Why the Hardest-Working Sales Teams Are Losing: The Willy Loman Problem

Death of a Salesman is a 75-year-old play. It is also the most accurate description of what AI is doing to modern sales that I have ever seen.

The most useful business case study I’ve encountered this year was written in 1949.

I watched it performed on stage last week.

A man walks out carrying two suitcases. He’s not acting exhausted. He is exhausted. Right there in front of you. And for the next three hours, you watch him disintegrate in slow motion.

The play is Arthur Miller’s Death of a Salesman, performed on a sold-out stage in Istanbul, with Halit Ergenç as Willy Loman.

One of the great plays. Pulitzer Prize winner. Taught in every major literature department. Often called the greatest American play ever written.

Watching it live hits different.

Miller’s language. The staging. The way grief moves through a family in real time. It reaches somewhere most business books never get close to.

I sat with that for a day.

Then I couldn’t stop thinking about your sales team.

I know what that sounds like.

But hear me out.

This isn’t a metaphor. Not a dramatic flourish.

It’s a case study in what happens when the rules of a market shift and the operator inside it refuses to update their mental model.

And the first thing Miller makes painfully clear is this: Willy Loman’s problem was not that he didn’t work hard enough.

---

The Problem Was Never the Work Ethic

Willy Loman worked hard.

His whole life on the road, city to city, state to state, putting in the hours, making the calls, shaking the hands.

Arthur Miller is precise about this. Willy isn’t lazy. He isn’t incompetent. He genuinely believed in what he was doing.

The problem was what he believed in.

Willy’s operating thesis was this: If you’re likable enough, well-connected enough, and work hard enough, the money follows.

Personality is the moat. Relationships are the asset. The salesman who everyone knows wins.

And to be clear, it was a real model.

It worked for a generation.

And then, quietly, it stopped.

The corporate, systematized model replaced the door-to-door era. What used to be relationship-driven became process-driven. What used to require charm now required systems.

Willy didn’t fail because the world got harder.

He failed because the world got different, and he kept running the same play.

The tragedy isn’t that he couldn’t win.

The tragedy is that he never updated his definition of winning.

But Miller doesn’t just show us the man trapped in the old model.

He also places another man right next to him. Quieter, less charismatic, easier to overlook who represents the model Willy refuses to see.

---

Enter Charley

There’s a character in the play most people forget.

Charley, Willy’s neighbor.

Steady. Quiet. Un-dramatic.

Throughout the play, Charley offers Willy a job. Multiple times. Willy refuses every time, because accepting would mean admitting that Charley figured something out that Willy didn’t.

Charley doesn’t sell on personality. He doesn’t have Willy’s charisma. He doesn’t need the room to love him.

He just runs a tight operation.

His son Bernard is the same kind of signal. Bernard is the kid Willy’s sons used to mock for studying too much instead of playing football. Later, Bernard grows up to argue cases before the U.S. Supreme Court.

That detail matters.

The person everyone underestimated was quietly building the kind of advantage that compounds.

Charley is not the opposite of Willy because he works less.

He is the opposite of Willy because he operates from a different model.

Willy believes the world rewards being liked.

Charley understands the world rewards being useful, prepared, and structurally sound.

That distinction is everywhere right now.

In every industry, there’s a Charley.

You probably don’t know who they are yet.

They’re not on stage.

They’re not posting about their sales culture on LinkedIn.

They’re quietly rebuilding their revenue architecture around AI. And in 18 months, their cost per acquisition is going to be structurally lower than yours, and you won’t be able to figure out why your pipeline feels increasingly uphill.

This is where the play stops feeling like literature and starts feeling like a board meeting.

Because the question every company is now facing is not “Should we use AI?”

It is: are we using AI to preserve an old model, or to build a better one?

The difference between those two choices is the difference between Klarna and Salesforce.

---

What Klarna Got Right. And What Klarna Got Wrong.

Let’s talk about the most cited AI case study of the last two years.

In early 2024, Klarna announced that its AI assistant, built in partnership with OpenAI, had handled 2.3 million customer service conversations in its first month alone.

• Response time dropped from 11 minutes to under 2 minutes.

• Customer satisfaction scores initially matched human agents.

• Repeat inquiries fell 25%.

The company had deployed the system for somewhere between $2 million and $3 million, projecting a $40 million profit improvement for 2024.

The board smiled.

The market applauded.

Sebastian Siemiatkowski talked about the future.

Then something interesting happened.

Customer satisfaction scores dropped as edge cases, emotionally charged interactions, and multi-step problem resolution overwhelmed AI trained to handle routine queries.

CEO Sebastian Siemiatkowski publicly admitted that the AI-driven transition negatively affected service and product quality. Klarna is now rehiring human staff.

This is not a story about AI failing.

This is a story about binary thinking.

Klarna didn’t build an AI-native system.

They performed a substitution.

They replaced humans with AI as if it were a one-for-one swap. Same jobs, different executor.

What they missed was that the model itself needed to change.

AI doesn’t replace people. It replaces the way work is structured.

And this is exactly where Miller’s family drama becomes useful again.

Because Willy is not the only character who refuses reality.

His younger son, Happy, sees the same collapse happening in front of him and keeps calling it progress.

Happy Loman is perpetually upbeat, perpetually deluded. He sees all the same signals his older brother Biff sees. The failure, the disconnect, the crumbling but he reframes everything.

Things are going great.

Dad’s fine.

We’re going to be fine.

When founders hear “AI can transform your sales function” and respond with “Let’s cut the team and plug in a bot” that’s Happy Loman energy.

It looks decisive.

It’s actually avoidance.

Avoidance of the harder question: how does the entire system need to be rearchitected?

The lesson from Klarna isn’t “AI doesn’t work”.

It’s this: Automation without systems redesign just creates faster failure.

So if Klarna shows the danger of substituting the executor without redesigning the system, the next question is obvious:

What does proper redesign look like?

That’s where Salesforce becomes interesting.

---

What Salesforce Actually Built

Now look at the other end of the spectrum.

Salesforce spent years being mocked for being behind on AI.

Then they built Agentforce and went all-in on what they called the “agentic enterprise.”

AI agents now successfully resolve 85% of Salesforce’s customer service inquiries and qualify its own sales leads 40% faster than before the advent of AI.

Overall, Marc Benioff says these AI agents are so effective that they are now doing 30% to 50% of all the work within Salesforce itself.

See: Fortune.

The numbers on the product side followed.

Agentforce ARR hit $800 million, up 169% year-on-year.

The company closed 29,000 Agentforce deals, up 50% quarter-on-quarter in Q4 alone.

Internally, Salesforce is reporting $100 million in annualized cost savings, with over 3,200 opportunities influenced by AI agents.

But here’s the part that usually gets buried: As a result, Benioff has announced that Salesforce won’t be hiring any additional software engineers, customer service agents, or lawyers.

But the company is hiring salespeople and customer success employees.

Read that again.

They’re not replacing everyone.

They’re replacing specific functions and redeploying humans to the work that compounds on human judgment: relationship-building, complex problem-solving, strategic expansion.

That is the Charley model.

Systematic. Patient. Structured.

Not dramatic. Not loud.

Just structurally better over time.

And the small business case is just as interesting.

Safari365, an Africa-based tour operator with just 35 employees, achieved 62% case resolution through Agentforce after completing a data cleanup that took longer than the agent deployment itself.

The data cleanup taking longer than the deployment.

That detail carries everything.

The limiting factor wasn’t AI.

It was the quality of the underlying system.

This keeps coming up.

The companies getting the best results from AI sales automation aren’t the ones with the biggest budgets.

They’re the ones who did the unsexy work of cleaning their data, mapping their process, and thinking clearly about where human judgment is actually required.

Bernard studied.

The jocks mocked him for it.

Then the world changed, and studying was the advantage.

And once you see that, another part of the play starts to look different.

Because Willy’s problem wasn’t only that he had the wrong model.

It was that he kept using the wrong signals to convince himself the model was still working.

---

Willy’s Hallucinations Were Metrics

There’s a recurring motif in the play that takes time to register.

Willy hallucinates.

He holds full conversations with his dead brother Ben, replays memories as if they’re happening now, and exists simultaneously in the present and in a golden version of the past where everything was working.

His family doesn’t call it hallucination.

They call it “Dad’s episodes”.

Most sales teams have their own version of this.

They’re called activity metrics.

Calls made.

Emails sent.

Meetings booked.

Pipeline coverage ratio.

These numbers feel real and important. They fill dashboards. They get presented in QBRs. People get promoted for hitting them.

But organizations using AI analysis reach 96% forecasting accuracy, while human judgment alone achieves only 66%.

That 30-point gap isn’t a rounding error.

That’s the gap between what you think your pipeline is doing and what it’s actually doing.

Willy’s tragedy wasn’t simply that he was working in the wrong direction.

It was that he had built an entire internal narrative around signals that felt true but weren’t.

By the time reality forced its way through, there was no runway left.

When you’re measuring sales team performance by effort rather than conversion velocity and CAC payback period, you’re living in Willy’s house.

And Willy’s house matters.

Because Miller makes the economics of the old model brutally literal.

Willy doesn’t just lose psychologically.

He loses financially, structurally, and too late.

---

IN COLLABORATION WITH CUEY

One AI Hallucination Can Cost You a Client

AI is great for fast drafts.

But for client work, Cuey cross-checks ChatGPT, Claude & Gemini in one tab so you can catch hallucinations before they burn you.

Built for founders and operators who can't afford to be wrong.

Claim 2 months of Pro on us with code NEXTBIG (in account settings).

---

The Mortgage Nobody Talks About

There’s a quietly devastating image in the play.

Willy has spent 35 years paying off the mortgage on his house.

The house is finally, finally paid off.

And in the same week, he dies.

His wife Linda says: “We’re free and clear”.

Nobody is there to hear it.

Miller uses this to show that Willy was always a beat behind. He spent his whole life paying for a life he never got to live.

The modern equivalent is the sales team cost structure that makes sense at $1M ARR and breaks at $10M.

Here’s what the math actually looks like at scale: A fully loaded enterprise AE in most markets costs $150K–$200K annually when you factor in base, variable, benefits, and management overhead.

Five of them is $750K–$1M per year.

That number doesn’t scale sub-linearly.

It scales linearly or worse, because more reps means more management, more enablement, more churn, more ramp time.

AI-driven outbound and lead qualification, built properly, handles the top-of-funnel work at a fraction of that number.

And unlike headcount, it doesn’t require ramp time, doesn’t churn, doesn’t have bad weeks, and gets better with data.

McKinsey data shows a 45% increase in sales productivity and a 12% reduction in sales costs from comprehensive sales automation implementations.

Bain & Company reports that early AI deployments in sales have boosted win rates by 30% or more.

But here’s the part that really matters: Those numbers compound.

The company running AI-native sales at year one isn’t just more efficient at year one.

They’re accumulating data, refining models, and expanding into markets you haven’t reached yet because they could afford the coverage.

You’re still paying the mortgage.

They own the house free and clear.

At this point, the play gives us more than one warning.

Willy shows what happens when you double down on the past.

Happy shows what happens when you reframe danger as optimism.

Biff shows what happens when you see the truth but can’t act on it.

Linda shows what happens when loyalty protects a broken system.

Together, they map almost every founder response to AI right now.

---

The Four Loman Traps

In the play, every major character represents a different response to a system that’s shifting beneath them.

I’ve seen all four of these responses in founders over the last two years.

1. The Willy Trap

Doubling down on what worked before.

More reps. More calls. More hustle. More relationship meetings.

“Our product requires a human touch”.

It probably does for some of the sales motion.

But most of the motion?

It requires speed and personalization at scale, which is exactly what AI does better.

The Willy Trap feels noble because it looks like commitment.

But often, it is just refusal dressed up as discipline.

2. The Happy Trap

Seeing the data and explaining it away.

“Our market is different”.

“Our customers expect a human”.

“We tried a chatbot in 2019 and it didn’t work”.

Happy Loman will tell you everything is fine right up until it isn’t.

The founders running this trap are usually articulate and confident.

That’s what makes it dangerous.

The Happy Trap feels strategic because it sounds calm.

But often, it is just denial with better vocabulary.

3. The Biff Trap

Biff is Willy’s older son.

He is the character who sees clearly, maybe the only one who does but freezes.

He knows the old story is broken.

He just doesn’t know what to build instead.

I see this in founders who understand the shift is real, understand they need to change, but get stuck in the enormity of the redesign and defer it quarter after quarter.

They are not in denial.

They are stuck between diagnosis and action.

The Biff Trap feels honest because the diagnosis is accurate.

But diagnosis without redesign still loses to someone already building.

4. The Linda Trap

Linda is Willy’s wife and the most sympathetic character in the play.

She loves Willy deeply and keeps everything running.

She’s also the one who, by protecting Willy from reality, extends his delusion the longest.

In a business context, this is the loyal sales leader.

High integrity. Real experience. Genuinely invested in the team.

But they keep defending the existing model because abandoning it would mean admitting that the last several years of strategy were pointing in the wrong direction.

The Linda Trap feels humane because it protects people.

But sometimes protecting the old structure only delays the harder transition everyone will eventually have to face.

None of these people are stupid.

None of them are bad.

They’re all rationally responding to the information available to them.

That’s exactly what makes it a tragedy.

And it’s also why the real danger isn’t the first-order effect.

The real danger is what happens after the first advantage compounds.

---

The Second-Order Effect Nobody Is Modeling

Here’s the thing about compounding advantages: They’re invisible until they’re insurmountable.

Companies that utilize AI see 83% revenue growth compared to 66% without AI.

That 17-point gap doesn’t sound catastrophic in year one.

In year three, it’s an unbridgeable chasm.

The mechanism is straightforward:

Your AI-native competitor acquires customers cheaper.

They reinvest the margin into more experiments.

They expand into adjacent markets.

They build a larger data set.

Their model gets better.

Their CAC drops further.

They can afford to undercut you on price and outspend you on distribution simultaneously.

This isn’t a hypothetical.

It’s the structure that played out in e-commerce, in fintech, in content, and in every category where a compounding technology advantage had a three-year head start.

Research shows sales reps currently waste about 40% of their time on leads that go nowhere, costing companies between $10,000 and $30,000 per sales rep every year.

Multiply that by your sales team size.

That’s the floor of what AI-native qualification saves before you even model the conversion improvements.

This is why the answer can’t be “let’s try a bot”.

That is too small.

The question is not whether AI can automate a task.

The question is whether your revenue system is being redesigned around a new cost structure, a new speed of learning, and a new division of labor between humans and machines.

So the practical next step is not a giant transformation deck.

It is a sharper diagnosis.

---

What To Actually Do With This

This does not need to become a transformation roadmap.

It does not need to become a 12-step framework.

Start with three questions that cut to the right diagnosis.

1. Map your sales motion and ask: where does human judgment actually change outcomes?

Not where it feels good to have humans.

Not where customers say they prefer it.

Where does the outcome, conversion rate, deal size, retention measurably improve with human involvement?

Be honest.

The answer is probably less of the motion than you think, and it’s almost certainly skewed toward late-stage and complex enterprise.

2. What’s your competitor’s cost structure going to look like in 18 months?

You know who’s moving fastest in your space.

What happens to your CAC disadvantage if they build what Salesforce built?

Model it.

The number will be uncomfortable.

Good.

That’s the right signal.

3. Are you redesigning the system, or substituting the executor?

Klarna substituted.

They replaced people with AI as if it were a one-for-one swap.

Salesforce redesigned.

They rethought which work required humans and restructured everything around that.

The difference in outcomes was not small.

The founders who win this next stage aren’t the ones who move fastest to cut headcount.

They’re the ones who think most clearly about what the system should look like and then build that system, with humans and AI doing what each actually does best.

And this brings us back to the stage.

Because Miller’s final scene is not about whether Willy worked hard.

It is about what all that work amounted to once the world had moved on.

---

Back to the Stage

The play ends with a nearly empty funeral.

Willy was sure hundreds would come.

All the people he’d helped. All the connections he’d built. All the goodwill he’d accumulated across forty years on the road.

But almost nobody is there.

His wife stands at the graveside and says the mortgage is finally paid off.

Free and clear.

Charley delivers the eulogy. Quietly. Without fanfare.

“A salesman is got to dream, boy. It comes with the territory”.

Miller meant it as compassion.

I read it as a warning.

The dream is not the problem.

The problem is when the dream becomes a substitute for updating your model.

When optimism about what used to work replaces honest accounting of what’s working now.

The death of the salesman, as a business model, is already in motion.

The question isn’t whether it happens.

It’s whether you’re on the right side of it when it does.

Willy Loman didn’t have advance notice.

You do.