Built for the User Who Doesn’t Exist

Part 1 · The user who is always connected

Let me say something before anything else, because some of what follows will be uncomfortable to hear. This is not an episode about why AI is bad. I’m not going to tell you it’s overhyped, or dangerous, or a fad that will pass. AI is here. It’s already inside Salesforce, inside the tools your people use every day and inside the plans your board approved this year. You can’t avoid it, and you shouldn’t try. Some of what it does will genuinely make working days easier for people who have wrestled with clunky systems for years.

But.

There is a person at the centre of almost every AI plan I’ve read this year, and out in the field, I have never met them.

I’m Connie Hazendonk. For more than ten years I’ve trained people to use Salesforce, in more than a dozen countries, in offices, depots, stores, warehouses and training rooms that were sometimes just a table at the back of a canteen. I’ve met thousands of users. Out in the field, I have never once met the one in this episode.

Simply said, you won’t find this person on an org chart. Nobody drew them. But you can describe them precisely from the decisions that have been made around them. They sit at a desk. They have a fast, reliable connection that never drops. They have time, not the ninety seconds between two jobs, but a calm ten minutes to think. They’re comfortable with AI. They use it at home as well as in the office, they know what to ask, and they type a clear, thoughtful question into a clean screen and get a useful answer back. They look exactly like the person in the keynote demo. They also look a lot like the people who approved the investment.

To be fair, this person does exist. I meet them all the time, just not where most of the work in an organisation happens. They’re in the technical teams who build and look after the platform. They’re in the project teams mapping processes and testing changes. They’re the analysts, the operations people, the strategy and finance teams whose whole job is to explore data, ask questions of it and work out what it’s telling the business. They sit at a desk, on a good connection, and they’re paid to be curious.

For them, AI is fantastic, and I mean that. Being able to ask your data a question in plain language, dig into the answer and then ask a better question is a genuine step forward. An analyst who used to wait days for a report can now chase a hunch in an afternoon. A project lead can see how a process really runs before changing it. A technical team can find out why something behaves the way it does without opening twenty screens. Exploring is exactly what conversational AI is good at, and these people spend their working day exploring.

So it’s no surprise that this is where AI gets designed, tested and proven first. These teams are closest to the technology, closest to the decision-makers and closest to the budget. The pilot works, because it was run with people it was always going to work for. The success story goes to the board. And the board, very reasonably, asks why everyone else can’t have the same.

That’s the moment the imaginary user is born. Not from bad intentions, but from a good result in the wrong place. The experience of a few desk-based teams quietly becomes the assumption for the whole organisation: everyone sits down, everyone is connected, everyone has time to explore, and everyone wants to.

But most of the people who keep a business running aren’t exploring anything. They’re doing. They’re on the road, on site, on the shop floor, in front of a customer, with a job to finish and very little time to finish it. They don’t want to dig into the data. They want one right answer, quickly, wherever they happen to be standing. And often, where they happen to be standing, there’s no internet signal.

Those are the people I meet when I go out to train. And in most AI plans, they’re the people nobody has met.

I’ve said before that AI is being built for the 20% who never needed convincing. What I’ve come to realise this year is that it’s also being built for the 20% who have a good signal. And they often overlap.

So, let me take you somewhere else. A rep who visits stores is sitting in a car park between appointments. Her phone is propped against the steering wheel and she has one bar of signal, which sometimes, to her frustration, completely disappears. The next store manager is expecting her in eight minutes. She wants to know what happened on the last visit, whether the promotion went in and what that store still owes. She does what she has been told is the future during a recent training session. She asks the new interface a question. And the question, goes nowhere.

A car park is the easy version. I’ve worked with people in far more challenging places: remote regions, mountain areas, complex border zones, where internet, and even 4G or 5G, is never guaranteed. For them, no signal isn’t an inconvenience or an edge case. It’s a normal workday.

Most of the conversational AI being sold right now needs a live connection to work. Claude, which Salesforce has now built into how people reach their data, has no offline mode. Every question has to travel somewhere and come back. In the boardroom, nobody notices, because the boardroom wifi rarely drops. The same for all those keynotes where Claude is showcased to everyone.

Here’s the irony. Salesforce already learned this lesson. Its mobile app was built to work offline, precisely because the people who use it so often have no signal. Somebody, years ago, stood in the field, watched the problem happen and designed for it. The newest layer of the conversation feels as if it was designed by people who have forgotten that day ever happened.

And this is the part that worries me most, because it isn’t technical at all. Missing data never explains itself.

When a rep in a dead zone can’t update the record, the dashboard doesn’t show a dead zone. It shows a blank. A visit with no notes. An activity that was never logged. A deal nobody touched for a week. The people reading that dashboard, a long way from the car park, draw the only conclusion the screen allows them to draw. In the dashboard, a dead zone looks exactly like a lazy rep.

What happens next is predictable, because I’ve watched it happen. Leadership tightens the screws. More mandatory fields, more reminder emails, more compliance reports, more pointed questions in the pipeline review. Every one of those measures lands on someone who was trying to do the job in a place the plan never imagined. She learns that the system doesn’t understand her day, and worse, that the people above her don’t either. That lesson is very hard to unteach.

In September, Salesforce went down around the world for a few hours. Boards noticed. Emails went round, questions were asked about resilience and backup plans. That’s the right reaction. But for some of your people, part of every working day is an outage. Nobody sends an email about that one, because nobody in the room where emails get written has ever sat in that car park.

This is not an argument against AI in the field. Field teams could benefit from it more than anyone, because they have the least time and the most need for the right answer at the right moment. It’s an argument against designing for a person who isn’t there. Before any AI capability is switched on for a role, somebody should be able to answer a very simple question: where will this person be standing when they need it, and will it work there?

If nobody in your organisation can answer that, the plan isn’t finished.

Connection is the easiest assumption to spot, because you can test it with a phone and a walk to the car park. The second assumption is much harder to see. That’s because the people it leaves behind will never tell you.

Part 2 · The user who already speaks AI

Now, let’s pivot the story to another group. There’s a moment I’ve seen in meeting rooms more times than I can count. Someone at the front asks a friendly, throwaway question: who here already uses ChatGPT? Most hands go up straight away. A few go up a beat later, halfway, while their owners glance around the room to check they’re not the only one. And a couple never go up at all. Those people look at the table.

I’ve been training people to use Salesforce for more than ten years, and I pay a lot of attention to the people looking at the table. In my experience, they’re usually the people the plan forgot.

Every AI plan I’ve seen this year is built around an imaginary user. They’re always connected, they always have time and, crucially, they already speak AI. They use it at home. They know what to ask. They understand roughly what’s happening when they ask it. That assumption is so widely shared at the top of organisations that nobody thinks to check it, and it’s easy to see why. Everyone a senior leader talks to uses AI. Their peers use it. Their board uses it. Their consultants use it. The vendor demos it. Their children use it for homework. From the boardroom, AI looks as familiar as email.

I’d go further than that, and I’ll include myself in it. Those of us who work in and around Salesforce live in a bubble. Admins, consultants, architects, developers, trainers: learning new technology is our job. When something new arrives we’re expected to pick it up, so we do, usually quickly. We use AI every day on the work we know best, which means we can see straight away when it’s helping and when it’s wrong. And because it works for us at work, we start using it at home. We plan a holiday with it, rewrite a difficult email, settle an argument about a recipe. Before long, it simply feels like part of ordinary life.

That’s exactly the problem. Once AI is part of your ordinary life, it’s very hard to imagine a life where it isn’t. Everything inside the bubble confirms it. Our events are full of it, our feeds are full of it, and everyone we talk to is experimenting with it. So when we design a rollout, write a training plan or advise a leadership team, we quietly design for people like us. Not deliberately. It’s just the only picture we have.

The people we should be designing for don’t live in the bubble. For them, learning new technology isn’t the job; it’s an interruption to the job. They didn’t choose this tool. They didn’t see the demo. Nobody asked them what they needed. And many of them have never had a single reason to pick up AI at all.

From outside the bubble, AI looks very different.

I’ve met plenty of users who have never opened ChatGPT, never opened Claude, never typed a single question into anything that answers back. Not because they’re behind or difficult. Simply because nothing in their life has ever asked them to, and they’ve had no reason to go looking. The version of AI they know comes from headlines: jobs being replaced, fake videos, machines making mistakes with great confidence. For them, AI isn’t a tool. It’s something slightly mystical, a little weird and, quite often, something negative. And they are now being told that the way they’ll use the system they rely on every day is to have a conversation with it.

I once trained a group of very experienced sales people, who were receiving their first iPads. The plan was to train them on Salesforce. It quickly became clear that before Salesforce, and long before anything resembling AI, they needed help with the device itself. How to unlock it, how to find an app, what to do when it updated overnight and everything looked different in the morning. Not one of them was incapable. They were skilled, experienced people who could spot an opportunity and sell the customer the perfect solution. But the plan had assumed a starting point that simply wasn’t there. And “the younger ones will pick it up” is not a plan.

Here’s why boards almost never hear about this. People who haven’t used AI won’t tell you. When a room assumes everyone uses it, admitting you don’t feels like admitting you’ve fallen behind. In a year when AI and job losses appear in the same headline every week, it can even feel like admitting you’re replaceable. So people stay quiet. They nod through the demo, they don’t ask the question they need to ask, and afterwards they go back to whatever they know works.

The silence after an AI rollout isn’t resistance. It’s shame. And because it’s silent, it never reaches the people making decisions as shame. It arrives as disappointing usage numbers, followed by an explanation that blames the users.

Then there’s what I call the blank box problem. The traditional Salesforce screen, for all its faults, told you what to do. It had fields with labels, a path to follow, a button at the top. You didn’t need to know what to ask, because the screen asked you. A chat box does the opposite. It sits there, empty, cursor blinking, and asks you to already know. For a confident user that’s freedom. For someone who has never used AI, it’s the most intimidating screen ever put in front of them, and it’s being sold to their leaders as the easy option.

And prompting is only the surface. Even when people learn to type a reasonable question, the real skill is judging the answer. With AI, users stop being the authors of their records and become their editors. They’re handed a summary, a draft or a recommendation and asked to decide whether it’s right. Editing takes more understanding than writing, not less. You have to know what a good answer looks like before you can spot a confident wrong one. That’s why I keep saying conversational tools need more training, not less. The skill hasn’t disappeared. It has moved from knowing where to click to knowing what to ask and whether to believe what comes back.

Now put the two assumptions together. The user in the car park with one bar of signal from earlier is very often the same person who has never touched AI. The field engineer, the rep on the road, the service technician in a plant room. They’re the furthest from the connection and the furthest from the familiarity. They are twice as far from the boardroom as the plan assumes, and they’re the people closest to your customers.

None of them will raise their hand to tell you. That’s not how this works. Which leaves one question that no AI strategy I’ve seen has answered: how do the people making these decisions ever get to meet them?

Part 3 · Meet the user who actually exists

Now, to finalise this essay, I’ll start by repeating what I said at the beginning, because it matters. I’m not against AI. It’s here, it’s in Salesforce, and some of it will make working days genuinely better. What I’m against is building for a person who doesn’t exist.

That person is in almost every AI plan I’ve seen. They sit at a desk with a perfect connection, they have time to think, and they already know how to talk to AI. The people they’re meant to represent are somewhere else entirely: in a car park with one bar of signal, in a basement plant room, on a mountain road, or in a training room looking at the table because they’ve never used ChatGPT and don’t want anyone to know.

After more than ten years of training people to use Salesforce, I’ve noticed that almost every rollout that struggled was designed for the first person and delivered to the second. So here is what I would do about it. Some of it will sound extreme. I’d argue that’s only because we’ve become used to making these decisions from a very long way away.

First: no AI sign-off until a decision-maker has spent a day in the field. Not a visit, a day. Not the version where the regional manager tidies the van and picks the best rep, but an ordinary working day with the people who will be expected to use the tool. Watch where the signal drops. Count how long there really is between jobs. Notice what’s on the screen when a customer is standing right there. And ask, without any judgement at all, whether they have ever used AI. If that sounds like too much to ask of senior leaders, consider this. You wouldn’t approve a new building without seeing the site. An AI rollout changes how hundreds or thousands of people do their jobs, and it’s routinely approved from a slide.

Second: ask two questions before every rollout, and refuse to proceed without honest answers. Where will the people using this be standing when they need it, and does it work there? And have the people using this ever used AI at all? Those questions take five minutes. In most organisations, nobody in the room can answer either of them. That isn’t a reason to skip them. It’s the clearest sign the plan isn’t ready.

Third: explain AI before you train the tool. For people who have never used AI, the first conversation can’t be about features, because fear doesn’t respond to a feature list. Start with what AI actually is and what it isn’t. What it’s good at, what it gets wrong, and why their judgement still matters more than its confidence. Talk honestly about what it means for their job, because they’re already wondering, and silence on that point will be filled with the worst headline they’ve read. Then give them somewhere safe to try it on something that doesn’t matter, before it’s put in front of a customer. Confidence is built by trying something and seeing it work, not by watching someone else do it.

Fourth: design for the hardest day, not the keynote. A tool that works in the van will work at the desk, but not the other way round. Decide, role by role, what has to work with no signal and what people should do when it doesn’t. Then train that fallback as seriously as you train the shiny part. And when a field team’s records look thin, ask where they were standing before you ask why they didn’t bother. A dead zone and a lazy rep look identical in a dashboard. They look nothing alike in real life.

Fifth: use the people who have met both users. In most organisations there’s a small group who sit in the boardroom for the kick-off and in the van for the rollout. Trainers are one of them. We’re in the room when the strategy is presented, and a few weeks later we’re in a canteen or a depot with the people it was written for. We hear what people won’t tell their managers, because we’re not their managers. We see the hands that go up a beat late. Most organisations bring trainers in at the end, to deliver the plan. Bring them in at the start, while the plan can still change. They may be the only people in your organisation who have met the user who actually exists.

None of this slows AI down. It does the opposite. The fastest way to waste an AI investment is to switch it on for people who can’t reach it, don’t understand it and won’t tell you either of those things. The slowest part of any rollout has never been the technology. It’s the months spent explaining disappointing usage, blaming the people at the end of the chain and buying another tool to fix what was really a gap in understanding.

So let’s go back to where we started. Somewhere in your AI plan there is a user. They’re connected, confident, unhurried and fluent. They live on a slide, and they’ve never had a bad signal or a moment of doubt in their life.

And somewhere else, there’s a rep in a car park. One bar of signal. A customer in eight minutes. In front of her is a blank box she’s been told is the future. To use it, she has to remember the right way to ask the question. She has to wait for the answer to come back over a connection that keeps dropping. Then she has to read what comes back and decide whether it’s right: whether the figures match what she knows about this store, and whether she can repeat it to the store manager without being caught out.

That might well take longer than the eight minutes she has. So she does what people always do when the system takes longer than the job allows. She walks in without it.

Both of them are in your plan. Only one of them is real.

The question is which one your board has met.