Skip to content
Gold and glass panels converging in a building's geometry

Investment thesis

A portfolio built to compound, so every business we add gives more to all of them.

Owning several companies is not a strategy.

Here each business has a job in the same system, and each one makes the next cheaper to build and better for the small businesses it serves.

1

Create demand

Free utility, search, directories and AI guidance attract people who already have the problem. Nobody pays to be useful to them.

2

Charge for the software

Once a provider can see real demand arriving, the software that runs their business becomes worth paying for: AI phone answering, scheduling, quotes and invoicing.

3

Grow with the customer

As more people and businesses use it, orders, ads and referrals grow, and each new business reuses what the last one built.

Where this sits

Berkshire combines cash flows. We combine capability.

Two models have defined how diversified companies get built. Each solved one problem and left another, and this portfolio is designed around both lessons.

The classic conglomerate

Combines cash flows

Berkshire Hathaway's businesses share a balance sheet, and capital moves to wherever it earns most without paying the toll of outside markets. They do not share systems: what the railroad learns never reaches the insurer. The combination is financial, not operational.

Vertical software

Sells one workflow per industry

Each new vertical needs its own sales, product and support teams, so overhead grows with every market entered. Growth expectations then push fees upward until the software becomes a toll on the small businesses it was built to serve.

Surana Holdings

Combines capability

Every business runs on our own engine, each with its own data, so a tool built for one is available to all thirteen. Capital still follows evidence, as in a conglomerate. What grows is the system as well as the cash, and fees stay low enough that customers stay.

The argument

The expensive part of a vertical marketplace is not the software.

It is the supply. Assembling every pro, shop, restaurant and pharmacy in a category — and making each record genuinely useful — is where the years go. Once that work is done, the software on top of it is the cheap part, and it is the same software every time.

What the first business paid for

Zerofi paid to learn how to load a state license board, how to rank a directory without letting a general contractor win every trade query, how to bill an ad click without billing a bot, and how to make a claim flow that a small business will actually finish.

What the others inherited

GymGuide’s advertising engine is Zerofi’s, reused. AutoDoctors’ repair guides are Zerofi’s DIY engine, reused. Every claim flow in the portfolio descends from the same design. The cost of building the next business is a fraction of the first, and falling.

zerofi.ai
The zerofi.ai home page: what needs fixing at home? Fair prices and real local pros, free, with a search by problem and place
Zerofi.ai, the first marketplace. Its claim flow, ad engine and DIY guides now run in other businesses.

Why the portfolio can compound

The same relationship can create several legitimate revenue paths.

Not forced bundling. Adjacent value, offered when the same customer genuinely has the next problem — and declined when they do not.

  1. Step 1

    Demand

    Search and free AI guidance

  2. Step 2

    Marketplace

    A qualified connection

  3. Step 3

    Business software

    Scheduling, quotes, invoicing

  4. Step 4

    AI agent

    Voice and automation

  5. Step 5

    Commerce

    Ads, referrals, transactions

Software and goods in one loop

The software runs the business. The goods it reorders come from us.

Most industry software stops at the screen: a restaurant still buys its coffee from a distributor who knows nothing about its sales. Here the business software and the supply line can belong to the same company — the software keeps the customer, and the goods carry the repeat volume.

Food and hospitality

  1. 1Live

    A café claims its RestaurantPro page and takes orders at a flat 3%

  2. 2Live

    It buys Moksha's wholesale coffee and matcha from the same dashboard, at prices the database sets

  3. 3Next

    Reorders prompted by its own sales, so the beans arrive before they run out

Trades and industrial

  1. 1Live

    A contractor runs jobs, quotes and calls on Zerofi's business software and Zoe

  2. 2Next

    Nirah's cable ties and fixings offered at wholesale inside the same workflow

  3. 3Live

    Stealth, a Made in USA line for buyers who need American-made product

Live means a customer can use it today. Next means planned, not yet built. Stealth is made in the USA by Nirah’s US manufacturing partner; no contract award is claimed. Beyond these steps, autonomous delivery closes the food loop and robotics the industrial one — the progression is on the Vision page.

restaurantpro.ai
The restaurantpro.ai home page: restaurants near you with no service fees, cuisines, and an AI concierge that picks from real menus
RestaurantPro: the restaurant claims its page and takes orders at a flat 3%.
Moksha Coffee products: whole-bean bags, a gift box with a mug, and branded hot and iced cups
Moksha: the coffee and matcha it can reorder from the same dashboard.

Why a low take rate holds

High fees pay people to leave.

Restraint is the promise. It also happens to be the position that lasts.

At a large share of the ticket

When a platform takes a large share of an order, the restaurant and its regular customer both save by going direct the second time. The platform has to spend to stop them, and the fee rises to pay for the spending.

At 3%, and $0 on a referral

Going around the platform saves almost nothing, and the operator would give up the tools that came with it. A take rate this low is not a promotion to be raised later — it is the reason operators stay, so raising it would undo the business.

Operating discipline

Capital follows evidence.

The portfolio stays lean by reusing its core systems and scaling a channel only once its early signal has moved. These are the six measures we watch.

Demand
Unpaid sign-ups and return visits
Supply
Claims and verified providers
Revenue
Paying customers and average revenue per customer
Retention
Seat, merchant and user cohorts
Economics
Margin after direct costs, including AI
Trust
Quality, privacy and permission incidents

Fund proof, not vanity.

Shared code lowers build cost. Brand-specific data boundaries preserve trust.

What is not on this page

No revenue projections, no forecast, no management case. Those exist, they are maintained carefully, and they are shared under NDA with people who have a reason to see them — not published to anyone who loads the page. If you are one of those people, ask.