AI Agent Controls: Atlanta Festival’s 2026 Success

If you don’t set up your AI permissions right, the bots will burn through your ad budget. It’s that simple. Defining exactly how much an AI can spend and who it can target, what we call targeting controls, is the line between a profitable campaign and a financial black hole. Let an AI run wild without oversight and you’ll find it blowing your money on audiences that don’t convert or completely misreading your targeting parameters. We saw this happen on a recent campaign, where putting tight controls on spend and targeting was the only thing that turned a struggling project into a profitable one. So, how do you actually implement these controls in a way that works?

Campaign Teardown: “Local Flavors Food Festival”

Let’s walk through a real-world example: the “Local Flavors Food Festival” campaign. The goal was to sell tickets for a food event in Atlanta, Georgia, featuring restaurants and food trucks from within a 25-mile radius of Centennial Olympic Park. The first two weeks, running from March 1st to March 15th, 2026, were a total mess because of sloppy targeting and an AI with way too much freedom on the budget. For the second half of the month, from March 16th to March 31st, we locked things down with strict AI permissions and targeting controls, and the results flipped completely. Here’s a breakdown of what happened in each phase.

Phase 1: Unrestricted Autonomy (March 1st – March 15th, 2026)

The initial plan was basically “let the AI figure it out.” We gave it a general goal of maximizing impressions and clicks with a total budget of $15,000 for the 15-day period ($1,000 per day), telling it to target people on Google Ads and Meta Business Suite interested in “food,” “festivals,” and “local events.” There were no spending caps for specific audience segments and we didn’t give it a real negative targeting list, other than blocking people outside the 25-mile event radius. It was a classic case of giving an algorithm too much trust and not enough direction.

Metrics & Performance (Phase 1)

• Budget Spent: $14,890
• Impressions: 1,250,000
• Clicks: 25,000
• CTR: 2.0%
• Conversions (Ticket Sales): 150
• Cost Per Lead (CPL): $99.27 (based on form submissions for newsletter, not direct sales)
• Cost Per Conversion (Ticket Sale): $99.27
• Return on Ad Spend (ROAS): 0.8x (Ticket price average $80)

The numbers from Phase 1 were brutal. A 0.8x ROAS means you’re losing money on every sale, specifically, 20 cents for every dollar spent on an $80 ticket. The $99.27 Cost Per Conversion was completely unsustainable. Digging in, we saw the AI had dumped almost 40% of the budget into super broad interest groups like “cooking enthusiasts” and “weekend activities,” which delivered tons of impressions but almost no ticket sales. A good chunk of the spend went to users in suburbs who, while technically within the 25-mile radius, showed no real intent to attend a local event downtown. This is a textbook failure of AI agent authority without guardrails.

Phase 2: Granular Controls & AI Permissions (March 16th – March 31st, 2026)

After the disaster of Phase 1, we obviously had to make some big changes. We came back with the same $15,000 budget for the next 15 days, but this time we put the AI on a much shorter leash. We rolled out a new framework for AI permissions and targeting controls with some hard-and-fast rules.

Spending Controls:

1. Daily Segment Caps: We capped the daily spend for each main audience segment. For example, our “Atlanta Foodies” segment (defined by local group memberships and interests) got a $300 daily cap, while the broader “Event Goers” audience was capped at $150. The AI couldn’t go over these limits.
2. Performance-Based Budget Shifts: The AI could shift up to 15% of a daily budget from an underperforming segment to a better one, but only if the target segment was hitting a Cost Per Conversion (CPC) under $40. Any budget shift over that 15% threshold required a person to approve it.
3. Overall Daily Cap: We slapped a hard daily ceiling of $900 on the AI’s spending, which gave us a $100 buffer for manual adjustments or for quickly scaling up a surprisingly good ad.

Targeting Controls:

1. Mandatory Exclusion Lists: We built out long negative keyword lists (e.g., “fast food,” “delivery service,” “recipes”) and added demographic exclusions like people under 21, since the event served alcohol. The AI was not allowed to override these.
2. Affinity Audience Prioritization: Instead of letting the AI guess, we told it to prioritize specific, high-intent affinity audiences we’d identified from past data, like “Frequent Diners” and “Local Event Enthusiasts.” We starved the broader, useless interest groups of budget.
3. Geofencing Refinement: We went beyond the simple 25-mile radius and layered on micro-geofences around dense residential and commercial areas like Midtown, Old Fourth Ward, and Inman Park, where we knew people were more likely to attend. The AI was instructed to spend at least 60% of its budget inside these tight zones.
4. Dynamic Creative Optimization (DCO) Constraints: The AI could still run DCO to test ad variations, but it was restricted to pre-approved brand guidelines and messaging themes. It couldn’t just invent new ad copy out of thin air without a human checking it first.

Metrics & Performance (Phase 2)

• Budget Spent: $14,950
• Impressions: 980,000
• Clicks: 35,000
• CTR: 3.57%
• Conversions (Ticket Sales): 480
• Cost Per Conversion (Ticket Sale): $31.15
• Return on Ad Spend (ROAS): 2.57x

The results were night and day. Yes, total impressions went down, which sometimes scares clients, but that just means we stopped wasting money showing ads to the wrong people. The CTR jumped to 3.57%, which proved the new, tighter targeting was working. The most important numbers tell the whole story: Cost Per Conversion crashed from $99.27 to $31.15, and ROAS shot up to 2.57x, making the campaign profitable. It’s a perfect example of how even a sophisticated AI needs clear boundaries to do its job properly.

What Worked and What Didn’t

The daily caps on each audience segment were a huge part of the turnaround. They stopped the AI from throwing good money after bad on speculative audiences and forced it to optimize where we were already getting traction. The mandatory exclusion lists immediately cut out a huge volume of irrelevant clicks, saving a significant chunk of the budget. Prioritizing our known affinity audiences and using micro-geofencing in places like Midtown also played a big part in reaching the right people. An unexpected learning was that the AI would still try to broaden its targeting within a segment if we gave it an inch, which just reinforces that you have to keep checking its actual audience distribution reports, not just the top-line performance metrics. You have to periodically audit the AI’s choices against your strategic goals, because its idea of a “best performing” segment might not actually align with your business objectives.

Optimization Steps Taken

After Phase 2, we built these lessons into a standard process. A human analyst now does a weekly audit, reviewing the AI’s targeting and spend allocation against our actual conversion data in Google Analytics 4. This lets us spot subtle drift, like when the AI starts expanding a geographic radius a little too far. We also created a feedback loop where human-approved negative keywords and audience exclusions get uploaded to the AI’s parameters automatically every 48 hours. We’ve even started A/B testing different levels of AI autonomy for budget shifts, comparing a 10% daily allowance against 20%, to find the right balance. This ongoing process lets us keep tuning the AI’s efficiency without completely neutering its optimization ability.

Relying on an AI’s “black box” optimization is a recipe for burning through your ad spend. AI has incredible processing power and speed, but it lacks the intuition and strategic context a human brings to nuanced market dynamics. Your job is to define the canvas and the color palette, then let the AI paint within those specific boundaries. Without clear AI permissions for spend and precise targeting controls, you’re effectively handing your company credit card to an algorithm that has an incomplete picture of your business goals. That’s a risk few businesses can afford, especially in a competitive market like event promotion in Atlanta.

The future of effective digital advertising is this careful dance between AI autonomy and human governance. It’s about using the AI’s power for massive-scale execution and optimization, while your team provides the strategic direction and financial guardrails to keep it on track. The “Local Flavors Food Festival” campaign is a clear example of how getting this balance right can transform a campaign from a costly mess into a highly profitable success.

Setting up strong AI permissions and targeting controls helps AI agents perform at their peak by keeping them focused on what matters. This hands-on governance ensures your marketing budgets are invested wisely which leads to better campaign performance and a strong return on advertising spend.