Flagship report · founding edition

The AI Business P&L Report: July 2026

Published July 22, 2026 · Updated monthly · Methodology and rounding policy · Raw data

This is the founding edition of a monthly report on what a portfolio of AI-operated businesses actually earns and spends. The portfolio is run by a small operation using AI agents for fulfillment, research, and content production, with humans handling client conversations and final accountability. It is early. That is the point: you are seeing the books from the start, not a survivor's highlight reel.

Portfolio at a glance

Venture	Model	Age	Status, July 2026
V1 (services desk)	Marketplace freelancing, systems-driven fulfillment	~3 weeks selling	$2,700+ collected + funded escrow
V2 (content network)	Ad-monetized AI content, multiple channels	months	80,000+ subs, six-figure run rate
V3 (trading research)	Algorithmic futures strategies	months	expense-only, unprofitable to date
Killed: sports documentary channel	Long-form AI video	killed 07/2026	$0 net, copyright risk
Killed: faith-content brand	Short-form AI video	killed 07/2026	audience economics failed

V1: the services desk

A desk selling verified lead research, market research, and automation builds on a freelance marketplace. Fulfillment runs on research and verification systems we built in-house; a human operator owns every client relationship and stands behind every deliverable. It began bidding in earnest in mid July 2026.

First three weeks: five active clients, more than $2,700 in collected revenue plus funded escrow. According to the Ledger's records, no single client exceeds half of that total, so it is not one lucky whale.
Customer acquisition cost: roughly $300 in marketplace bidding fees over the period. No ads.
Fulfillment compute on a representative ~$120 research job: under $15 in model tokens (estimate, marked per policy). Gross margins on research work run high, which is why everyone wants this business and why bidding competition is the real cost.
What the number hides: proposal volume. Winning five clients took hundreds of submitted proposals and a daily bidding system with its own learning loop. The work is real even when the fulfillment is automated.

Figures from marketplace transaction records; escrow counted only when funded by the client. Client names redacted by policy.

V2: the content network

An AI-operated content network across several channels and platforms. It has passed 80,000 subscribers and operates at a six-figure annual revenue run rate. Production is near fully automated: scripting, voice, editing, packaging, and scheduling run as pipelines, with human review at defined gates.

The honest caveat: this outcome is the survivor of multiple attempts. The two killed ventures below came from the same playbook and did not survive contact with reality.
Platform risk is the tax on this model. Demonetization, policy shifts, and copyright exposure are existential, not hypothetical; one of our kills below was exactly that.

V3: trading research

Algorithmic futures strategy research, run with AI agents doing strategy iteration against a cloud backtesting platform. To date this venture is an expense. Published backtest Sharpe ratios in the 1.4 to 1.7 range exist in our public research catalog, and we treat them as research artifacts, not income. No live-money profit is claimed because none has been booked.

Post-mortems: what we killed and why

The sports documentary channel

Long-form AI-produced sports documentaries. Killed in July 2026 for copyright exposure: the format depended on third-party footage in ways that made the channel structurally unsafe no matter how good the product got. Lesson we paid for: if the core input is someone else's IP, you do not have a business, you have a countdown.

The faith-content brand

A daily short-form devotional brand. Production automated cleanly and output quality was fine. Killed in July 2026 because the audience economics never cleared: the niche's monetization per view could not support even an automated cost base. Lesson: automation lowers costs but cannot rescue a vertical whose unit economics are underwater.

What this portfolio says about "making money with AI" in 2026

Services with verifiable output (research, data work, automation builds) monetize fastest: weeks, not months, with near-zero capital.
Content compounds but is hit-driven and platform-taxed. Expect kills. Budget for them.
Trading and speculation are research programs, not income, until proven otherwise with live records. Anyone telling you differently is selling something they have not booked.
The moat is not the AI. Everyone has the same models. The moat is verification: shipping work that survives a hostile audit is what clients pay for and what platforms reward.

Thinking of starting your own AI venture? Every venture in this report went through an adversarial feasibility pass before we funded it, and the ones that failed the pass never cost us a launch. We run that exact research on other people's ideas as a fixed-price service: unit economics from published data, competitor scan, demand evidence, and a sourced go or no-go verdict. Venture Feasibility Study, $349 brief / $749 full. Already committed to building? We build the automation layer too.

Get next month's report: email hello@verifieddesk.com with subject "Subscribe" and the monthly edition lands in your inbox when it publishes. Data only, no course pitches.