The User Donna Was Built For

Donna was founded on a specific, recognizable moment. A field sales rep finishes a meeting, gets in the car, drives home, and feels like the meeting went well. A week later, the rep is standing in front of their manager who asks what the next steps are. The rep hesitates. Did you take notes? Not really. Did you update the CRM? Not really. Who wants to update their CRM? The founders' bet was that almost nobody does, and that voice is a format the rep will actually use because it fits the natural moment between the meeting and the next appointment.

The mechanics followed from that bet. A field sales rep spends hours every week in a CRM. The CRM is the central touchpoint for their organization. If the rep fills it in poorly, the whole company downstream inherits the gap. Donna exists so the rep can skip the typing and talk. She turns a spoken recap into structured CRM updates, action items, and reminders, so the rep gets the meeting documented without opening the CRM at all.

The Ten Percent

“Only about ten percent of our end users have ever used AI. And by used AI, I mean they have typed a single query into ChatGPT. That is what we are working with.”
- Gheerwijn Clicque

This number landed in the room because most of the product people there have been building AI features for users who already have a mental model for what AI is. Donna's audience does not have that model. They do not have intuitions about hallucinations, context windows, prompt injection, or why it sometimes confidently gets things wrong. They do have very strong intuitions about what CRMs are supposed to do, what a good sales meeting looks like, and what their manager expects in the next pipeline review.

The product design problem that follows from this is not a small one. Every affordance, every piece of language, every confirmation prompt, and every recovery path has to be legible to someone who cannot fall back on ten years of AI habits. That forces a level of explicit design that a product built for power users can skip.

Training Users Is Not Optional

Gheerwijn spent a few minutes on a topic most product teams would rather avoid. Donna's users need training. Not a quick-start guide. Actual, explicit training on how to talk to an AI assistant.

The first surprise for new users is that Donna cannot handle yes or no answers. If a rep tries to shortcut a question with a single word, the speech-to-text layer loses the context and Donna's downstream processing falls apart. Users have to speak in complete sentences, which is not how most of them talk on the phone in the car. That one constraint alone requires explicit instruction, ideally from the customer success team before the rep ever opens the app.

infoWhy the Training Cannot Be Avoided

What Donna actually does is distill an existing sales workflow into a structured CRM update. Every company's CRM setup is different. The level of detail expected in a visit report, the custom fields, the triggers for task creation, all vary. The Donna team cannot ship one product that is instantly right for every customer, and they cannot automate around the human factor of a sales rep who has never been asked to be concise with a computer before.

Gheerwijn flagged this as a scaling problem the team is aware of. High touch customer onboarding works at their current size. It will not scale forever. But he pushed back on the idea that the solution is to make Donna simpler. The CRM workflow underneath is genuinely complex, the customers are enterprise accounts with millions of contact records, and the problem Donna is solving is not one that responds well to radical simplification.

The Product Is the UX

The second half of the talk pushed back against a frame that almost everyone in the audience walked in with. Traditional SaaS product teams treat UX as a layer that sits on top of the product. There is a backend, there is business logic, there is an API, and then there is a UI that presents all of it. UX is the thing designers argue about on Friday afternoons.

Gheerwijn does not work that way, and he made the case that nobody building conversational AI can afford to.

“The UX is the product. Our product is the UX. There is nothing in between. We have built something that interacts with an end user, and that end user does not know what is going on under the hood. There is no mental model to correlate it to. For us, the most important thing is to realize that, and to approach every iteration as a design thinking exercise.”
- Gheerwijn Clicque

This shift is not just rhetorical. It changes how the team works. Prompts are not written by solution engineers in isolation, they are designed through the same process that used to produce wireframes. Every conversational flow is treated as a user journey, mapped, tested, and iterated. The solution engineers and the product owners sit in the same room as the UX designer. The artifact of that collaboration is not a mockup. It is the prompt that will shape every future conversation with that feature.

The Day Donna Had an Off Day

The most charming moment in the talk was an accidental design outcome the team decided to lean into. Because Donna is positioned as a named, personified assistant rather than a feature, users started talking about her like a colleague. Which meant, in the early days when things went wrong, they did not say the product is broken. They said, Donna seems to be having an off day.

“From a product and business point of view, that was amazing. If a product is not working, you would expect users to churn. But with Donna, they would say, she seems to have an off day, I will come back tomorrow. That is a very different relationship with software than we are used to.”
- Gheerwijn Clicque

The team saw this as a signal that the humanization was resonating and began deliberately reinforcing it. The product gets a little more human with every iteration. The frustration users express when something breaks is closer to interpersonal frustration than software frustration, and Gheerwijn observed that it is a kinder failure mode than the one that normally kills an AI feature.

lightbulbAnd Yes, Enterprise Customers Asked About Donnie

An audience member raised the gender question directly. Enterprise customers had pushed back on the female assistant stereotype. The team has experimented with a male voice, which some languages handle worse than others, and is also iterating on a version they are calling dark Donna. The point, Gheerwijn said, was that the product has become personal enough that the conversation about it includes its identity, not just its features. That is a design outcome most SaaS products never reach.

The Workflow Underneath Is Ten Years Old

The part of the talk that resonated most with the other PMs in the room was Gheerwijn's honesty about what Donna is actually replacing. Sales reps who have spent a decade running their workflow a specific way do not describe that workflow as broken. It is simply their workflow. The team cannot walk in and announce that a new AI tool will make everything better, because from the rep's perspective nothing was bad to begin with.

This creates a design problem that is different from the one most consumer AI products face. Donna is not competing with the idea of doing less. It is competing with the idea of doing the same thing the rep has always done, slightly differently, with an added layer of learning. The only way through is to make the new workflow genuinely easier within a week, and to be explicit with both the customer and the rep about what the learning curve looks like.

The Salesforce Workshop Nobody Expected

Gheerwijn shared an anecdote that landed hard. The Donna team ran workshops with people from Salesforce itself, walking through the real workflows their enterprise customers had. At the end of the workshop, the Salesforce team said openly: this is not how you are supposed to use Salesforce. The customers doing things that way were enterprise accounts with real budgets and real pipelines. The gap between how Salesforce imagines it is used and how it is actually used is the gap Donna lives in, and the team has stopped pretending it does not exist.

When 90% of Your Users Have Never Used AI: What Donna Taught Us About Designing for the Unprepared

Download the Slides

The User Donna Was Built For

The Ten Percent

Training Users Is Not Optional

infoWhy the Training Cannot Be Avoided

The Product Is the UX

The Day Donna Had an Off Day

lightbulbAnd Yes, Enterprise Customers Asked About Donnie

The Workflow Underneath Is Ten Years Old

The Salesforce Workshop Nobody Expected

This article is for Consortium members

Key Takeaways

Related Editorials

Why AI Is a Product Challenge, Not a Technology Challenge

Chat Bolted on Legacy: Why Slapping a Chatbot on Professional Software Is Product Slop

One Agent Became Two: How Donna Split Conversation from Execution to Survive Enterprise Scale