Productside Webinar

Proto to Proven

A live working session on validating your personas

Date:

10/14/2026

Time EST:

1:00 pm
Watch Now

Your persona has a name, a face, and a backstory. It might not have a single customer behind it. 

AI can produce one in 90 seconds that looks exactly like a validated one. Same boxes, same confidence, opposite predictive value. Most teams can’t tell which is which, and the roadmap gets built anyway. 

In this working session you’ll start with an AI-generated persona, pressure-test it against real customer data, and watch it break. Then you’ll take the same workflow back to your own. 

What You’ll Learn:  

  • How to tell a proto-persona from a validated one, and why that changes every prioritization call you make 
  • Which persona assumptions carry the most risk, and which data already sitting in your systems can kill them today 

Introduction and Housekeeping

Welcome and Speaker Introductions

Rina Alexin & Tom Evans | 00:00:00 – 00:02:31

All right, welcome. We’re going to wait a couple of minutes for people to come in. As you arrive, please let us know where you’re joining from in the chat. I’ll go first.

Welcome, everyone. This is a really great webinar. Today we’re going to be talking about something I hope we do more than once a year: how to actually validate personas. From proto persona to proven persona. My name is Rina Alexin. I’m the CEO of Productside. Unfortunately, you don’t get Roger today, but you get me instead. I’ve been running Productside for the past eight years and I’m excited to be here with our very own Tom Evans. Tom, why don’t you give a quick introduction?

Tom Evans | 00:01:20 – 00:02:31

Absolutely. Tom Evans. I happen to live here in Austin, Texas, and I’ve been a trainer and consultant with Productside for 16 years. It’s been a wonderful 16 years, with lots of great experience and lots of opportunities to interact with our product community and help improve the practices of product management around the world. Excited to lead this webinar today.

Rina Alexin | 00:01:55

Tom, you’re always really good at making good decisions, and I listen to you, so that’s why I moved over here to Austin, which is lovely.

Housekeeping and Productside Overview

Rina Alexin | 00:02:31 – 00:05:44

A few housekeeping items before we get into the webinar. I always encourage everyone to chime in the chat and share where you’re calling in from. It’s a good idea to say something first because it makes it a lot easier to ask questions later.

For those of you who don’t know us: at Productside, we are an outcome-driven product partner. We live and breathe outcomes, our own and our clients’. When we partner with organizations, we know that transformation is more than just content and learning new practices. That’s the easy part. The hard part is actually adopting those practices, making the right trade-offs, improving your strategy, and communicating it effectively to everyone in the organization who relies on your decision-making.

Two components of our complete transformation solution:

  • Tailored engagements: we tailor all programs to your specific context so they address the real gaps and challenges your teams face, not generic ones
  • Invested experts: you’ll meet one of them today. Tom is wonderful to work with, and when we say invested, we mean deeply committed to driving results for our clients

A couple of quick logistics:

  • Please use the chat freely and drop questions in the Q&A box. We welcome the discussion, it makes things more interesting for everyone. Do not be shy. Give us your hardest questions.
  • Everyone will receive a link to the recording after this webinar ends, so don’t worry about recording it yourself
  • Connect with us on LinkedIn. The link is in the chat. It’s where product leaders from all over the world share leadership best practices, discuss how to make product management more effective, and network with peers

And now, for the fun stuff.

Understanding Proto Personas vs. Validated Personas

Webinar Agenda and Why Personas Matter

Tom Evans | 00:05:44 – 00:09:02

Here’s what we’re going to do today. Four things:

  • Talk about the difference between a proto persona and a validated persona, and what gets in the way of actually validating personas
  • Meet Rachel, an AI-generated persona, and find out whether she’s actually a good persona
  • Run an exercise with real data so you can see how to validate whether what AI came up with actually holds up
  • Learn how to use internal data you already have to take a first step toward building a persona that is based on evidence, not just assumptions

The key point here is this: a proto persona is not yet a full persona. When we build a proto persona, it’s usually based on ideas, anecdotes, or opinions. It doesn’t become a real persona until we validate it with customer evidence of some kind. We’ll talk about what forms that evidence can take.

Why do we care about personas? The single biggest value I see is that they give everyone in the organization a shared focus on the same users, buyers, and influencers around your product. Here’s what that means across functions:

For product management:

  • Helps us understand who we’re building for
  • Gives engineering and development teams a clear picture of who they’re developing for, and the more empathy they have for that user, the better solutions they’ll create
  • When teams feel disconnected from the user, that’s when you get experiences that don’t live up to expectations

For product marketing and sales:

  • Helps us understand who we’re targeting in our go-to-market
  • Guides what we say, how we message, and what content will resonate
  • In complex B2B sales, having a clear definition of each persona, their problems, and how to communicate value to them is genuinely valuable

Poll: Who Built Your Current Personas?

Tom Evans & Rina Alexin | 00:09:02 – 00:12:09

Before we go further, let’s find out who built your current personas. We have a poll. Take a moment to select the team that built yours, or if you’re not sure, we have an option for that too.

I’m very happy that nobody so far is clicking “What personas?” (though I have definitely worked with companies where that was the answer).

Results: about 40% said the product team, another 40% said marketing, and about 20% said product and marketing did them separately.

What I’ve seen in organizations is that there’s no standard rule here. In larger organizations, marketing often takes ownership and then passes the personas on to the rest of the org. The challenge there is that a set of five or seven personas built for a large enterprise won’t necessarily apply at every product level. That’s where product managers and product marketers need to step in and do their own persona work.

The other thing worth noting is that product and marketing often have different purposes for personas, which is why they sometimes end up doing them separately. I’m a strong proponent of creating a shared set wherever possible: it gives the whole organization a consistent language. But I’m also realistic that there are times where product needs and marketing needs diverge. What matters most is that you’re actually doing the research, which is exactly what the next slide is about.

The Impact of Persona Research

Tom Evans | 00:12:09 – 00:14:37

Some data worth knowing. Nielsen Norman Group found that just over 50% of personas being used were built mostly without research. So roughly half are built without empirical evidence, and just under half do use research to build them. That’s the good news.

Here’s the stronger data point, from research by Cintel:

  • 70% of companies that missed revenue goals never interviewed buyers to understand and create their personas
  • 82% of those that did conduct qualitative persona research achieved their revenue goals

This is correlative, not causal, but I think it has strong implications. There is likely a meaningful cause-and-effect relationship in there.

And here’s the challenge: using AI, we can literally build a persona in about 90 seconds. When we do this in training classes, it takes maybe two or three minutes. But we still have to remember that until that persona has been validated with true customer evidence, it is just a hypothesis. It is a proto persona. It is not something to put full trust in.

So how much trust should we actually put into it?

Evaluating Persona Creation Methods and AI Pitfalls

Tom Evans & Rina Alexin | 00:14:37 – 00:20:58

Let’s look at some of the different ways personas get built and how much confidence we should have in each.

  • Team workshop. A group comes together in a conference room and builds personas based on opinions. It looks pretty finished and has some useful qualities. There is evidence of a kind, because hopefully the people in the room have customer-facing experience. But we haven’t validated it with the right level of rigor.
  • AI-generated personas. The output is beautiful. It looks polished and complete. But what we don’t know is where AI got its evidence. Is it actually evidence that relates to our specific target customer segments? There’s no way to know. I wouldn’t put a lot of trust in an AI-generated persona, but it is a great starting point and a solid hypothesis.
  • Internal data. This is what we’re going to do in today’s exercise. While it’s not the same as conducting full interviews and surveys, which will always give you the strongest evidence, using internal data gives you much more confidence than workshop opinions or AI output alone. You’ll see that today: using real data, we get some pretty good results.
  • Interviews and surveys. The gold standard. Highest confidence. Takes the most time and effort, but produces the most reliable evidence.

A story that sticks with me: I was doing work with a credit union about five or six years ago. One of the branch managers said to me, “Marketing shares these personas with us, but they don’t look anything like the customers who actually come into the branch.” That is exactly the problem that happens when we build personas in internal workshops or with AI and never validate them against real customers. Customer-facing teams see the gap immediately.

Now, a word specifically about AI. There’s research from Stanford suggesting that AI tends to agree with you more than it challenges you. Its default is to please. And from Nielsen Norman Group, when they compared AI-generated research participants against real customer research, the AI output was described as too shallow, too generic. It doesn’t get into the depth of what users would actually say or feel.

Rina Alexin: Something that sticks with me: I remember a founder who was building a company to help product people test against synthetic users. At the end of his pitch, he asked the audience to give him feedback on his product. So, he couldn’t ask his own synthetic users. He still needed real humans to give him feedback.

Tom Evans: Synthetic can get you started and help you frame hypotheses, but in the end, it comes down to real human interaction to truly understand what is going on.

Interactive Exercise: Analyzing and Validating AI Personas

Introducing Rachel and Poll: Which Claims Do You Trust Least?

Tom Evans & Rina Alexin | 00:20:58 – 00:26:31

Let’s get into the exercise. But first, let’s meet Rachel.

Rina Alexin: Before Tom starts, a quick heads up: we’ll be running an interactive exercise in Mural. Ideally you’re at a laptop. The link to the Mural board is in the chat now. This is a visitor link, so you don’t need to sign up. Just click and join us.

Tom Evans: So here is Rachel. We used a simple prompt to generate a persona for a finance operations manager who buys expense management software for midsize companies. It looks fantastic. It looks well thought out and reasonably complete. But the question is: how much of this can we actually trust?

Our next poll asks: which of Rachel’s claims do you trust least?

Rachel’s claims include:

  • Primary pains: late and incomplete expense reports, missing receipts, manual data entry, employees hate the current tool
  • Buying authority: she is described as the decision maker, approver, and champion. She owns the budget and makes the final call
  • Desired outcomes: more expense reports submitted on time and complete, less chasing, better policy compliance
  • Jobs to be done: get everyone’s expenses in, approved, and paid without chasing people; feel in control at month end

Or, you could trust all of it. That’s also an option.

Results: the two biggest responses were her buying authority and her desired outcomes. I’m glad nobody trusted all of it. And you’ll notice I didn’t ask about bio and demographics, because those are the least impactful for guiding product decisions. If we get the picture wrong or the age estimate off, that’s not going to bust our business case or break the roadmap. Getting her decision-making authority wrong, or getting her primary pain wrong, those will.

Rina Alexin: Tom, that’s exactly the kind of time we need to take with AI output. Especially if you’re earlier in your career, or if you’re a leader managing someone earlier in their career: we’re hearing more and more that product leaders are frustrated with teams who are not questioning their AI output.

Tom Evans: Exactly. You own the decision. AI can hallucinate and it can produce something confidently that isn’t accurate. If I do step-by-step prompting, building incrementally, and challenging the output along the way, I get a much better result than when I just say “create something for me.”

Exercise Part 1: Challenging AI Hypotheses in Mural

Tom Evans & Rina Alexin | 00:26:31 – 00:32:05

All right, let’s get into the exercise. We’re going to do two parts. Part one is a quick one.

Here’s what to do in Mural:

  • Right-click the Rachel persona image and select “Download image” to save it to your computer
  • Right-click the green-screen prompt box, select “Copy text” to copy the prompt
  • Open your preferred AI tool (Copilot, ChatGPT, Claude, etc.), paste the prompt, and upload the persona image

What the prompt does: it asks the AI to evaluate the persona and identify which claims it would trust least, which ones need the most validation, and what evidence would be needed to validate them.

This is actually a great general technique when working with AI: after AI gives you a result, challenge it. Ask it: “Out of everything you just told me, what are the top three things I should investigate further? Which of these claims might not hold up?”

Here’s what I got when I ran it:

  • That Rachel is the actual buyer and decision maker: high risk. The evidence needed is confirmation of whether finance ops actually makes the final call or whether the controller or CFO is the real decision maker.
  • That ease of use will cause employees to submit expenses on time: high risk. This assumes that the behavior problem is caused by the tool, not by other factors outside the product.
  • That more timely and complete expense report collection is the primary desired outcome: worth validating. There may be a deeper outcome she cares about, like faster month-end close, that this is just a proxy for.

Rina Alexin: I got the same first two. Decision authority was my top pick. I genuinely don’t think this person can make the purchase. And the ease-of-use assumption is one I’ve heard come up repeatedly when I ask product leaders about bets that didn’t pay off. Usually the bet had an underlying assumption that changing the product would change user behavior, and the realization afterward was that the behavior was influenced by something completely outside the product. My third was “employees hate the current tool,” which sounds true but is actually really hard to validate.

Tom Evans: Yeah, that one is probably fourth or fifth on my list. Let’s keep moving.

Exercise Part 2: Building Evidence-Based Personas with Data

Tom Evans | 00:32:05 – 00:38:17

Part two is where we show you how to use AI with actual data to create a much more credible proto persona. This is the approach I have significantly more confidence in.

For this exercise, we have two data sources in Mural:

  • CRM ticket data: double-click the image and select “Download the file” to save it
  • Internal interviews: conducted with customer-facing team members, not the customers themselves, but people who interface with them daily. Same process: double-click and download.

Once you’ve downloaded both files, you’ll load them into your AI tool along with the prompt from Mural. The prompt asks AI to build a persona for a finance operations manager based on the actual data.

Here’s what I got when I ran it:

On buying authority: the data showed that finance operations is almost always listed as the champion in CRM records, not the decision maker. Multiple interviewees described finance ops as the champion. Signed deals were usually executed by the controller or CFO. So we now have data-backed evidence that debunks one of Rachel’s biggest claims.

On jobs to be done: the evidence points to getting expenses approved, reconciled, and posted correctly at month end, rather than simply “getting expenses submitted without chasing people.”

On desired outcomes and pains: the data fills in specific, evidence-backed bullet points, each labeled with which source it came from.

The other thing I want to do: make this more succinct before presenting it. I asked AI to consolidate it down to the top three to five points in each area, and it tightened the output significantly. Then, to make it ready for stakeholders, I asked it to apply the data to the persona template format. You feed it the Rachel template image as a reference, and it generates a new persona, same format, different content, based on what the data actually said.

One thing to notice: the new persona has no demographic or biographical information because neither the CRM data nor the ticket system had any. That’s okay. Demographics are the least important part. What matters is getting the pains, the buying authority, and the desired outcomes right.

Formatting and Comparing Validated Personas

Tom Evans & Rina Alexin | 00:38:17 – 00:46:32

Let’s call the data-backed output “the new Rachel” and compare it to the original AI-generated Rachel.

Here are the most important differences the comparison surfaced:

  • Core job: the original focused on keeping expenses submitted and paid without chasing people. The evidence-based version shifts it to keeping month-end close on schedule by managing approvals and reconciliation. A different center of gravity entirely.
  • Primary problem: original said missing receipts and incomplete reports. Evidence pointed more toward reconciliation effort and close-cycle delays.
  • Decision authority: this changed most significantly. Rachel in the original is the decision maker. Rachel in the evidence-based version is the champion. The controller or CFO makes the final call. Getting this wrong loses deals.
  • Employee experience as the primary driver: the original treated it as the most important factor. The evidence said it’s important but not the primary one. Close performance and operational visibility matter more to the person you’re actually selling to.
  • Mobile app as a key feature driver: the original flagged this. The evidence said it was not necessarily true for this persona and this use case.
  • Compliance as a top desired outcome: mentioned in the original. The evidence said close performance and operational visibility appear more often and more strongly.

Tom Evans: So, you can see that our data definitely debunked some of what the AI-generated persona said. Getting decision authority wrong is the biggest one, because that directly affects how deals close and how you message to buyers. Getting the primary pain wrong means you’re probably solving for the wrong problem, which affects your roadmap. Getting the age estimate slightly off is not going to hurt you.

Rina Alexin: Tom, there’s one correction I’d point back to when building the prompt: you mentioned that the influencers section ended up focused on internal influencers. What we really want to know from a marketing perspective is who externally influences their decisions. That would be a refinement to bring back into the prompt.

Tom Evans: Exactly right. That’s a specific correction to make in the instructions before the next time you run this workflow.

Leveraging Internal Data for Persona Validation

Tom Evans | 00:46:32 – 00:49:14

So, how do you go forward with this in your own organization?

Today we used support tickets and CRM records, and we supplemented those with a few internal interviews with customer-facing team members. Here’s how to think about each data source and what it tells you:

  • Support tickets: surface the most common pains, what customers bring up most often, and where volume spikes
  • CRM deal data: tells you what the buying process looks like, who plays what roles, what pains drove or stalled deals, and if you have win/loss data, who your best customers actually are and where you should be targeting
  • Customer-facing interviews (CS and sales): gives you deeper “why” behind what the data surfaces. Not the same as talking to the customer directly, but often very close

When going out to your customer-facing teams, ask almost the same questions you would ask a customer: where are customers running into problems, what’s causing those pains, what do they like about the solution, what don’t they like? If you already have a persona, present it and ask: does this look like the customers you actually see? Are we on target?

That brings us to the end of the content. Tom already mentioned it, but our Persona Pack is available for download. Stick around for the QR code and the final section.

Q&A and Closing Remarks

Q&A: Entering New Markets and Aligning Teams

Rina Alexin & Tom Evans | 00:49:14 – 00:52:38

Question:
Everything we covered is useful for figuring out who we’re already serving using internal company data. How can AI help product managers who are trying to enter new markets or add new personas?

Tom Evans:
A couple of approaches. First, look at your existing CRM data: are there any deals or opportunities outside your typical market segment? Even deals you didn’t win can be indicators of adjacent opportunities you’ve already pursued but weren’t quite the right fit for.

Beyond that, I’d use AI for initial external research:

  • Start by describing your current market, who you typically serve, and the problems you solve, then ask AI to generate a list of other market segments worth considering
  • Once you have a candidate list, use deep research mode to run a more complete market analysis: where is growth occurring, who are the competitors, where is there white space
  • Treat everything that comes back as an initial hypothesis, then prioritize which to dig into further

What may have taken a couple of months through traditional research, you can now move through in a day to get to a candidate list. The deep validation still has to happen, but you’ve dramatically compressed the time to that first prioritized list.

Question:
What advice do you have when product and marketing have different perspectives on personas?

Tom Evans:
This goes back to what we covered earlier. Product personas are primarily about understanding user needs and building the right product around them. Marketing personas are primarily oriented toward go-to-market: how do we reach and communicate with buyers and influencers? Those purposes can diverge, especially in larger organizations with complex product portfolios.

My honest take: in a large organization, I might just accept that marketing will build buyer-oriented personas and I’ll build user-oriented ones. That’s okay. In a smaller organization, the opportunity to create a shared set that serves both is much higher, and worth pursuing. If you’re the one driving it, make sure you’re sitting down with both product and marketing and asking: what do you actually need a persona to tell you? Then build from there.

Rina Alexin: I’d also encourage a collaborative session between marketing and product. Often the user and the buyer are two very different people, and marketing needs to reach the buyer while product needs to serve the user. But they end up caring about a lot of the same underlying data. And personas can go stale, which is a conversation worth having together even if you end up needing slightly different outputs.

Q&A: Overcoming Resistance and Navigating Regulated Data

Rina Alexin & Tom Evans | 00:52:38 – 00:58:26

Question:
How do you get buy-in from leadership for doing persona work? I’ve seen teams resist taking time out of their day to refresh personas that have gone stale.

Tom Evans:
I’ve heard this often. “We’ve been in this industry for 20 years. We know our customers.” And yes, you knew them, but buyer behavior has genuinely changed over the last five to six years. That’s probably not enough to sell leadership on it by itself, though.

Here’s what I’d do instead: show, don’t tell. Getting a download from your CRM or ticket system may take jumping through a few hoops, but once you have that data, look how quickly you can actually work with it. I’d come back to leadership and say: “I looked at our actual data. Here’s what it’s telling us about our customers right now. Here’s what has changed.” That evidence-based reasoning is a much stronger argument than “we should refresh our personas.” The work we did today took about 30 minutes of actual AI prompting. That story will land.

Question:
What about teams working in regulated industries like financial services or healthcare where there are data privacy considerations around customer data?

Tom Evans:
The key distinction is personally identifiable information, or PII. Most of the data we need to build a good persona is not PII. As long as you can get a file that has been screened of any PII, name, contact information, account numbers, you should feel comfortable and confident working with that data to build personas.

In healthcare, there may be additional considerations around diagnoses or medical history. That’s a conversation to have with your regulatory team. But it should not prevent you from getting good data to work from. You can absolutely get what you need for persona building without having personally identifiable information attached to it.

Resources, Upcoming Events, and Wrap-Up

Rina Alexin & Tom Evans | 00:58:26 – 00:59:52

Thank you so much, Tom. This was a great presentation and I genuinely enjoyed going through the exercise myself. Some really good AI prompting in there, and reminders that people often forget to ask for evidence when they’re working with AI.

Before we close, a few things to share:

Thank you all again for joining us today. I hope you learned something. I certainly did. And from all of us at Productside: let’s do product better, together.

Webinar Panelists

Tom Evans

Nearly 5,000 product managers across 150-plus companies have learned from Tom’s expertise. He knows exactly how to unlock product management potential.
LinkedIn

Rina Alexin

Rina Alexin, the CEO of Productside holds a BA with honors from Amherst College and an MBA from Harvard Business School. She is also a member of the AIPMM.
LinkedIn

Webinar Q&A

Persona validation is the process of testing every assumption inside a persona against real customer evidence, such as interviews, CRM records, support tickets, and sales data, to confirm whether the persona accurately reflects how customers think, buy, and behave. Product teams need it because AI-generated personas look polished and complete but are built on generic assumptions rather than verified customer data. Nielsen Norman Group research shows that just over 50 percent of personas in active use were built mostly without research, and Cintell data cited in Productside’s session found that 70 percent of companies that missed revenue goals never interviewed buyers to validate their personas. An AI persona is a hypothesis until it is tested against real evidence; treating it as fact leads teams to build for a customer that does not actually exist.
Validating AI-generated personas with internal data starts with the two richest sources most product teams already have access to: CRM records and support ticket systems. CRM data reveals who actually champions or blocks deals, why opportunities stall, and what pains surface in win-loss patterns. Support ticket data captures the problems customers report most often in their own words. Productside trainer Tom Evans demonstrates a repeatable workflow: load both data sources into an AI tool alongside the proto persona, then prompt the AI to evaluate each claim, flag which assumptions carry the highest risk, and surface the evidence that supports or contradicts each one. The output is a data-backed, evidence-cited persona that replaces assumptions with something product, marketing, and sales can all act on with confidence.
AI-generated personas are not reliably accurate without validation. Research from Nielsen Norman Group found that AI-generated research outputs are often too shallow compared to those built from real qualitative interviews, and Stanford research cited in Productside’s session found that AI tools tend to produce outputs that agree with the user rather than challenge their assumptions. The riskiest persona assumptions to leave unvalidated are those that directly drive roadmap priorities, pricing, and go-to-market messaging. In Productside’s live exercise with a finance operations manager persona named Rachel, the two highest-risk assumptions were decision-making authority (whether the persona actually owns the budget and makes the final purchase call) and primary pain (whether the core frustration driving the persona’s behavior is the one the product is solving for). Getting either of those wrong sends sales to the wrong person and puts the roadmap in the wrong direction.
A proto persona validation workflow for product managers moves a draft persona from assumption-based to evidence-based through five repeatable steps. First, list every claim the persona makes, covering role, buying authority, primary pain, desired outcomes, and jobs to be done. Second, rank each claim by its risk to product and go-to-market decisions: assumptions that affect roadmap priorities and deal outcomes rank highest. Third, gather real customer data from CRM records, support tickets, and interviews with customer-facing team members. Fourth, load that data into an AI tool alongside the proto persona and prompt it to compare each claim against the evidence, flag gaps, and identify which assumptions need the most investigation. Fifth, revise or retire claims that the data does not support, and document the evidence behind the ones that hold up. Productside’s Persona Pack includes templates for each stage of this process.
Product managers and marketing teams often generate personas for different purposes, and those purposes do not always produce the same output. Product focuses on understanding who uses the product and what problems they are trying to solve. Marketing focuses on who to target, how to message to buyers and influencers, and what content will move them through the funnel. These needs overlap significantly, and Productside recommends running a collaborative session between both functions before building personas, so that each team’s requirements are understood upfront. In smaller organizations, a single shared persona set built from validated customer data can serve both functions well. In larger organizations with complex buying groups and diverse product lines, separate persona sets may be necessary. In either case, the underlying data sources, CRM records, support tickets, and customer-facing team interviews, are equally valuable to product and marketing, and both functions benefit from grounding their personas in real evidence rather than AI-generated assumptions that have never been tested.