The Trust Deficit Premium

Nobody has a line item called trust. Everybody pays for it.

I've spent twenty years in businesses where that payment is large and badly accounted for. US mortgage servicing. Residential property management. Industrial access control. Indian real estate. Different countries, different regulators, different decades, and underneath all of them the same problem: a stranger has to be made trustworthy enough that someone will hand over money or keys.

What changes between markets isn't whether you pay. It's where the payment shows up in your accounts. And that turns out to decide what AI is worth to you.

Two ways to pay

In a high-trust market, the cost of establishing trust is paid as friction. You fill in a form. You wait for a confirmation. You take an afternoon off to be present for something. The cost is real, it's measurable if you bother to measure it, and it sits on nobody's balance sheet. It's absorbed by customers, a few minutes at a time, and it never appears in a financial statement.

In a low-trust market, the same cost is paid as capital expenditure. You build a site office so there's a person to ask. You build a sample flat so the buyer can stand inside something real. You maintain a branch network, a field force, a layer of relationship managers, a channel of intermediaries with reputations in their own neighbourhoods. All of that is trust infrastructure. All of it is depreciating on your balance sheet, indistinguishable from operations.

Same underlying cost. Completely different accounting treatment.

Now introduce a technology that can answer questions at three in the morning, remember every previous conversation, never get bored on the fortieth repetition, and never quietly steer a buyer toward the unit that pays better commission.

In the high-trust market, that technology shaves friction. The customer waits less. Good, genuinely, but it's an efficiency gain against a cost that was never on your books, so it's correspondingly hard to see in the P&L. This is, I think, a large part of why enterprise AI returns have been so disappointing in advanced economies — the thing being saved was never being counted.

In the low-trust market, the same technology is doing something else. It's substituting for capital expenditure. And that you can see.

Why this isn't just an India argument

I first noticed this in American mortgage, not Indian property, which is why I'm fairly confident it isn't a story about India.

US mortgage and default servicing is, in trust terms, a low-trust business operating inside a high-trust economy. Nobody believes anybody. Every assertion needs documentary proof, every document needs verification, every verification needs an audit trail, and the whole apparatus exists because the consequences of misplaced trust in 2008 were severe enough to be legislated about. The result is an industry that spends enormously on establishing facts that are, in principle, already known.

When we built asset management and renovation tracking systems for a US residential portfolio of a couple of hundred million dollars, the savings that mattered weren't in the obvious places. Automating the renovation tracking took manual effort from days to minutes and produced a mid-six-figure annual saving. Retiring a reporting process built on an eighteen-megabyte Excel model produced a low-six-figure one. Putting ROI, IRR and gross margin behind a single click cut the time to a buy, sell or renovate decision in half.

None of those were AI. All of them were the same mechanism: a cost that existed because verifying something used to require a person, removed once verifying it didn't.

The pattern generalises. Wherever trust is expensive and the expense has been capitalised into physical or human infrastructure, there's a large, visible, bookable return available to whoever removes it. Wherever trust is cheap, the return is real but diffuse, and it shows up as customer satisfaction rather than margin.

The test

Here's how to find out which market you're in, and it takes about twenty minutes.

Walk your balance sheet and your headcount, and mark everything whose primary function is to make your organisation credible to a stranger. Not to deliver the product. Not to operate the business. To be believed.

For most Indian consumer-facing businesses, the number is much larger than people expect. For most Western SaaS businesses, it's close to zero, because the trust was established by the category and the contract rather than by a building.

That number is what AI is actually competing against in your business. Not your software licences. Not your process costs. Your trust infrastructure.

And if it's large, you have something most of the enterprise AI world doesn't: a return that's visible in the accounts rather than inferred from a survey.

What happened when we tested it

We ran a residential launch last year with no sample flat and no site office. 8,838 people registered against 1,419 homes released. Six times oversubscribed, ₹640 crore allotted, across about five weeks. The AI handled roughly 45,000 conversations.

Eight percent of buyers completed without speaking to a human at any point.

I'd like to be precise about what that does and doesn't prove. It's one launch, in one micro-market, over five weeks. It's internal data, not research. It proves that the trust infrastructure was substitutable in that instance, at that price point, for that buyer. It does not prove it generalises, and I'd be suspicious of anyone who claimed otherwise on a single data point, including me.

What it did establish, to my satisfaction, is that the sample flat and the site office were not physics. They were price — things that had been true for so long they'd stopped feeling like decisions.

The more interesting finding was the one I'd have bet against. Sixty-eight percent of registrations still came through channel partners, on a launch running entirely on AI. The trust infrastructure didn't collapse. One part of it turned out to be substitutable and another part didn't, and the line between them wasn't where I'd drawn it.

The part I'm still unsure about

There's a version of this thesis that goes too far, and I want to mark it rather than pretend I've resolved it.

If AI can manufacture trust cheaply, and trust infrastructure is a large part of the cost base in low-trust markets, then in principle the advantage flows to whoever moves first — and low-trust markets could leapfrog. That's the exciting version.

The World Bank's July 2026 background paper complicates it considerably. Across 1,563 frontier firms in 83 countries, adoption ran 63% among those with digital foundations and 16% among those without. If you can't integrate the technology, the size of the prize doesn't matter. Most firms in low-trust markets are in the second group, and the trust infrastructure they'd be replacing is precisely what they've built instead of digitising.

So the premium is real, and access to it is not evenly distributed. I think that's the honest statement. The businesses that capture it will be the ones that did the unglamorous work first, and they'll capture more of it than any comparable business in a high-trust market would.

That's worth knowing if you're deciding where to spend the next three years.