Pantera
Sign in
All articles
WhitepaperJul 202612 min read
A briefing for real estate leaders

Horizontal AI gives you an assistant.
Vertical AI gives you a system.

Generic AI is a phenomenal assistant. It becomes a liability the moment it's treated as the infrastructure for underwriting, valuation and investment decisions. That distinction is the whole of this argument.

Caleb Dunn
CEO, Pantera
The state of play

Everyone is piloting. Almost no one is succeeding.

88%

are piloting AI. Investors and landlords, in JLL's survey of 1,500+ industry decision-makers.

5%

achieved their goals. Almost no firm reached the outcome it wanted. The early effort is built almost entirely on horizontal tools.

The choice

The same rent roll, in two hands

General-purpose
Horizontal AI
Built for everyone

A lawyer, a marketer, a student, an analyst, across any task you can describe.

Starts from a blank page

Works only with whatever context you happen to give it in the moment.

Can summarise a rent roll

It will read it back to you, but it stops there.

Built for one industry
Vertical AI
Knows the domain

Understands a tenancy schedule, a rent review, a void, an exit yield from the outset.

Embedded in the workflow

Structure, data, rules and where the work happens are all purpose-built.

Knows where that rent roll goes

How it feeds voids, reviews, the expiry profile, cashflow and exit assumptions.

Vertical does not mean building your own model. The model is not the moat. The moat is the workflow, the fixed structure, the proprietary data, the audit trail and the accumulated context of your firm, all built around the model.

The heart of it

Where horizontal AI breaks

The failures don't look like failures. There's no error message. The output is fluent, confident and well formatted. It simply can't be trusted.

Inconsistent outputs

Ask the same question two ways, get two answers. Two analysts reach different numbers from the same building.

No audit trail

The reasoning lives in a private chat no colleague can see, isn’t version-controlled, and may not exist in six months.

No fixed structure

A horizontal tool rebuilds the underwriting skeleton from scratch on every prompt. No guarantee two deals share a basis.

No proprietary context

It starts cold. None of your comparables, historical deals or house view. Articulate, and anchored to nothing.

A worked example

An unauditable decision is a liability, however much time it appeared to save.

An analyst uses a generic tool to summarise the tenancy schedule on an £80m office purchase. The model reads three headline rents as net-effective, skipping 12 months rent-free and a £2m capital contribution. The IC paper looks immaculate and the deal closes. Six months later the client questions the model, but the prompt that produced it is long since deleted. When your PI insurer or the RICS asks “how did you arrive at this figure?”, “the AI told us” is not an answer you can defend.

Removed by design

What vertical AI does differently

Consistency from structure

The framework is fixed and applied the same way each time. Same deal, same shape of answer.

Every output is auditable

Version-controlled, with a record of inputs and assumptions behind each number, defensible to an IC or auditor.

Your context is built in

It already holds your comparables, assumptions and prior deals. Every output grounded in what your firm knows.

Checks run on every deal

Flags gaps, inconsistencies and inputs that look aggressive against your own history, automatically.

The example that looks easy

“£42.50 psf”… but which £42.50?

Horizontal AI sees
£42.50 psf

A number on a page. It extracts what looks like the answer and presents it with the same confidence as the truth.

Vertical AI asks
Which £42.50?

Headline rent? Net effective? Passing rent? ERV? Zone A? Or a comparable adjusted for rent-free periods and incentives, the figure that actually matters.

Already proven

The proof is already in another industry

Law had ChatGPT too. The serious firms still moved to vertical tools built around the work, and the market voted with its feet.

30%
less contract review time across 4,000+ lawyers at A&O Shearman.
<48hrs
deal turnaround, down from a week, winning business on speed alone.
$16bn
combined valuation of the leading vertical legal-AI platforms by mid-2026.
In this context

The winning AI interface in real estate won't be a blank chat box. It will be the underwriting workflow itself.

Pantera is not ChatGPT wrapped in a property interface. It's a system built around the actual workflow of real estate underwriting, and it does not replace judgement; it removes the friction that delays it. Every assumption belongs to the analyst, every model belongs to the firm.

Our mission

To build the best-performing and most trusted AI platform in real estate investment.

End of article
Written by Caleb Dunn, CEO of Pantera · Jul 2026
Back to all articles

The question is no longer whether to use AI. It's which kind.

Only one is built to earn your trust. See what vertical AI looks like on one of your own deals.