The Silent Infrastructure

The idea was always seductive: take the $500B US home improvement industry and bring it online.

Between 2013 and 2019, a wave of startups set out to revolutionize interior design in the U.S.

Companies like Modsy, Havenly, Laurel & Wolf, Home Polish, and others were hailed as the next big thing in consumer design-tech. They promised to democratize interior design by combining visual rendering tools, data-driven personalization, and embedded commerce. They raised hundreds of millions of dollars collectively, secured deals with Fortune 100 retailers like Home Depot, Crate & Barrel and Bed, Bath & Beyond, and landed features in every publication from TechCrunch to The New York Times.

We were one of them. In fact, we closed a pilot with a large Fortune 100 retailer, designing thousands of dorm rooms with our own custom-built design engine. We were invited to present to their board. Our second pilot was priced at hundreds of thousands of dollars.

VCs flooded in. So did the press. “Pinterest meets Shopify,” they called it.

It felt like all of us design-tech startups were on the cusp of transforming how America furnished its homes.

We were wrong.

The Dream That Died

By 2022, nearly all of us had either shut down, been acqui-hired, or pivoted into niche services. Modsy, for example, went from raising over $70M to abruptly laying off its entire staff and halting consumer operations.

The crash was quiet, but total.

What went wrong?

At the time, the prevailing analysis was surface-level: Design isn’t scalable. AI isn’t good enough. Consumers aren’t ready.

But underneath that was a deeper misalignment — one that reveals not just why those startups failed, but what the next generation of startups must do differently.

The Value Was Real. The Model Wasn’t.

By most traditional consumer metrics, design-tech startups were doing things right. Customer satisfaction scores were high. Session lengths and engagement were strong. Retention, for those who bought, was promising. Even the products — shoppable, interactive, visually rich — were universally loved.

But the business model couldn’t support it.

Firstly, here’s what we startups learned the hard way: people love beautiful rooms, but they don’t value design enough to pay for it.

The reason? 92% of U.S. home renovators do not hire an interior designer. (2025 U.S. Houzz & Home Study: Renovation Trends, pg. 23)

They don’t view design as a service. They see it as something they can do themselves — or ignore altogether.

Good design is invisible.
Turns out, monetizing the invisible is impossible.

Secondly, the design itself became a source of value leakage. Users browsed 3D designs and purchased similar products elsewhere. Startups were doing the hard work of inspiration — while someone else captured the transaction.

Thirdly, customer acquisition costs (CAC) were high — because design isn’t something people shop for often. Most consumers redesign a space once every 5–10 years. Retention was negligible. Word-of-mouth didn’t scale.

Finally, rendering costs were high. Design labor — even when outsourced — was manual and slow. To break even, startups needed either (1) massive repeat usage, or (2) high-margin product purchases.

They had neither.

So the model got stretched in both directions:

• To consumers, the service felt like magic — but still not quite worth paying for.
• To investors, the vision sounded like SaaS — but behaved like a services business.

Some startups tried pivoting to private-label furniture. That meant warehousing, logistics, returns, and thin margins. They were no longer SaaS companies. They were trying to become Wayfair, but with less capital and more overhead.

The Buyer No One Could Find

One of the most valuable consumer segments in the U.S. home improvement economy is also one of the hardest to reach: movers.

Every year, roughly 5 to 6 million homes are sold in the U.S. — out of a total housing stock of over 140 million units. That means just 3–4% of adult Americans become new homeowners annually.

But these 4% punch far above their weight.

Movers account for over 50% of all home improvement and furnishing spend. That includes furniture, flooring, lighting, paint, window treatments, appliances, and decor. (Harvard’s Joint Center for Housing Studies, pg. 6)

Why? Because they’re making decisions they can’t defer. New layout → new furniture. Empty walls → new inspiration. Tight timelines → high intent.

From a retail standpoint, movers are dream customers: high purchase intent, high basket size, low price sensitivity (relative to the transaction they just made), urgency built in.

But they’re almost impossible to target directly.

Retailers spend billions trying to infer mover intent from change-of-address forms, credit card activity, email sign-ups, zip code lookups and ad click behavior.

These signals are noisy, delayed, and often arrive after the big purchases are already made.

By the time a mover walks into a store or gets a “Welcome Home” email, they’ve usually bought what they need — or made trade-offs because no one showed up at the right time.

Now here’s the twist: there’s one industry that sees movers first, engages them deeply, and captures their attention during their most formative purchase window.

Real estate.

Real estate platforms — MLSs, brokers, portals, and agents — don’t just engage with movers. They help create movers.

They know when a user is thinking about moving. They know which property they’re considering. They know the layout, condition, and style of the home. They own the surface where these buyers spend hours analyzing room by room.

In an internet that obsesses over intent signals, real estate has the clearest one of all. The buyer isn’t just visiting. They’re moving in.

If you’re a brand that sells into the home — furniture, appliances, paint, decor — you couldn’t ask for a better opportunity.

The only problem? That opportunity has never been wired into startups’ distribution strategy.

Why Real Estate Should Care (And Already Does)

The good news? Real estate leaders are beginning to wake up to the opportunity.

Portals like Realtor.com and Redfin have already rolled out early experiments in this direction — visualization tools that allow users to redesign, restage, or renovate listings on their platforms. From AI-enhanced staging interfaces to renovation preview tools, the push toward interactive listing media has clearly begun.

But these are still just early signs. The engagement is real, but the infrastructure behind it remains fragmented.

And most importantly, these tools don’t yet plug into monetization at scale. They’re cool features — not yet commerce layers.

What real estate has today is a preview of potential. What it needs is a system.

Because beyond engagement lies something far more valuable: the ability to turn attention into intent, and intent into transaction.

Why Everyone Wants This — But No One Can Afford It

Generative AI is the key here. But not the way it is being used by national portals today. Let’s assume you do want to turn every listing into an interactive, AI-powered retail canvas.

Here’s what that looks like at scale:

• U.S. real estate portals (Zillow, Redfin, Realtor.com, etc.) see over 8 billion sessions a year. (Similarweb)
• Let’s say just 50% of those engage with interactive design tools.
• Let’s assume 10 redesign tweaks per session are done using AI.
• That’s 40 billion generative AI API calls per year.
• Each image editing API call costs around 3–5 cents using a standard A100 GPU cloud setup. For the purpose of this guesstimate, we’re not even factoring in APIs like OpenAI’s image model — because that can be 5x more expensive per image, making the economics even more prohibitive.

That’s an over $1.5 billion annual compute bill, just to make this possible.

This is the bottleneck that will hold everyone back — from portals and brokers to brands and buyers.

Not to mention: generative AI still can’t do the one thing that matters most.

It can’t insert real products into real homes. It can’t handle photos from bad angles or dim lighting. It can’t match furniture SKUs with 3D-accurate scale. It can’t show your couch in your new home. It can’t design using real retail catalogs — at volume.

Even the best AI tools today are only good for fantasy. They don’t sell products. They sell “vibes.”

AI can’t show real products in real rooms — and that’s basically everything.— Business of Home

The AI Stack This Actually Requires

Most generative AI demos look magical — until you try to scale them or ask them to behave like infrastructure.

In reality, for this vision to work, AI must:

• Generate design imagery from any listing photo, even low-quality or cluttered ones.
• Incorporate real products from retailer catalogs — not just AI hallucinations.
• Allow users to include their own existing furniture in generated scenes.
• Create dimensionally accurate layouts with correct scale and geometry.
• Maintain hundreds of stylistic variations tailored to user preferences.
• Do all this at zero incremental cost per image, with no human input.
• And finally — integrate with listing workflows, not just sit on a separate app.

Until now, no platform has cracked this.

The System We’ve Quietly Been Building — Together

This time, “we” means more than just a startup.

It’s the unlikely but powerful pairing of an AI company and one of the largest real estate organizations in the United States — coming together not just to imagine a new future, but to build the rails that make it possible.

We didn’t get here overnight.

We started with real estate agents — serving over 15,000 paying professionals through our virtual staging service. Then we launched a consumer-facing AI staging tool that went viral, used by over 2 million users worldwide. That gave us clarity on what users actually wanted, and what AI simply couldn’t do — yet.

But it wasn’t enough to crack the AI. We needed to rewire the entire way listing media flows through the ecosystem — from the moment a photo is uploaded to how it’s enhanced, published, and monetized. That’s where our partnership with the MLS came in. Together, we’ve re-architected the backend of listing distribution so that interactive, shoppable media doesn’t just bolt on — it gets embedded at the protocol level.

This isn’t just a product or a feature. It’s a foundational shift in how real estate media becomes retail infrastructure.

And it only works because the AI finally does. Here’s what it took to get there:

And we solved something else along the way: the cost problem.

Most generative AI systems are expensive because they create visuals in real time — burning compute every time a user clicks. We flipped that model. In our system, a listing photo is processed once through a stack of over 10 specialized generative models — extracting layout, depth, object masks, semantic context, and more.

After that, a million users can interact with that photo — customizing styles, swapping products, tailoring budgets — with zero additional compute cost.

It’s not just scalable. It’s sustainable.

What Retail Gets in Return

For retailers — especially those in furniture, flooring, paint, and décor — this is a moment of channel transformation.

They’ve spent years trying to catch movers post-sale through address change data, credit card signals, or geo-fencing. But that’s like trying to close the barn door after the furniture has shipped.

Now, for the first time, they can be present during the decision — not just after.

Years ago, when we pitched this vision to the board of a major big-box retailer, the response was an emphatic yes. They greenlit a six-figure pilot to bring commerce directly into dorm layouts and design surfaces for student housing.

Later, the leadership of a national home improvement chain told us what they really wanted wasn’t more ad impressions — it was the ability to show their products inside the home photos people were already obsessing over.

Retail doesn’t need another feed. It needs infrastructure — and this is it.

The Dream Rebuilt

By now, you’ve heard your share of AI-powered pitches.

You’ve seen products demo beautifully but deploy poorly. You’ve seen tech promise scale but forget economics. And you’ve seen startups chase inspiration when they should have chased infrastructure.

But this time is different. Not just because AI has evolved. But because we’ve grounded the innovation in three things the first wave overlooked: Distribution. Timing. And trust.

This isn’t about convincing users to behave differently. This is about meeting them in the exact moment they already care — when they’re staring at the kitchen of a house they might buy — wondering if their couch will fit — wondering what’s possible.

It’s not a guess. It’s not an ad. It’s real.

So here’s the ask — for each of you reading this:

If you passed on Modsy or Havenly, you were probably right. But if you pass on this — you’ll be wrong. For very different reasons.

Because this time, the room isn’t just designed. It’s connected. It’s shoppable. And it’s right there — inside the listing.

P.S. This series is called Not in a Nutshell for a reason. Some truths deserve more than a scroll. Essay two, The Indictment, is here on notinanutshell.com. Judge harshly.