How to use AI to optimise running ad campaigns without a full time analyst
To optimise running ad campaigns without a full-time analyst, split the analyst's job in three. Let Google's and Meta's own bidding handle auctions. Let an AI tool or agent watch the account daily, spot waste and draft specific fixes. Keep a person for a short weekly review that approves or rejects those fixes under a hard spend cap. The setup takes a day; the routine takes about 30 minutes a week once tracking is right.
What does an analyst do that you need to replace?
Before choosing a tool, list the jobs. A paid media analyst on a small team usually does six things:
Pulls numbers. Exports, pivots, weekly reports.
Spots anomalies. A campaign that suddenly overspends, underspends or stops converting.
Cleans search terms. Adds negatives for queries that spend and never convert.
Watches pacing. Checks whether the month's budget will land on target.
Diagnoses. Works out why cost per acquisition rose: bids, budget, creative, landing page or tracking.
Tests. Plans and reads ad copy and landing page tests.
AI handles the first four well today. The fifth it can draft, with evidence, for a person to judge. The sixth still needs someone to decide what is worth testing.
Step 1: What must be in place before AI can help?
Every automated decision rests on the data under it. Do this first, or the AI will optimise the wrong thing quickly.
Conversion tracking. One primary conversion per goal, firing once per real action. Google's docs describe conversion tracking as providing "key insights into users' actions after viewing or clicking an ad" (Google Ads API docs). Test it with a real purchase or form fill.
Targets. Write down the target cost per acquisition or return on ad spend for each campaign. A tool cannot tell you a campaign is "too expensive" without one.
A hard spend cap. A daily and monthly ceiling on each ad account, set by a person.
Clean structure. Campaigns named by goal and market, so a report reads without a key.
One owner. A named person who approves changes, even if it is only 30 minutes a week.
Step 2: Which work should the platforms' own AI do?
The platforms already automate the auction. Use that rather than fighting it.
Bidding. Google's Smart Bidding strategies, such as Target CPA and Target ROAS, set a bid in each auction. Meta's delivery system does the same for ad sets.
Daily pacing. Google's demand-led budget pacing, rolled out to Search campaigns, spends more on high-demand days and less on slow ones while staying within daily and monthly caps (Google).
Distribution across ad sets. Meta's Advantage+ campaign budget spreads one campaign budget across ad sets by performance.
What the platforms do not do is tell you when a campaign should not exist, when a search term is irrelevant to your business, or when to move money from Google to Meta. That is the gap an AI tool or agent fills.
Step 3: What should an AI tool or agent watch for you?
Set the tool to watch for a short list of problems, each with a threshold. Optmyzr's guide to small accounts gives useful starting points (Optmyzr):
Spend anomalies. Alerts when spend, clicks or conversions move far from their average, which catches budget typos and broken tracking within a day.
Pacing. A check that campaigns have reached about 50% to 60% of the monthly budget by the 15th.
Wasted keywords. Pause candidates with no conversions in 60 days and spend above three times the CPA goal.
Wasted search terms. Negative candidates with no conversions in 30 days and spend over a set figure, such as $400.
Declining keywords. Keywords that converted in the prior 90 days but not the last 30, flagged for a person to investigate.
Adjust the figures to your volume. A small account needs longer windows, because thin data turns noise into false alarms.
Then decide which of these may act alone. A sensible start for a team with no analyst is: alerts run alone, everything that changes spend waits for approval.
Step 4: How should you review and approve changes?
Replace the weekly report with a weekly decision session. Thirty minutes, same day each week:
Five minutes: the headline. Spend against plan, conversions against target, by channel. Ask the tool in plain English if it supports that.
Fifteen minutes: the queue. Read each proposed change with its evidence. Approve, reject or edit. Reject anything you cannot explain to a colleague.
Five minutes: what changed last week. Did last week's approved changes help? Compare each against the same number of days before it.
Five minutes: one test. Pick a single ad copy or landing page test for the week.
Between sessions, only alerts reach you. If you are getting more than a few a day, your thresholds are too tight.
Step 5: How do you know it is working?
Track four numbers for the first 90 days:
Wasted spend share. Spend on keywords, search terms and campaigns with no conversions, as a share of total. It should fall.
Time to fix. Days between a problem appearing in the data and a change going live. It should shrink.
Approval rate. The share of proposals you approve. Very low means the tool is miscalibrated for your account; near 100% means you may not be reading closely.
Cost per acquisition against target, by campaign.
Take a baseline before you start. Without one, you are measuring a feeling.
What mistakes do teams without an analyst make?
Automating before tracking works. The tool optimises toward a broken signal and scales the wrong campaigns.
Letting rules raise spend. A rule that raises budget on good ROAS keeps raising it as returns fall. Keep raises behind approval and a cap.
Changing too often. Daily edits keep campaigns in the platform's learning phase.
Nobody owns the queue. Proposals pile up unread and the tool becomes an expensive report.
Ignoring advice. Some findings have no safe automatic fix, such as a landing page problem. Someone has to act on them.
Which tools fit a team without an analyst?
All checked on the vendors' own sites on 24 October 2026.
Optmyzr
Best for: Google-heavy accounts that want audits, alerts, rules and one-click fixes.
Pricing: from $299 a month, or $209 billed annually, for up to $25,000 of monthly spend (Optmyzr pricing).
Limitation: many tools to choose from; it helps to know what you are looking for.
Madgicx
Best for: Meta-first ecommerce brands; AI Marketer audits the account and recommends actions (Madgicx).
Pricing: shown in the app after sign-up and tied to spend; 7-day free trial (Madgicx pricing).
Limitation: Meta first.
Birch (formerly Revealbot)
Best for: teams comfortable writing rules; Birch AI drafts a rule from a plain-language request.
Pricing: Pro at $99 a month for up to $10,000 of spend (Birch pricing).
Limitation: rules are only as good as the person writing them.
Albert.ai and Smartly
Best for: large advertisers. Both price through sales (Albert pricing, Smartly).
Limitation: usually beyond a small team's budget and setup capacity.
Where does Nagent's NIA fit?
We make Nagent. Nagent is multiplayer AI for growth: people and AI agents in one shared context, delivering the number. NIA is the AI performance marketing lead on the Marketing team, led by MIRA, the AI Marketing Chief of Staff. A sub-agent works each channel under her: Google Ads, Meta ads (Beta) and ChatGPT ads (Beta) today.
She covers the analyst jobs above without the export step. On the Advertising screen you ask about a connected Google Ads account in plain English, such as "Where is budget being wasted?" or "How did the last 14 days go?", and she answers from the account's stored data and can draw a chart. Insights and Canvas give prebuilt views of spend against return, cost per conversion, campaigns, search terms and trends.
For Step 3, her Agent optimise tab runs a diagnosis over the window you choose. She checks the window is fresh and well covered first, a keyword pause must clear a wasted-spend floor and a click floor, and each fix is validated as a dry run. Apply queues it on Actions with what it does and a risk note, which is your Step 4 queue. Each change that lands is recorded with the day it landed, so its effect can be set against the window before it, which is Step 5.
She also works in the Marketing room alongside your team, so a question asked there gets an answer in a thread that others can see.
What does NIA not do?
She does not act alone on spend. She starts at L2 Assist, so every live change waits for a person.
She does not set up tracking. On Google Ads, for a leads or sales objective, she will not build without an enabled conversion action, and the check names what is missing.
On Meta she works at campaign level: pause, resume, budget changes and creating an empty campaign. Ad sets, audiences, ads and creative stay in Ads Manager.
She does not pick your tests or write your strategy. She drafts; a person decides.
If a 30-minute weekly review is the most time you can give, see how NIA works and the autonomy ladder that governs what she may do alone.
Frequently Asked Questions
Can ChatGPT run Google Ads campaigns for me?
A general chatbot can draft keywords and ad copy, but it cannot see your account unless it is connected to it, and it cannot apply changes safely without approvals and spend limits. A connected ad agent reads the live data and proposes changes for a person to approve.
How much time does AI-assisted ad optimisation take each week?
Once tracking and thresholds are set, about 30 minutes a week for one owner on a small account, plus a few minutes for any alert. Expect more time in the first month while you tune thresholds.
What should I automate first without an analyst?
Start with alerts and pacing checks, which change nothing and catch the costliest problems. Add proposed changes with approval next. Leave any automatic spend increases until last, behind a hard cap.
Does AI replace a PPC analyst?
It replaces most of the reporting and monitoring. It does not replace the judgement about goals, tests and trade-offs, so someone on the team still needs to own the account.
Want to see this routine on your own account? Book a demo.