The Demo Is Not the Product

A working demo used to be expensive, and that was the entire point of it.

When it took a team six weeks to get something clickable in front of an executive — six weeks of somebody's actual life, spent wiring screens together in InVision and arguing about which three flows mattered enough to fake — the demo was a credential. Somebody had thought it through, the hard parts had been touched, and the estimate behind it meant something, because getting that far had cost something real.

That's over. A working demo now costs a week and a decent prompt, and sometimes it costs a weekend.

Your executives are still reading it as a credential.

So what exactly did that demo prove? That's the strategic problem, and it isn't really an AI problem at all. It's a pricing problem. The demo isn't a smaller version of the product; it's a different product entirely, and your organization is pricing one with the other's numbers.

I call it the demo discount. Somebody shows a working thing on a Tuesday, the room feels the risk drop, and everything after that gets quietly marked down. Nobody writes the discount anywhere — it just sits inside the yes, unexamined, until somebody has to go build the thing.

Two economies, and only one of them got cheaper

The demo economy is generation: one person, one machine, the happy path, sample data, and nobody watching it overnight.

The product economy is everything else. It's integration with systems that predate you, permissions nobody has documented since the last reorg, real data that's always messier than the sample, and every edge case your prompt never saw. Then there's a whole category that didn't exist three years ago, which is how on earth do you test output that comes back different every time you ask?

AI collapsed the first economy to almost nothing, and it left the second one exactly where it was. In a few places it made the second one heavier.

Watch the pattern move across industries and it keeps the same shape. In a bank, the demo skips model risk review — and the review is the schedule. In government it skips procurement and accessibility, which are the actual road, and in a scale-up it skips the data plumbing, which is where the quarter quietly goes.

The demo skips the expensive part, and then gets used to price it.

A demo tells you the thing is possible. It tells you nothing about what it costs to be real.

What one week bought me

Years ago I built a recommendation system, because my CTO wanted one and I wanted to be the person who handed it to him, and I had something demoing inside a week.

It demoed beautifully, and that's the part I'd like back.

Building it for real took three to four months. Then came a year of minding it — steady, unglamorous attention that never appeared on a roadmap, never got staffed, and never once came up in a planning conversation as actual work. It was a lemon — hard to run, hard to change, hard to hand to anybody else.

Eventually I added measurement, because I wanted to watch the thing working, and what I got instead was the discovery that almost nobody was using it.

Nobody had told me, either. There was no meeting and no confrontation, no moment where somebody sat me down and broke the news gently. I built the instrumentation that should have shipped with the product in the first place, and then I sat there and watched a flat line.

Here's what I'd been telling myself for sixteen months: the demo landed, so the idea was sound.

The demo proved the thing could work, and nothing in that entire year proved that it did. I never separated those two sentences, and the gap between them cost my team the better part of a year. That one is on me.

One week to demo, four months to real, twelve months of minding it — and I never wrote those three numbers next to each other until long after they could have helped anybody. Yikes.

The distance to real

Write them next to each other. That's the entire discipline, and it's less work than the last meeting you sat through.

The distance to real is the gap between the day something demos and the day it runs in production with actual users on it, and you measure it as a multiple. A week to demo, sixteen weeks to production, and your distance is 16x.

The multiple itself isn't the problem, because a 16x distance is completely fine if the room knew it was 16x when it said yes. Mine didn't know, and mine priced a four-month build plus a year of quiet attention off a Tuesday afternoon that happened to go well.

A demo you can price is an asset. A demo you can't price is a liability in a nice suit.

So go and pull the number. Take the last three initiatives your organization approved after seeing something working. For each one, write down the date of the demo and the date it reached production. If it never got there, write today's date and let that sit for a second.

Now you have three multiples, which is your organization's real conversion rate from idea to running software, and I would bet a considerable amount that nobody has ever said it out loud in a portfolio review. Why would they? Nobody has ever been asked for it.

Three is enough. You're not building a model here, you're ending an argument.

Why this one is yours

You might think this belongs to engineering, and it's true that they know exactly what the demo skipped, because they're the ones handed the remainder and most of them can name every piece of it before the meeting ends.

But they're describing work and you're pricing a portfolio. Those are different conversations in different rooms, and only one of them is the room where the yes gets given.

Your exec sees the demo, your team sees the distance, and you're the only person in the building sitting where both are visible. That makes the translation yours whether you want it or not.

Annie Duke has a line I think about constantly: you can't judge a decision by how it turned out, because good decisions lose all the time and bad ones get lucky. The demo is an outcome; the distance is decision quality. Confusing the two is how a room full of genuinely smart people ends up funding the same lemon twice and calling it bad luck. Sound like any review you've sat in?

None of this is an argument for slowing down, by the way. Cheap demos are good, so build more of them — build ten this quarter, because the whole advantage of this moment is that proving an idea barely costs anything now.

Just stop letting a cheap demo buy an expensive yes.

Do this before your next portfolio review

Go find your last three demo-driven yeses, put them on one page in two columns — demo date, production date — and work out each multiple in the margin.

Then take that page into the review, not as an accusation but as arithmetic.

What's your number, and who else in the building already knew it before you asked?