summary of my observations at the following link: ?https://www.dhirubhai.net/pulse/what-ive-observed-experimenting-generative-ai-perez-au-d-ccxp-8bz7e/. ?

This final installment shifts from analysis to action. Here, I present practical strategies designed to minimize the risks inherent in Generative AI persona creation. May they serve you well.

?

My Over-Arching AI Persona Strategy

The risk mitigation tactics presented below support an overall strategy that harmonizes Generative AI with traditional customer research. Based on my experience comparing personas created with real people with their Generative AI counterparts [https://www.dhirubhai.net/pulse/what-ive-observed-experimenting-generative-ai-perez-au-d-ccxp-8bz7e/. ], AI personas aren't good enough to stand on their own. It's crucial to cross-check them with target audience data to minimize the high probability of error. Furthermore, adopting a blended approach supports deeper customer understanding for informed decision making.

To be clear, the tactics I present here are theoretical and await practical testing. They are best viewed as hypotheses that I believe hold promise for mitigating Gen-AI persona’s considerable risks. I offer them along with my rationale for your consideration. I look forward to evaluating them throughout the course of my CX work and invite you to join me.

?

Five Tactics to Reduce the Risks of AI Persona Creation

?1.????? Use Gen-AI as a Support Tool Only

I support anyone’s choice to reject instant Gen-AI customer persona creation. (I was inspired to write this article series in response to the extensive promotion of this practice throughout the AI community). The allure of customer personas in minutes is understandable, but the risks are great. Abstinence is the best and most obvious way to avoid them.

Yet for insight seekers employing traditional market research methods, Generative AI tools like Chat GPT and Bard can be powerful efficiency enhancers. I’m awestruck by the sheer number of hours I’ve saved using ChatGPT to develop interview guides. Similarly, I utilized DALLE-AI to create this article's cover image, drastically reduced the time I typically spend combing through hundreds of stock photography images. I see myself saving both time and money when creating graphical elements for personas and other CX deliverables with Generative-AI. Incidentally, I’m on the lookout for AI-powered design and data visualization tools that can swiftly, securely, and seamlessly transform my customer research content into visually appealing formats. As someone with limited graphic design ability, the time I could save formatting CX deliverables like personas and journey maps would be a total game-changer for me. If you’ve encountered a commercial tool that meets all three of these criteria, I’d love to hear all about it.

A Few Words of Caution:

I’ve encountered a few additional ways in which people are using Generative AI tools to support persona creation. These include:

• ?Analyzing and synthesizing your customer data to produce quicker insights
• Conducting sentiment analysis on text data
• Creating data visualizations
• Cleaning data prior to analysis

?

I don’t address any of these potentially persona-enhancing roles here. While artificial intelligence is a powerful tool for performing these functions, at this time I simply wouldn’t entrust my critical business data to an AI system unless it’s specifically privacy preserving and I’m confident all the appropriate safeguards are in place. Privacy-Preserving Artificial Intelligence technology (PPAIs) enable organizations to utilize their proprietary data in AI-driven applications without compromising data privacy and security. At this time, neither ChatGPT nor Google Bard are categorized as PPAI’s.

2.????? Understand the Risks and Devise a Mitigation Plan

For those who choose to delve into full blown Gen-AI persona creation, I recommend a deliberate approach to reducing the risks. I’d start by itemizing the risks specific to your business context and devising corresponding mitigation tactics for each one. Several primary risks are described in article one of this series [https://www.dhirubhai.net/pulse/why-i-dont-recommend-generative-ai-customer-personas-micheleigh/]. Since one of the potential negatives associated with Gen-AI personas is undermining executive support for all customer research, I’d be prepared to articulate these risks and explain their implications to executive stakeholders. This reinforce the importance of authentic customer intelligence and ward-off potential cuts to your budget. (See what I just did there?) ??

?

?3.????? Gain Upfront Commitment for the Entire Persona-Creation Process

Secure an upfront commitment from executive leadership for gaining authentic customer insight. Start by involving project sponsors in defining meaningful goals prior to getting started. Then frame persona work as a key milestone in the pursuit of your business objectives. Presume that an AI persona is just a stepping stone in the process, not the endpoint. Establish a realistic timeline and budget for customer exploration and validation. These measures will go a long way toward curbing senior leaders' inclination to shortcut customer research, risking costly mistakes.

4.????? Physically Distinguish Assumptive Personas from Valid Ones

One of the reasons executive leaders fail to appreciate the value of intelligence-rich personas is that they’re difficult to differentiate from low value ones. Business leaders frequently believe they already know their customers, and in fact, they do know a lot about them. In leaders’ view, it seems unnecessary to invest in furthering customer understanding, which in turn has led to a proliferation of personas built on internal business assumptions.

No matter how flawed, assumption-based personas look the same as their genuine counterparts. No one can tell the information isn’t valid based on their outward appearance. The same holds true of AI-generated personas. I’ve yet to encounter a voice of customer project where the business hasn’t overlooked something crucial or harbored significant misunderstandings about customers. The ability to create invalidated personas in minutes could serve to worsen this trend, leading to costly mistakes, missed opportunities, and further eroding executive support for customer research.

Professionals engaged in customer research can ward-off these possibilities by clearly distinguishing between assumptive and research-based personas. An effective way to achieve this is to create visible distinctions between the two types so that anyone viewing them can recognize their validity at a glance.

For example, assumptive personas could initially be presented in black and white, then gradually transitioned to color as the assumptions therein are validated. (However, for the benefit of individuals with color vision deficiencies, it’s inadvisable to rely solely on color to convey meaning. Employing additional indicators such as textual descriptions, markers, or shapes help to ensure distinctions are accessible to people who perceive color differently). In addition, prominently displaying a disclaimer and including verbiage such as “preliminary, validated/invalidated, hypothetical, assumed,” etc., in persona titles, category headings, and watermarks offer increased clarity. By omitting details such as names and images, hypothetical personas can be further differentiated from research-based ones. Persona designers might even include a list of hypotheses to test within the body of the persona to emphasizing the underlying need for validation. They could take this even a step further by creating a first draft persona consisting entirely of hypotheses to test, only converting them to declarations once they've been substantiated.

Below are two examples illustrating how persona differentiation might be achieved:

Example #1--A Partially Validated Assumptive Persona Created with ChatGPT

Example #2--Corresponding Research-Validated Persona

5. Eliminate Blind Handoffs

Present your personas—don’t just hand them off and leave it to others to interpret what they mean and how they should be utilized. Educate your organization about the goals of persona creation, the data on which they’re built, how they're useful, and the differences between assumptive and research-based personas. Make sure the risks of over-relying on AI and internal assumptions as proxies for customer intelligence are clear to everyone.

In fact, one-off persona reveals aren’t ideal either. Personas achieve their utmost usefulness when they become a recurring and instinctive part of everyday dialogue and business decision-making. Better yet, rather than handing-out personas and serving as the company expert on customer attributes, invite colleagues across the organization to join you in efforts to better understand your target audiences. You’ll likely find this significantly enhances empathy and understanding of the customer's perspective, fosters a sense of responsibility for the customer journey, and cultivates a culture centered on the customer's interests.

Some Concluding Thoughts

Thank you for coming along on my journey through Generative AI persona creation--I hope it brings you value. While my initial call for caution has not waned, I'm optimistic about AI's potential to enhance CX deliverable creation when it's used alongside primary research and risk mitigation tactics. (I believe that the issues described here apply not to persona creation but to other CX deliverables as well such as customer journey maps).

Reflecting on the journey, I feel confident that at least at this time, creating Gen-AI personas in minutes is not worth the risk. I'll also be interested to see how much time will actually be saved given the enduring need for human validation and frequently added exploratory research.

I'll be experimenting with the tactics outlined here and invite you to do the same. Thanks again, and I welcome your candid feedback regarding their feasibility, merits, and weaknesses in the comments section below.

Bonus Tip--How NOT to Validate AI-Generated Personas

As we seek to understand the capabilities and limitations of AI for our profession, I offer this final recommendation. If you choose to use Generative AI to craft first draft personas, there is one extraordinarily enticing practice that I sincerely wish to discourage--a validation process that looks something like this:

Stacy, a well-intentioned CX professional, employs a Gen-AI tool and a popular prompt that’s been circulating on the internet to craft a first pass customer persona. Within minutes, she has what appears to be a reasonable depiction of her target customers’ attributes, and she organizes them into a professional-looking persona template. Appreciating the need to validate these assertions, Stacy schedules a series of meetings with representatives of her target audience to gain their feedback. During these sessions, she walks participants through the preliminary persona, asking how well it describes them and reflects their views. Reviewers assure her that the content is consistent with their key characteristics. So, in very little time, Stacy confirms that her AI customer persona is largely on track, requiring few additions or corrections.

I anticipate that many CX professionals would default to a validation process much like this. In fact, I’ve been guilty of using a similar approach but for validating stakeholders' assumptions with customers. While straightforward and speedy, I do not recommend this approach because it's highly susceptible to acquiescence bias. Also known as agreement bias, my concern is the common tendency for research subjects to agree with statements regardless of their true viewpoints. I became painfully aware of this possibility while watching videos of myself validating CX deliverable during customer meetings. In an overwhelming number of instances, interviewees would nod and affirm the content of our CX prototypes when asked to evaluate them. Excessive agreement occurs for lots of reasons. Some people are naturally inclined to go along to be perceived favorably and avoid criticizing. It's especially true of cultures that emphasize harmony, and it can also occur due to differences in language interpretation with both native and non-native speakers. Perhaps more significant, the act of agreement requires minimal cognitive effort. Disagreeing, analyzing, and formulating a nuanced response places heightened demands on subjects, leading many to opt for the less challenging path of agreement. While quick and easy, asking customers to validate a series of persona-related statements, whether AI-generated or originating from some other source, is not a reliable method for uncovering truth. (This is yet another hypothesis that would be interesting to test--comparing the results of Stacy’s confirmatory customer validation approach with customer answers to exploratory questions free of preconceptions).

?