The short answer
- An AI media buyer is software that reads your ad accounts, decides where budget and bids should go, and makes the changes in Google, Meta and other platforms, ideally after you approve each one.
- It beats a human at volume and consistency: checking every campaign daily, every search term weekly, and comparing cost per result across channels on the same basis.
- Keep humans on offer, creative direction, channel strategy and big budget calls. Automate audits, pacing, negatives, routine budget shifts and reporting.
- Evaluate one by what it can actually write to per channel, whether every change needs approval, and whether it respects how each platform's learning works.
What an AI media buyer does
AI media buying covers the operational loop a media buyer runs every week: read performance, find waste and winners, shift budget, adjust targeting and keywords, refresh ads, and report.
The difference from older media buying software is that the AI does the analysis and drafts the change itself. Traditional tools give you dashboards and rules. An AI media buyer reads the dashboards for you and proposes "move $50 a day from campaign A to campaign B, here is why."
It is not the same as the platforms' own automation. Google's Smart Bidding and Meta's Advantage+ optimize inside one platform toward the goal you set. An AI media buyer works one level up: which campaigns get budget, which channel gets the next dollar, which keywords to exclude, which ad sets to cut.
AI media buyer vs human media buyer
| Task | Human buyer | AI media buyer |
|---|---|---|
| Daily pacing checks | Spot checks on big campaigns | Every campaign, every day |
| Search term review | Top terms, when time allows | Full list, every review |
| Cross-channel comparison | Spreadsheet, often monthly | Same metrics side by side on demand |
| Spotting a broken tracking setup | Good, if they look | Good at flagging anomalies, needs human confirmation |
| Offer and landing page judgment | Strong | Weak |
| Creative direction | Strong | Drafts variants, needs direction |
| Reading context outside the data | Strong (a sale next week, a stockout, a PR issue) | Only what you tell it |
| Cost | Salary or agency fee | Software subscription |
The best setup for most teams is one human buyer with an AI media buyer doing the first draft of every routine task. The solo media buyer playbook shows how one person can run several channels this way.
How cross-channel budget moves work
The most valuable thing an AI media buyer does is move money to where it returns most, across channels that report results differently.
Step 1: Put channels on one basis
Meta, Google and LinkedIn each count conversions with their own attribution windows. Compare cost per result and ROAS from each platform, but sanity check the totals against your own analytics or backend so you are not moving budget toward the platform that counts most generously.
Step 2: Look at the marginal dollar, not the average
A campaign with a $40 average cost per acquisition may be paying $90 for its last 20% of conversions. Before adding budget, look at what happened the last time spend went up: did conversions rise in proportion, or did cost per result climb?
Step 3: Move in steps the platforms tolerate
- Google Ads: Google's help center says a campaign can spend up to twice its average daily budget on a given day, and no more than 30.4 times the average daily budget over a month. Budget changes shift the monthly cap, so plan moves against the month, not the day.
- Meta: Meta says an ad set needs around 50 optimization events in the week after its last significant edit to leave the learning phase, and that budget changes can count as significant depending on their size. Large, frequent budget swings on Meta keep ad sets learning.
A worked example
A store spends $6,000 a month: $3,000 Google Ads, $3,000 Meta. Google search returns a 4.2 ROAS, Meta prospecting 2.1, Meta retargeting 5.0 but capped by audience size. Google search is losing 30% impression share to budget.
- Move $600 a month from Meta prospecting to Google search, where the data shows unmet demand.
- Leave retargeting alone; more budget there buys frequency, not new buyers.
- Re-check in 14 days whether Google's incremental conversions held near the old ROAS. If cost per result jumped, move half back.
The campaign budget allocator runs this kind of split from your own numbers.
What to automate and what to keep human
| Automate (with approval) | Keep human |
|---|---|
| Search term reviews and negative keywords | Which products or services to advertise |
| Pausing ads and keywords that spend without converting | Offer, pricing and promotions |
| Routine budget shifts between campaigns | Entering or leaving a channel |
| Duplicating winning ads with new copy | Creative concept and brand voice |
| Weekly and monthly reporting | Targets: what ROAS or CPA counts as good |
| Account audits and anomaly flags | Deciding if a flagged anomaly is real |
The rule of thumb: automate anything with a right answer in the data. Keep anything that needs context the account cannot see.
Where AI media buying goes wrong
Most bad outcomes come from a short list of failure modes. Each one has a simple check.
- Optimizing to a broken signal. If the conversion action double counts or fires on page view, the AI will move budget toward whatever inflates it. Check conversion setup before turning on any budget moves, and review it again after site changes.
- Chasing the platform that counts most generously. Platform-reported ROAS from a long view-through window can make one channel look twice as good as it is. Compare against backend orders or analytics totals.
- Too many small edits. Daily nudges keep Meta ad sets in learning and make results impossible to read. Batch changes weekly unless something is clearly broken.
- Cutting the top of the funnel. Prospecting usually shows a worse last-click cost than retargeting or brand search. Cut it to zero and retargeting audiences shrink a few weeks later. Judge prospecting on new customers, not last-click ROAS.
- Ignoring seasonality. A drop in the week after a sale is not a reason to cut budget. Tell the agent about promotions, stockouts and launches.
None of these need a smarter algorithm. They need a person approving changes who knows the business, which is the case for keeping approval on every change.
Evaluation checklist for AI media buying software
- Write access per channel. Ask for the list: which platforms it can change, and which it only reports on. "Supports TikTok" can mean reporting only.
- Approval on every change. Each proposed change should be editable before it applies. Avoid tools that spend autonomously by default.
- Evidence with every proposal. The data it read, the change, and the expected effect.
- Respect for learning phases. Does it batch budget changes, or does it make a dozen small edits a day?
- Cross-channel view. Can it compare Google, Meta and others in one report?
- Account-level depth. Search terms, Quality Score, impression share, audiences, not just top-line spend.
- Pricing model. Flat subscription or a percentage of ad spend. A percentage grows with your budget whether the work does or not.
- Trial on a real account. Give it a problem you already know about and see if it finds it.
Sprites as an AI media buyer
Sprites is an AI marketing agent that works as an AI media buyer across your ad accounts. You connect them, it reads the data itself, and every change arrives as an editable approval card that you approve before anything is applied.
- Google Ads: audits search terms, wasted spend, Quality Score and impression share; creates and edits campaigns, ad groups, ads, keywords and negatives; changes budgets and pauses or enables. See the Google Ads agent.
- Meta: analyzes performance, creates and edits campaigns, ad sets and ads, builds custom and lookalike audiences, checks pixel health, changes budgets. See the Meta ads agent.
- Microsoft Advertising, LinkedIn and Reddit: creates and edits campaigns and ads, and changes budgets and status.
- TikTok: reporting and analysis only.
Pricing is a flat monthly plan, not a share of spend: Build at $99 for one ads agent (Google Ads or Meta), Launch at $495 for Google Ads and Meta, Grow at $985 for all six ads agents. Each starts with a 7-day trial for $1. Details are on the pricing page.
Start by connecting the account with the most spend and asking for an audit. The first proposals show quickly whether the agent sees what your best buyer would.