3 Ways Enterprise Retailers Are Using AI to Fix Paid Media Waste

Most large retailers are wasting part of their paid media budget. Believing that is the easy part(some will even admit it).

This piece is about three specific places where AI is recovering wasted ad spend right now  each one backed by real numbers from retailers operating at serious scale. No theory. No "imagine if." Just what's working in production today. Here's the shift worth paying attention to in 2026: AI has finally moved to production.

First, why is paid media waste so hard to see

At enterprise scale, waste doesn't show up as one big, obvious problem. It hides in plain sight.

You're running hundreds of campaigns across a dozen channels. Creative is produced faster than anyone can quality-check it. Budgets shift between platforms based on rules written months ago. And your attribution model   if you're honest about it can't reliably tell you which touchpoint actually drove the sale. So spend keeps flowing toward whatever looked good in last week's report, not what's working right now.

Reporting tells you what already happened; it doesn't change anything in real time. Closing the gap takes systems that can act, not just observe, enter AI.. 

Here are the three places it's making the biggest difference.

1. Smarter creative, produced at a scale humans can't match

Creative is both a bottleneck and a quiet cost center. Generic or underperforming ad creative burns impressions you've already paid for, and your design team can only produce so much of it. So most retailers run fewer variations than they should, test less than they'd like, and leave performance on the table.

Generative AI changes the equation. The right Gen AI system produces high-performing, on-brand ad creative automatically   at volumes no human team could reach.

The clearest example we've seen in production is GenAds, a generative AI tool we built with Mercado Libre and AWS. Tens of thousands of smaller sellers on the platform couldn't afford professional design work, so their ads underperformed. GenAds takes a plain product photo and returns a polished, contextual banner   automatically, every week, across tens of thousands of advertisers at once. The result was a 25% average lift in click-through rate, with fresh creative generated continuously rather than in occasional batches.

CMOs Takeaway: better creative stops being a headcount problem. AI turns it into a capability that scales with your catalog instead of your design budget.

2. Budgets that defend themselves in real time

When you're managing hundreds of campaigns, something is always overspending while your team sleeps. ROAS erodes in the gaps between reporting cycles   a campaign saturates, a channel underdelivers, and by the time anyone notices, the money's gone.

This is where AI does what spreadsheets and weekly reviews simply can't: it forecasts marginal performance, reallocates budget across channels in real time, and pauses what's bleeding   continuously.

Mercado Libre invests more than $1.2 billion a year in marketing. We built an AI-powered media optimizer (the engine behind our Mixilo product) that decides how that spend is allocated across campaigns and channels to hit target ROAS. The machine learning models forecast performance using saturation curves, then adjust bids and budgets automatically. The outcome: a 9.6% increase in ROAS, alongside a 55% reduction in time spent on manual campaign management.

On a budget that size, every point of ROAS is a meaningful amount of money put back into the business. And the second number matters just as much as the first   when AI handles the constant monitoring, your team stops babysitting dashboards and starts doing the strategic work that actually moves revenue.

3. Automating the operational grind that slows everything down

Some of the most expensive waste in paid media isn't in the spend at all   it's in the operation around it. Manual creative uploads, platform-by-platform campaign setup, and geotargeting that has to be reconfigured by hand every time priorities shift. It's slow, repetitive, and impossible to scale without  hiring more people.

AI agents and automation handle that grind end-to-end, so your team's capacity stops being the ceiling on how fast you can move.

Rappi runs paid media across nine countries. Their marketing team was manually uploading every creative into Meta, TikTok, and Google Ads, following different rules for each. We built Megatron, a production-grade automation platform that handles creative deployment and dynamic geotargeting automatically   including an optimization engine that translates business-defined geographies into platform-compatible targeting. The impact: 5x more creative deployed at the same headcount, 83% faster time-to-market, and a 60–70% reduction in manual operational effort.

The point worth sitting with: this isn't about replacing your marketers. It's about freeing them from work that was never a good use of their time in the first place.

The thread connecting all three: measurement you can trust

As impactful as these solutions are, they will only go so far if you're still struggling to identify what worked. if you can't tell what actually worked.

Smarter creative, real-time budgets, and automated operations all depend on a feedback signal that's honest about which touchpoint drove the conversion. Last-click attribution isn't that signal; it overcredits the final ad and quietly misleads every optimization decision downstream.

This is why we built a deep-learning multi-touch attribution model for Mercado Libre that measures the true incremental lift of each touchpoint across the customer journey, accounting for channel interactions and timing. Better measurement (MTA done right) is the foundation the other three plays stand on: it points your optimization at what genuinely drives revenue, not at what merely happened to be last in line.

For marketing leaders building or rethinking their martech stack, this is the part that's easy to skip and expensive to get wrong.

The one question that separates production AI from demo AI

The distance between a polished demo and a system managing billions of dollars in live ad spend is enormous, and it's exactly where most "AI for retail" falls apart. A demo handles the happy path. Production handles the edge cases, the platform limits, the fallback rules for when the model can't decide, and the monitoring that catches a problem before it costs you money.

That's the bar we built Muttdata to clear. The three results cited in this post aren’t projections. 

• +25% CTR
• +9.6% ROAS

These numbers are real results. They reflect the impact of real systems running today for some of the largest retailers in LATAM.

Meet us at eTail Boston (Aug 10–12)

Hobson Sherrill from our team will be on the floor. If paid media efficiency is on your 2026 roadmap, grab 15 minutes with him here.

Not attending eTail? No worries, you can contact us here. 

You can also explore the full case studies behind these numbers: Mercado Libre, GenAds, and Rappi.