Skip to main content

​
Overview

We use Laso Finance to build Laso Finance, and paid acquisition is one of those vectors. An autonomous PPC operator runs our advertising. It is paid on the incremental contribution margin it generates from new customers. The pay goes to its Laso wallet, gets loaded onto its own Laso reloadable card, and the card buys the next round of ads. The recipe works for any business that can tie revenue back to a customer. The agent earns only when the customers it acquires make money, so its budget grows only as fast as it finds profits. Parts of this loop run today, parts are switched off until they’re proven. What is live today is updated as each step is turned on.

​
Tools Used

​
1. Set the objective and the pay

Give the agent one objective, in writing:
Maximize incremental contribution margin generated by genuinely new users acquired through paid advertising.
In simpler terms, “earn more and spend those funds via advertising”. Then pay it that number. Our agent is allocated 100% of the attributable contribution margin generated by the new users it acquires:
Direct variable costs are costs that exist as part of a transaction. Processing, network fees, referral costs, or anything else directly attributal to the purchase. Fixed costs (such as salaries) are not deducted.
Example: An ad-acquired user buys a $100 USA non-reloadable card. Laso retains $4.80. If variable direct costs for that purchase come to about 10% of that revenue, the contribution margin is $4.32, which accrues to the agent.
Paying out the full margin means the agent should, in theory, become self-financing. Every dollar it spends comes from profits it generated.

​
2. Measure all the way to money

Most PPC setups stop here:
Revenue starts when a customer actually transacts, and different products have different margins. A campaign with a cheap signup CPA (cost-per-acquisition) can still lose money if its users never fund, never transact, or only use low-margin products. The instrumentation needs to reach the margin generating activities:
As a result, the agent’s first actions were verifying each of the conversion events, adding them where needed, and attributing value per conversion.
The OpenAI Ads Manager Conversion Events tab listing events such as Human account created, Human purchases, Agent Venmo purchase, Agent PayPal purchase, Agent US card purchase, Agent first funded and Agent signup started, each with its base event, pixel, identifier coverage and event count.

Conversion events in OpenAI Ads Manager, one per product and funnel step, split across a human pixel and an agent pixel.

Three rules make attribution safe to pay on:
  • New users only. We capture the click ID and UTMs on landing, and only credit the agent operator from its own campaigns.
  • Attribution is permanent. Once an account is credited to a campaign, later transactions are credited to the campaign, funding the agent.
OpenAI Ads: the Conversions API needs the click reference oppref, but oppref is a reserved name in the tracking-parameters field. Pass it under another key, such as openai_oppref={oppref}, and map it back to oppref in your capture code.
The OpenAI Ads Manager Edit Ad dialog. The tracking parameters field lists the supported placeholders and contains openai_oppref={oppref} followed by the campaign, ad group and ad ID parameters.

The openai_oppref template in an ad's tracking parameters.

​
3. Pay only for settled margin

Keep a ledger of what the agent has earned, and follow three rules:

Settle before paying

Never pay for a click, a signup or a pending order. Only pay once the revenue has been recognized, and the variable costs can be applied to the transaction

Charge for the variable costs

Refunds and chargebacks append a negative row instead of rewriting the original. If the original was already paid, the negative row offsets future earnings.

Make payouts idempotent

Each payout covers an explicit list of ledger rows, and is confirmed against a specific, unique transaction ID.

​
4. Give the agent its own wallet and card

Now, Laso provides value to the process. The agent holds its capital the same way the agent in automated payroll does.

Create a managed wallet

Create a managed agent wallet to receive the agent’s payouts, and set a spend limit on it.

Create a reloadable card

Create a reloadable card and add it as the payment method on each ad platform.

Top up for near-term spend

When a payout lands, the agent tops up the card from the wallet with GET /fund-card-balance.

​
5. Run it daily, with memory in git

Our agent lives in a ChatGPT Project, but you can set this up with other harnesses such as Claude Code or Hermes. It runs the following steps once a day:
  1. Read its charter and state.
  2. Pull the ad and conversion data through through the ChatGPT Ads Manager plugin
  3. Decides on next course of action to improve conversions.
  4. Sends a report of the current status and next actions.
The plugin lets the agent read campaign, delivery and conversion data and make campaign changes directly from ChatGPT, without a human copying numbers out of Ads Manager.
The Laso PPC Operator daily report for October 3, 2026. Economics: 349.28 dollars of spend, 3 attributed account creations, and 0 dollars of attributed order sales, so contribution-margin metrics cannot be calculated. Actions taken: a full review, and no budget or campaign changes.

An excerpt from one day's report. With no attributed sales, the agent left budgets alone.

Its long-term memory is defined by two files in the repository:
  1. A charter (objective, pay, attribution rules, duties)
  2. A state file (verified facts, experiments, failed hypotheses, lessons).
Every day’s learnings are updated and committed in the state file. The following day, the agent picks up from that point. When measurement breaks or landing pages can be improved, it opens a pull request like any engineer.

​
What is live today

​
Build this yourself

First, install the ChatGPT Ads Manager plugin into your ChatGPT agent.

​
Quickstart

Then, run the following prompt:

Setup the agentic PPC operator for my company.

​
Setup

For a more intentional setup, use the following Laso tools in addition to the ChatGPT Ads Manager plugin.

Managed agent wallet

Give your agent a wallet to hold its own capital.

Reloadable cards

Give it a card it can pay ad platforms and vendors with.

Spend limits

Cap what the wallet can spend.

MCP server

Connect your own agent to Laso.