We Asked an AI Agent to Optimize Our Marketing Campaigns. It Asked Us Questions First.

The Usual Failure Mode

Every marketing team has done this: hand raw campaign data to an AI tool and ask, "What should we do differently?"

The AI responds instantly with generic advice: - "Consider increasing spend on your top-performing channel" - "Test new creative variations" - "Optimize for mobile"

Technically reasonable. Strategically useless. Because the AI didn't know: - That the "top-performing channel" is already at budget ceiling - That the creative was refreshed last week - That 80% of the audience is already on mobile

The advice sounds intelligent but lacks context. And in marketing, context is everything.

We wanted an agent that understood this principle. So we built one that doesn't advise until it asks.

Build an agent that interviews before it advises.

Download the Campaign Optimization Interview Loop TACT Blueprint.

Get the Blueprint →

The Interview: A Simulated Conversation

Here's what happens when the Campaign Optimization Interview Loop processes a batch of campaign data. Instead of jumping straight to recommendations, it runs a structured interview with the marketing manager.

🤖 Agent:

I've ingested the weekly campaign data from your SharePoint folder (5 campaigns across SG, MY, and TH markets). Before I generate recommendations, I need to clarify a few things:

Question 1: The Singapore Instagram campaign shows a 340% increase in spend from last week. Was this an intentional budget surge (e.g., product launch), or does this look like an anomaly?

👤 Marketing Manager:

That's intentional — we launched the new product line on Monday.

🤖 Agent:

Noted. I'll weight the SG Instagram campaign as a launch phase and compare it against launch benchmarks rather than BAU metrics.

Question 2: The Thailand LinkedIn campaign has no conversion tracking data for the past 3 days. Is this a known tracking issue, or should I flag it as a data gap?

👤 Marketing Manager:

Known issue — the UTM parameters broke on Thursday. We're fixing it.

🤖 Agent:

Understood. I'll exclude the affected 3-day window from the Thailand LinkedIn performance analysis and note the gap in my report.

Question 3: You have two campaigns targeting the same audience segment in Malaysia (Facebook and Google). Is this intentional (cross-channel reinforcement) or an overlap that should be resolved?

👤 Marketing Manager:

Intentional — we're testing which channel drives better conversion for that segment.

🤖 Agent:

Got it. I'll run a channel comparison analysis for that specific segment and include a recommendation on budget allocation between the two.

Thank you. I now have sufficient context to generate your optimization report.

Why the Interview Changes Everything

That 90-second exchange prevented three analysis errors:

Every wrong assumption an AI makes without context leads to a wrong recommendation. The Interview Loop eliminates the most dangerous assumptions before a single recommendation is generated.

The TACT Architecture: Proactive Interviewing

T — Trigger: Event-Based (Weekly Campaign Data Ingestion)

Trigger: When an item is created or modified Configuration: Weekly ingestion of marketing campaign data into the designated SharePoint folder

The agent activates each time new campaign data arrives — typically on a weekly cadence aligned with reporting cycles.

A — Agent: Campaign Optimization Interview Loop (Orchestrator Type)

The agent runs three stages:

Stage 1 — Data Normalizer: Cleans and standardizes campaign data from different regional formats. Google Ads exports look different from Meta Business Manager exports. The Normalizer brings them into a common schema.

Stage 2 — Interviewer: This is the signature capability. The Interviewer scans the normalized data for ambiguities, anomalies, and missing context. For each finding, it formulates a specific question for the marketing manager. It pauses the workflow until the human responds.

Stage 3 — Strategic Advisor: Armed with clean data and verified context, the Strategic Advisor generates performance analysis and optimization recommendations.

System Prompt:

You are a Campaign Optimization Interview Loop agent. Your workflow:

1. Collect Input Data: Gather all relevant source data, documents, and information.
2. Consolidate & Structure: Organize and standardize the collected data.
3. Analyze & Process: Execute proactive interviewing. Actively pause to ask the user for missing context before outputting localization strategies or recommendations.
4. Validate Results: Review the processed output for accuracy.
5. Distribute Output: Format the final results and share with stakeholders.

C — Connectors & T — Tools

The Final Output: Context-Enriched Recommendations

After the interview, the Strategic Advisor produces a report that is fundamentally different from what a context-blind AI would generate:

Campaign Optimization Report — Week of 17 Feb 2026

Agent: TACT Interview Loop v1.2 | Context interviews completed: 3 questions

---

Singapore — Instagram (Product Launch Phase) - CPM: $4.20 (vs. launch benchmark $5.50 — outperforming) - CTR: 2.8% (vs. launch benchmark 2.1% — excellent) - Recommendation: Maintain current spend for the remainder of launch week. Reduce to BAU budget (-60%) starting Monday. The launch metrics justify the surge investment.

Thailand — LinkedIn - Note: 3-day data gap (17-19 Feb) excluded due to UTM tracking failure - Adjusted CPL (excluding gap): $12.40 - Recommendation: After tracking is restored, re-evaluate over a full 7-day window. Do not make budget decisions based on incomplete data.

Malaysia — Facebook vs. Google (Same Segment Test) - Facebook: CPA $18.20, ROAS 3.1x - Google: CPA $24.50, ROAS 2.4x - Recommendation: Facebook is outperforming Google for this segment by 29% on CPA. Propose shifting 40% of Google budget to Facebook for next week's A/B continuation.

The Principle: AI Should Ask Before It Answers

Most agentic AI workflows are designed to run autonomously from start to finish. This one is deliberately designed to pause.

The pause isn't a limitation — it's the feature. In marketing, the gap between data and strategy is filled by context that only humans possess: intent, history, organizational knowledge, and upcoming plans.

The Interview Loop bridges that gap. It gives you AI that's smart enough to know what it doesn't know — and humble enough to ask.