AI for ad management starts with drafts: one channel, weekly reviews
On small budgets, AI drafts ads for one channel, you review weekly, and a person caps and decides budget and delivery.
Key points
- Start AI with ad drafts, then expand
- First task: drafts for one channel
- Each week, keep ads that drew inquiries
- Set caps; you decide budget & delivery
Contents
Start AI with ad drafts, then expand step by step
A&A perspective
Bringing AI into ad management starts with drafts of ad copy and images (ad drafts), and widens what you hand over one step at a time. It means handing ad work to AI: making ad drafts, setting up each place you advertise (a channel, such as search ads or social media ads), delivery (deciding when and where ads run and putting them live), and splitting the budget.
From the sources
Anthropic, the company that builds the AI model Claude, presents a customer story about Advolve, a small company that uses AI to automate the entire online customer acquisition process for businesses.
From the sources
According to its co-founder, the company started with a system for producing ad images and copy. It breaks each ad into layers (background, art, fonts and copy), and AI generates and evaluates each layer. The basis is data from past ads that performed well.
From the sources
From there, it expanded to configuring, running and optimizing campaigns across several platforms. The co-founder says it also uses data to move budget between platforms automatically.
A&A perspective
Automate & Augment (A&A), the company writing this article, thinks the same order works for a small company trying AI. If something goes wrong at the drafting stage, a person can stop it before the ad runs. If something goes wrong at the budget and delivery stage, the mistake turns straight into ad spend.
From the sources
The case study also gives figures, such as how much work time was cut and how much customers' revenue per ad dollar rose. The co-founder says its customers include some of the largest companies in Brazil. The case does not say what the figures were compared with or over what period.
A&A perspective
The story is Anthropic presenting one of its own customers, not a third-party check. A&A does not treat the figures as results a small company can expect; it reads the case as a model for the order in which to expand.
Make your first task ad drafts for one channel
A&A perspective
The first task to hand to AI is drafting ads for one channel you already use. With a single channel, it is easy to see which draft brought in inquiries. With two, you cannot tell whether a result came from a different draft or a different place.
A&A perspective
Give the AI three things: your current ad copy and images, what customers most often ask about, and the kinds of requests you cannot take on. Listing what you cannot take on cuts drafts that would attract inquiries you would have to turn down. For copy alone, a chat AI such as ChatGPT is enough.
Hypothetical example
For example, say a 20-person company that maintains commercial equipment runs a small search ad. The owner gives AI the current ad copy and a common question, "I can't get parts for an old machine," and asks for five versions of the ad copy.
Hypothetical example
The owner picks two of them and runs them alongside the current ad. After that, once a week, the owner checks which version brought inquiries. This example is hypothetical, not a real company's result.
A&A perspective
A person reads every AI draft before it runs, checking for wording that is not accurate and for promises the company cannot keep. A person decides whether it runs.
A&A perspective
Whether the test is worth it depends on what this work costs you now, in time and in outside fees. So before you start, write down how many hours a week go into ad drafts and how much you pay outside help.
| Step | What AI does | What a person does |
|---|---|---|
| Hand over | Uses the three inputs as material for drafts | Gives the current ad copy and images, common customer questions, and the kinds of requests you cannot take on |
| Draft | Writes ad copy drafts for one channel | Picks drafts and runs them alongside the current ad |
| Before it runs | Does not decide whether it runs | Checks for inaccurate wording and promises the company cannot keep, then decides whether it runs |
In a weekly review, keep the ads that drew inquiries
A&A perspective
In the weekly review, decide which drafts to keep by whether they brought inquiries, not by how many clicks they got.
From the sources
In "Building effective agents," an article for developers on how to build AI systems, Anthropic describes a pattern in which one AI produces a draft and another evaluates it and has it revised, in a loop. It says the pattern works best when there are clear evaluation criteria.
From the sources
Advolve's system used data from past ads that performed well as the basis for evaluation.
A&A perspective
A small company does not have those criteria yet. A&A sees the weekly review as the time to build your own.
A&A perspective
If you sell to businesses, base the criteria on inquiries rather than clicks. A draft with many clicks will not lead to deals if most of the people clicking are only looking for free materials.
A&A perspective
Look at three things in the review: clicks per draft, the number of inquiries, and what the inquiries are about. Give each ad its own landing page address so you can tell which draft an inquiry came from. Let AI tally the numbers and draft the next versions; a person decides which drafts stay.
A&A perspective
If you get only a few inquiries a month, one week of numbers will not separate a good draft from a weak one. In that case, look at four weeks together. If you still cannot decide, write down that you could not, and run the same drafts for another four weeks.
Set caps, and let a person decide budget and delivery
A&A perspective
For budget changes and ad delivery, a person sets the spending cap, and a person decides each change.
From the sources
The same Anthropic article says that systems in which AI acts on its own judgment bring higher costs and the potential for compounding errors. It therefore recommends extensive testing somewhere separate from live use, with appropriate guardrails.
From the sources
As ways to keep control, it also mentions capping how many times the AI repeats a step and having it pause for a person's check along the way. The Advolve case lists, among the AI's jobs, validating data in real time and checking that the process has no gaps or inconsistencies.
A&A perspective
A&A reads this as a safeguard built into the system, separate from a person's check.
A&A perspective
On a small ad budget, a single budget change shifts that month's spending. Until the review has given you criteria, A&A recommends that a person decides on budget and delivery.
A&A perspective
A person sets two caps: the most you will spend on ads in a day, and how much you may raise spending in one change. The AI's part stops at writing proposed changes and the reasons for them.
A&A perspective
In a week when the system that collects clicks and inquiries (your tracking) stops, tell two cases apart: results were zero, or the numbers simply were not recorded. Do not raise the budget while numbers are missing.
A&A perspective
Some argue that the budget split is exactly what should be automated for the biggest gains, and Advolve does move its budget allocation automatically. A&A sees that as a later stage, once evaluation criteria and data are in place. When you can pick the drafts to keep several weeks in a row, and you have a way to notice when tracking stops, move on to having AI propose budget changes.
Summary
A&A perspective
Start AI with ad drafts, then expand step by step
A&A perspective
Make your first task ad drafts for one channel
A&A perspective
In a weekly review, keep the ads that drew inquiries
A&A perspective
Set caps, and let a person decide budget and delivery
A&A perspective
When you run drafting and reviewing in house, the knowledge of which drafts bring inquiries stays with your company.
Bringing AI into ad management starts with ad drafts, where a person can stop a mistake before it runs. Let AI draft for one channel, keep the drafts that bring inquiries in a weekly review, and have a person set caps and decide on budget and delivery.
FAQ
Can I try this if an ad agency runs my ads?
Yes. Use AI only to make the ad drafts and hand them to the agency, leaving delivery and budget with the agency under your current contract. Once a week, review which drafts ran together with the agency.
Can I expect the results from the overseas case?
You cannot count on them. The case's figures do not say what they were compared with or over what period. The revenue-per-ad-dollar figure comes from customers that include some of the largest companies in Brazil. Start small and check against your own number of inquiries.
When can I let AI adjust the budget?
Once your reviews let you pick the drafts to keep several weeks in a row, and you have a way to notice when tracking stops. Even then, a person sets the most you will spend on ads in a day and how much you may raise spending in one change.
Sources & editorial note
Primary pages read for this article. Publication dates below belong to the sources; access dates record our research.
- Advolve automates digital marketing with Claude
Anthropic · Publication date not stated
Accessed 2026-10-22 - Building effective agents
Anthropic · 2024-12-19
Accessed 2026-10-22