The Tandem AI bet: how office leasing will finally become a true marketplace

Web2 upended small-ticket, recurring service providers: stock brokers, travel agents, etc. The yet undisrupted: large, infrequent transactions where human trust matters a lot. AI will change that.

Hey readers!

This is a long one (too long to fit in an email, in fact), but a topic that’s real close to my work today. I’ve written a fair bit about the future of the office market, and something I’ve been thinking a lot about over the past couple of years is how AI will change the industry.

This is focused on office leasing, but I think many of the takeaways are generally applicable to all professional service industries: lawyers, doctors, residential real estate agents, accountants, etc. Combined these industries are worth trillions of dollars, and I believe all of them will experience significant change as a result of LLMs. But maybe not in the ways you might imagine.

This is an exciting time, and I’m thrilled to share our learnings and approach (so far).

If you like it, please drop a comment or consider sharing!

All about scalability!

Learning from history: all the failed marketplaces that came before

In the last 20 years, the number of tech startups that have died trying to disrupt real estate brokerage is uncountable.

The story was clean, the task surmountable.  The job not so hard:

1. Find spaces.

2. Coordinate tours.

3. Collect feedback.

4. Find and share more spaces.

5. Negotiate a price.

6. Sign docs.  

And yet the current experience is terrible.  Buying a home, leasing an office -- months and months of work, countless emails, a huge headache. Archaic industries that operate no differently than they did 50 years ago. 

Tech could do this.  Build a marketplace, price properties algorithmically or create a bidding system no different than eBay. Facilitate self-touring with the property owner. Automate document origination and streamline signing processes.   This isn't complicated stuff, and the tech behind Uber and Airbnb - tech giants of yesteryear - is more than good enough to do it.

And… they all failed.  Zillow, Redfin, Opendoor got close, but ultimately retreated.  Big pockets, lots of investment, no success.  On the commercial side, nobody even made it to the 50 yard line. 

Of course, each failed for a confluence of reasons (I won’t talk about CoStar and the litigious behemoth of an anti-innovation machine it is), but the common denominator with all of them was trust. 

The question that all the smart, hard working, well funded people behind these tech startups failed to truly appreciate:

If a platform could offer the same process as the “traditional” way with objectively better outcomes, why might people still not flock to it?

The tech assumption, an assumption that underlies the failure of many ambitious startups, is that, assuming you can reach people (distribution), if the outcomes are better, people will adopt the new way of doing things.

It’s a logical assumption that completely ignores human psychology.

In big, infrequent, risky-seeming transactions, fear is real. When afraid, people trust people over machines.

Imagine you were on the market for a new home. You’re considering a traditional agent — someone recommended to you from a close friend — when one of these new tech companies comes to you and says they can offer you the following if you choose them:

1. You can buy a home in half the time.

2. You will get the guaranteed best price available for a home of this type on the market.

3. They will guarantee your money-back if anything goes wrong in the process (for example, if you discovered post-close that the original home inspection was faulty).

Would these sway you away from the agent?

For most people, the answer is no.  The reason is trust.

Here's why:

1. Big ($) decisions taken infrequently are scary.

   1. Buying a home, renting an apartment, leasing an office. They're scary because the costs of failure are high, and, because you don't do it often, you don't know what you don't know.

2. Fear is a potent emotion.

   1. In moments of fear, our brains turn inward, away from conscious, deliberate, objective thought. In these moments, we rely more on our instincts, our "gut".

3. In moments of fear, our biases control.

   1. Our instincts (the "thinking fast" parts of our brain) are developed on thousands of years of evolution. 

4. One of those biases is the bias to trust human beings above inanimate objects (and other animals), and particularly human beings that look, smell, talk, and act like we do.

   1. And yes, computers count as inanimate objects.

These reasons are the reasons why self-driving cars are so unpopular.  When the risk is high, people trust humans over machines, even if those machines are assured to be better in all measurable ways.

With each of these failed startups, the techies continued to fight, believing that if they could just improve the outcomes, offer a little more of a guarantee, eventually the people would take it.

Eventually, they might.  The trust/fear paradox of humans and technology is not a binary — because fear is not a binary. Humans trust computers to trade their stocks and buy their flight tickets, so it’s only reasonable to assume they will eventually trust computers to buy a home too.

So I’m not arguing that eventually better outcomes will win. But, “eventually” doesn’t cut it in startupland.

For the big ticket items, the infrequent decisions like buying a home, “eventually” is too far away to be worth fighting for today.  Emerging companies are resource constrained -- results are needed soon, not in 20 or 30 years when the generations might turn over.  In the meantime, betting on mass behavior change against a potent and evolutionary human emotion is foolish.

Where tech couldn’t do all of it, it couldn’t do any of it. Brokers don’t buy gizmos.

The lesson from all this is that any technological breakthrough for big ticket, risky, non-recurring services must augment or enable the human being, rather than replace.

Fully self-serve law firms, primary care doctor's offices, real estate brokers are just not going to happen.  For these things, people trust other people, and without that trust you don't have a customer.

So let’s assume the human will, for some non-negligible period of time, remain involved in these activities. How can tech help those humans do better?

The problem was, until now, all the tools created for these professionals were not disruptive.  Software to help a physician take better notes or a lawyer to more easily templatize documents -- these are certainly helpful to the service providers, and might lead to marginal increases in productivity, but they don't fundamentally change how they do their jobs.

The issue with tools that only make marginal improvements: they’re fully at the will of the buyer. They do not enjoy the forceful effects of capitalism in its glory; instead they exist at the whim of marketing and ads and enterprise sales agreements etc., etc.

When it comes to real estate agents and brokers, the challenge of non-disruptive tech is even greater. Even if they help out marginally, the cost is born entirely on the individual because most agents and brokers either work for themselves or are commission-paid 1099 contractors.

This independent contractor arrangement is a huge reason any tech introduced to help make the real estate buying or leasing process smoother has failed to gain real traction: if it’s not enough to fully disrupt the brokerage model, the onus falls on the individual broker to buy the tech and put it to use. And when every dollar of a deal is a dollar in your pocket, most brokers won’t pony up.

This explains why the whole office brokerage process, take for the sheer existence of email as a way of transmitting messages, is essentially the same as it was 50 years ago.

AI, acting as an autonomous co-pilot, is different. It has the potential to truly disrupt and flip the whole model on its head.

To explain how, I'll share how we are putting AI to work at Tandem.

Meet Kristen and her superpowers

Kristen (my now co-founder) was the best office broker in the city of San Francisco. When I met her 14 months ago, I knew she had an impressive track record: she had helped hundreds of companies find office space, and almost all of them came back to her for their second, third, fourth spaces.

Kristen and I driving to one of our first office tours together

But truthfully at the time, I had heard this from lots of brokers, and I didn't really know what made one better than the rest. 

What I saw in the last year helped me understand this industry and what makes a truly great broker.

More importantly, understanding Kristen’s skillset helped me understand where the human being is needed in this process - and why tech has struggled to overcome the trust/fear technology paradox:

1. She listens to customers. Not so much the words they say, but the message between the words, the melody behind the lyrics. Facial reactions, awkward interactions, delays in email responses.  She sees these things and makes sense of them. What is this person actually looking for?  What are they afraid will go wrong?

2. She takes action in the customer’s interest, even when they may not be aware of it.  She picks up the phone and takes the trash out (sometimes, literally, if that's what it takes).

3. She balances empathy and transparency in her communication.  Without being rude, she's able to help customers understand why things aren't the way they want them to be.  She's able to step above her own emotions and deliver with empathy.

These things — the things that make Kristen so great — are hard for computers to do, and frankly, even too hard for the best LLMs to do today. AI can’t read emotions well, it struggles with requests that are not clear, specific, direct. And it certainly struggles to delivery empathy in the way a real human being could.

But - and this is a big but - there is still a HUGE opportunity.

Before Tandem, Kristen spent most of her time doing things that did not rely on her superpowers.

This is what a day in Kristen’s life looked like, pre-Tandem:

It was shocking to me to see how much time was consumed by drafting emails. But I guess it shouldn’t have been. An average deal was hundreds of emails long, and most were highly transactional (“can we bring dogs to this office?”).

As a result of the amount of time each deal would take, Kristen could only work on a handful of deals at a time. Of course, you’d think being a part of a brokerage firm would help introduce some assembly-line efficiencies, but the reality is that almost every broker in real estate is a 1099 commission-based independent contractor (that includes everyone working for the big firms like CBRE, JLL, Cushman). Without a salary, there’s little incentive to truly collaborate and improve processes. It’s a lone wolf, eat-what-you-kill industry.

Instead, Kristen was responsible for the whole customer journey on her own, compounding the inefficiencies:

Since brokers don’t buy gizmos (see above), what was really needed was not a new set of tools, but a true marketplace. A place where the individual buyer could meet the supplier.

This would make the whole customer journey much more efficient — in all the same ways Uber made taxis more efficient, Expedia made travel more efficient, Amazon made shopping more efficient. Marketplaces deliver efficiency across the customer journey — the outcomes are almost always better than individual agents acting out of their own goodwill. (And there’s lots of research to back that up).

But, thanks to the fear/trust technology paradox, a marketplace in a space like ours had no ability to come to fruition. A marketplace meant no broker, and no broker meant no trust.

With the latest LLMs enabling human-like co-pilots, a marketplace doesn’t have to mean no broker — meaning, there’s still a space for the human being in the transaction.

By freeing up Kristen to flex her superpowers, our AI copilot enables a true marketplace (and all its benefits), while still delivering on client expectations of trust

How might this work?

Well, we started by looking at all the points in the customer journey where Kristen’s superpower skills (listening, empathy, exceptional communication) are put to work.

Digging in deeper, what we discovered was that it wasn’t throughout these activities where her trust-building abilities were most crucial, but rather at micro-moments peppered in. Quick check-in calls, a change in tone in an email message.

Can tech do the rest?

Our bet is that, yes, it can. The key factor is that Kristen’s superpowers - those peppered moments - remain as much a part of the journey as they were before. Here’s what it will take:

1. Centralized marketplace with all the expected leasing functionality (imagine Airbnb for long-term office — browse spaces and setup tours with the Host or landlord, put in bids or accept an algorithmically-generated price, review drafted lease docs, sign, pay rent, manage access control, etc.). This is what all those early startup attempts tried and failed at doing…. because they didn’t have #2 and therefore they didn’t deliver with trust.

2. AI co-pilot that interprets all inbound client messages and drafts all outbound messages. Every message gets reviewed by Kristen or someone else on the team to ensure it maximally delivers trust and empathy throughout, and we continually train the model to improve where/when/how it prompts human involvement.

Kristen and team watching the machines do their THING!

The point of this model is that it allows the technology and the human to each do what they are uniquely qualified to do.

This dramatically changes Kristen’s day. She can spend 100% of her time flexing her superpower - empathy and trust, and let the computers do the rest.

For mid to large deals, this may still include check-in calls with short clients. But when you account for all the time wasted on things she wasn’t uniquely qualified to do before, the difference in outcomes is huge.

We’re still early on our AI journey (delivery is still quite manual), but we’re already servicing 10-20x the volume of clients as a typical broker.

We think we can take this as far as 100x by building internal tooling that allows an account executive to focus on what they really should be focused on - delivering trust.

Imagine an AI copilot that:

• Organizes all the account executive (AE)’s clients by stage and priority

• Pre-drafts responses to any inbound messages, and organizes them for quick review and approval by the AE at the end of each day

• Proactively alerts the AE when clients hit points in the journey when they are most likely to need a human to lean on; pre-drafts responses

• Learns the AE’s tone and voice to iteratively improve messaging and personalization

Having fun along the way!

In short

Before LLMs, the options were either a marketplace without trust, or a broker with trust. Clients need trust, so brokers won.

Today, AI copilots can deliver human involvement when needed most.

Because that involvement can be peppered throughout the journey, the volume of clients that can be served is an order of magnitude greater, which is what makes the marketplace possible without sacrificing the trust.

Of course, none of this is easy. The hard part, though, is no longer tech - it’s building a marketplace with true liquidity and scale.

Thanks for reading this long one :) Looking forward to sharing more on our path as we embark further!