Thursday, 2:20 PM
The owner of a kitchen design studio opens her ad account after two weeks of not looking. The daily budget runs out before noon every day. Among the search terms that brought clicks are "free kitchen design," "used kitchens" and "carpentry course." Not one of them will ever pay her for a kitchen.
Google didn't get it wrong. It did exactly what it was asked to do. Nobody was watching.
The scene is an illustration, not a client case.
That is the real problem of managing Google and Facebook campaigns in a small business. There is no shortage of smart tools. What's missing is someone who goes over what they do, every day, and tells you in plain language what happened and what is worth changing.
In short
- Google and Meta already run AI of their own (Performance Max, AI Max for Search and Advantage+). They decide a lot, and don't always explain.
- An AI agent for your business works around that automation: it checks the account every day, flags anomalies, reviews search terms, suggests negative keywords, prepares ad drafts and writes a weekly report in plain language.
- Every budget change, campaign launch or new ad waits for your approval. The agent proposes, you decide.
- The big value is connecting the data in the platform to real inquiries, so you know what brought customers and not just clicks.
What is left to do if the platforms already use AI?
A fair question, and the answer isn't flattering to the marketing of "campaign management with AI." A large part of what goes by that name already happens inside Google and Meta, without anyone selling it to you.
| Platform | Its automation | What it decides |
|---|---|---|
| Google Ads | Performance Max | One campaign that reaches all of Google's inventory: Search, YouTube, Display, Discover, Gmail and Maps. You provide a goal, a budget, assets and audience signals, and the system looks for conversions. |
| Google Ads | AI Max for Search | An upgrade to a Search campaign: broader search term matching, ad text customization and final URL expansion. |
| Meta | Advantage+ | Applies AI to audience, placements and budget. Some campaigns are end to end, and some tools cover a single stage. |
This information comes from the official help centers and documentation of Google and Meta. Google also publishes a claim about AI Max: advertisers who turn it on "typically" see 14% more conversions or conversion value at the same CPA/ROAS. Mind the footnote: it is an internal Google figure from 2025, and only for non-retail advertisers (source). A number like that is a reason to test, not a promise that you'll get it.
So what is left? The image that explains it best is an autopilot, and the person in the cockpit.
The platform is the autopilot. It is very good at holding course, making thousands of small decisions a second and adjusting to air pressure. But it doesn't know where you really want to go, and it won't tell you "I noticed the plane is climbing to an altitude you didn't ask for."
The agent is the one who watches the instruments all the time and reports to the captain. It doesn't replace the autopilot. It sees what the autopilot is doing, spots when something is off and proposes a correction. The controls stay with you.
The platform's automation
The autopilot
- Manages bids, audiences and placements
- Optimizes toward a goal you set
- Knows the account, not the business
- Doesn't know what happened after the customer left their details
- Works inside a single platform
An AI agent connected to the business
The one watching the instruments
- Checks both platforms together, every day
- Alerts you and explains in plain language what changed
- Knows the business, the services and the customers you want
- Can connect a click to a real inquiry
- Proposes, and you approve
What an AI agent does in campaign management: six jobs that come back every week
These are not "new capabilities." They are the jobs a good campaign manager does, and that a business owner usually postpones for lack of time. The agent does them steadily.
Daily monitoring and alerts
- What it does: checks spend, clicks, inquiries and errors, and compares them to what's normal.
- What reaches you: a message only when something is off: a budget that ran out early, a campaign that stopped, an ad that was disapproved.
- What it doesn't do: it doesn't fix the budget on its own.
Search terms review
- What it does: goes through the search terms report and flags searches unrelated to the service.
- What reaches you: a short list of suggested negative keywords, with a reason for each.
- What it doesn't do: it doesn't add a negative keyword without approval.
Ad drafts
- What it does: prepares headlines, descriptions and image ideas in the business's tone.
- What reaches you: two to three versions to choose from.
- What it doesn't do: it doesn't upload a new ad without approval.
Weekly report in plain language
- What it does: summarizes what happened on both platforms, without jargon.
- What reaches you: five lines: what worked, what didn't, what changed and what's recommended.
- What it doesn't do: it doesn't present results as proof when the sample is small.
Connecting to real inquiries
- What it does: compares the ad data with the inquiries that arrived by email, form or CRM.
- What reaches you: a picture of which campaign brought serious inquiries and which brought only clicks.
- What it doesn't do: it doesn't access customer details beyond what was set up.
Automation settings check
- What it does: makes sure the platform's automation isn't doing things you didn't ask for.
- What reaches you: an alert if a new setting appeared, for example auto-applied recommendations.
- What it doesn't do: it doesn't change settings without approval.
A few details checked here, from the official documentation, that are worth knowing:
- Negative keywords work with the automation too. Google defines them as a way to exclude search terms from campaigns, and writes that in an AI Max campaign they continue to be honored (source). In Performance Max you can also add brand exclusions and negative keywords.
- The search terms report doesn't show everything. Google stresses that it hides terms with little activity for privacy reasons (source). The agent looks at what is there and doesn't claim to have seen the whole picture.
- Final URL expansion and ad text customization are settings, not gifts. The AI Max documentation says that when they are both on, asset pinning isn't preserved. Someone who notices this can decide whether to allow it.
- "Auto-apply recommendations" applies changes to the account on an ongoing basis. It can be turned on and off at any time, and it doesn't increase the budget (source). A good agent checks that it is off, or that you know exactly what it applies.
- In Meta there is control over creative enhancements at the ad level. The developer documentation notes that for each enhancement you can choose
OPT_INorOPT_OUT(source).
Who decides what
This is the most important part of the article. In paid advertising a mistake costs money the moment it happens, so the agent is never allowed to initiate one.
Responsibility in campaign management: the agent, you, and no one
The agent does on its own
- Reads data from the accounts
- Monitors and sends an alert on anomalies
- Analyzes search terms and prepares a list of suggestions
- Writes the weekly report
- Prepares draft headlines and descriptions
A draft for your approval
- Any change to a daily or monthly budget
- Launching, pausing or expanding a campaign
- A new ad, or a change to a live ad
- Adding negative keywords
- Changing a goal, audience or automation setting
Never
- Changing the payment method or billing details
- Deleting a campaign or its history
- Granting a new person access to the account
- Spending beyond the cap you set
- Working in an account it wasn't connected to
How is this enforced in practice? In two layers. The first is the rule defined for the agent, and the second, more important one, is the permission in the platform itself. According to Google's documentation, there are access levels: read only, standard and admin (source). At Meta there is a similar distinction between ads_read, which allows reading reports, and ads_management, which also allows managing ads (source).
What this means: you can start with read-only access, where the agent monitors, analyzes and proposes, and no change can happen by accident. Only once you trust the process, and only for what you want, do you open edit permission, and still with approval before any financial change. We wrote about this principle in the guide to AI agents for business, and we'll expand on it in the article on permissions, approvals and security.
What a search term looks like on its way to you: from click to negative keyword
This is the smallest and most important process in managing a Search campaign. Here you can see where the automation is, where the agent is and where you are.
From the search terms report to a negative keyword
- The agent pullsall of the week's search terms, with spend and inquiries
- The agent filtersflags terms unrelated to the service, and separates the clear ones from the "it depends"
- You approvego over the suggestions in Telegram, approve or decline
- The agent applies and checksadds what was approved, and next week checks that it really stopped
Notice step two. A term like "used kitchens" looks irrelevant to a studio that designs new kitchens, but "wooden kitchens" could be exactly what a good customer searches for. The agent can flag and explain, but deciding who your target audience is belongs to you. That's why it sends a list instead of acting alone.
One week with an agent managing campaigns alongside you
This is what an ordinary week might look like, in short messages. Pick a day and see what the agent sends and how you answer. It is an illustration based on an imaginary business, and the numbers are made up.
Sunday, 08:00 · Weekly report
The week in brief:
- Google: 41 clicks a day on average, 9 inquiries. Two of them are unrelated (people looking for a job)
- Facebook: the video ad brought 6 inquiries, the image ad 2
- The sample is small, so this is a tendency, not proof
Three suggested actions below. I didn't change anything.
Monday, 09:10 · Search terms
I went through the search terms. I suggest adding these as negative keywords: "carpentry course," "carpentry shop jobs," "kitchen template download."
3 negative keywords · spent on them this week: ₪112 (example figure)
I left "used kitchens" aside, because it isn't clear whether that audience is relevant. What do you think?
Add the three. Block "used."
Wednesday, 11:20 · Drafts
I prepared three headline versions for the Facebook ad, based on conversations the business had:
A · "A kitchen that fits your home, not a catalog"
B · "Kitchen planning with a measurement at the customer's home"
C · "A new kitchen with no budget surprises"
All three are drafts only for now.
A and C. Run them with the current budget, no increase.
Thursday, 15:40 · Alert
Alert: the daily budget in the Search campaign ran out at 11:00, when it usually happens at 16:00. Clicks went up, inquiries didn't.
Possible reason: broader match for searches. I suggest cutting the budget by 20% until we check. I haven't done anything.
Cut it by 20% until Sunday.
Friday, 08:00 · Who really got in touch
A comparison between the ads and the inquiries that reached the email this week: of 11 inquiries, 4 asked for a quote and 3 booked a meeting.
Meetings booked from: Google 2 · Facebook 1
Please update me after each meeting on whether the inquiry became a customer, so we know what really worked.
An illustration. The names and numbers are made up and don't represent any particular client or result.
Notice that in every message the agent asks for a decision. The budget didn't drop on its own on Thursday, and negative keywords weren't added on their own on Monday. It's less "magic," but it's what gives you control.
The missing data: which click became a customer
In most small businesses, the Google and Meta report stops the moment the customer leaves their details. What happened next? Did you talk to them? Did they buy? Without the answer, the algorithm optimizes for "inquiry," and may prefer cheap inquiries that don't bring money.
Google describes this itself. The guide to importing offline conversions explains that an ad can start a path that ends in a real-world sale, and recommends connecting it back to the account with "enhanced conversions for leads." It also says that from June 15, 2026, uploads move to the Data Manager API and are blocked in the Google Ads API. A technical detail, but it shows what mostly happens with such connections: they need maintenance, and whoever set them up once and forgot may find they no longer work.
This is where the agent is especially useful, because it already sits inside the channels the inquiries arrive in: email, forms, calendar. At the end of the week it can ask you "which of the 11 inquiries became a customer," and carry the answer into the report. Worth remembering:
- Privacy. If the details connected to the account are customer details, that is a regulated matter. In Israel, check what is allowed and what isn't in the article on Amendment 13 before passing on lists; readers elsewhere should check their own local privacy rules.
Where this doesn't fit
We don't think every business needs an agent for campaign management. These are the cases where we'd say "not now":
- No tracking of inquiries. Without measurement you won't know what works, and neither an agent nor a person can decide.
- A tiny budget and little data. The agent can save time, but it can't create a sample. If three inquiries come in a week, no report can say what "works."
- A business that doesn't know who its target customer is. An agent suggests negative keywords based on what you've defined. Without a definition it will guess, and you'll be approving guesses.
- Someone looking for "somebody to take care of this." If you don't have five minutes a week to approve, even a good agent won't change anything. The whole "you approve" part needs you.
Your weekly checklist: 10 minutes of approval
Even with an agent, there is a part that is yours. Here is the list worth going through once a week, after the report arrives.
The business owner's weekly check
How much does it cost?
The cost of the agent itself is built from the same components as any agent: setup, external services and support. The full breakdown is in the article on the cost of an AI agent. Two things worth keeping separate:
- The ad budget is the money you pay Google and Meta, directly from your credit card. The agent doesn't touch it and doesn't manage it.
- The agent's cost is for the work around it: monitoring, analysis, drafts and reports.
How to start: the first three weeks
- Week one: read only. Connect the agent to the Google and Meta accounts with view access. It sends one report and alerts you to anomalies. Check that it understands the account correctly.
- Week two: suggestions. It adds suggestions for negative keywords and ad drafts. You approve, and someone else (or you) applies them manually.
- Week three: execution with approval. Open a limited permission for certain actions, such as adding approved negative keywords. Budget, launching and a new ad stay under explicit approval for every action.
If you're building the agent yourself, there is a step-by-step guide. And if you're still not sure what an agent is, the simple explanation is here. The convenient channel for talking to it will usually be Telegram, and we wrote about the reasons in the article on an AI agent in Telegram.
Frequently asked questions
What is managing Google campaigns with AI?
Managing Google campaigns with AI combines two things: the automation Google runs itself, such as Performance Max and AI Max for Search, and an external AI agent that monitors the account, goes over the data, prepares suggestions and drafts, and reports in plain language. Sensitive actions, such as changing a budget or launching a campaign, are approved by the business owner.
Can an AI agent change my budget without asking?
Only if you allow it, and we recommend you don't. With the right setup, every budget change, launch or new ad waits for your approval. In addition, you can limit the permission in the platform itself, for example to read-only access.
What is the difference between an AI agent and Performance Max or Advantage+?
Performance Max and Advantage+ are tools inside the Google and Meta platforms that manage bids, audiences and placements. An AI agent for your business operates outside them: it follows both platforms together, knows the business, connects the data to real inquiries and proposes decisions to you. It doesn't replace the platform's automation but oversees it.
Can an agent also manage Facebook campaigns?
Yes. Facebook campaigns are managed in Meta Ads, and the agent can read data, summarize performance, alert you to anomalies and prepare ad drafts. According to Meta's documentation, there is a permission for reading reports (ads_read) and a separate permission for managing ads (ads_management), so you can start with read-only access.
Will the agent guarantee better results than my current campaign?
No, and nobody can guarantee that. The agent saves time, catches anomalies early and connects data that wasn't connected. If your campaigns improve, it will be the result of decisions you approve and of a business with demand, not of the agent alone.
Who sets this up at Platin Digital?
Platin Digital is an OpenAI Select Partner, and builds AI agents for businesses with whichever models and tools fit each business best, from OpenAI, Anthropic, Google and others. With each business we define which connections to turn on, which permissions to give and what requires approval, and we always start with narrow access.
Sources
- https://support.google.com/google-ads/answer/10724817
- https://support.google.com/google-ads/answer/15910366
- https://support.google.com/google-ads/answer/15913066
- https://support.google.com/google-ads/answer/2472708
- https://support.google.com/google-ads/answer/2453972
- https://support.google.com/google-ads/answer/10279006
- https://support.google.com/google-ads/answer/9978556
- https://support.google.com/google-ads/answer/2998031
- https://www.facebook.com/business/ads/meta-advantage-plus
- https://developers.facebook.com/blog/post/2025/06/03/advantage-plus-campaign-experience-for-sales-and-app/
- https://developers.facebook.com/docs/marketing-api/advantage-catalog-ads/standard-enhancements/
- https://developers.facebook.com/docs/marketing-api/get-started/authorization
