top of page

Agent Recipe 1 — Daily Pacing Monitor

Task
Detect pacing and KPI issues early during active campaigns.


When to use
Daily (or twice per day) for active campaigns.


Inputs

  • Campaign table for today / yesterday / last 7 days

— Campaign name

— Spend
— Impressions
— Clicks
— Installs/Conversion
— CPA/CPI
— ROAS (if available)
— Country
— Channel

​

  • KPI targets:

— Target CPA/ROAS
— Daily budget

​

  • Alert thresholds:

— Overspend >15%
— CPA drift >20%
— Volume drop >25%

​

Expected output

  • Executive summary

  • Priority issues

  • Opportunities

  • Recommended actions with reasons

  • Data warnings

​

QA check

  • Is attribution lag affecting results?

  • Is volume large enough to trust the signal?

  • Is the agent using the right KPI (not vanity metrics)?

​

Next action

Approve 1-2 actions and log them in the daily optimization notes.

Prompt (copy-paste)

Agent Recipe 2 - Creative Research Agent

Task

Generate testable creative angles for a new UA test sprint.


When to use

Before launching new creative tests or when performance starts to slow.


Inputs

  • Product/ app description

  • Audience profile

  • GEOS

  • Campaign goal (install / trial / purchase / deposit)

  • Top-performing creatives (if available)

  • Weak creatives (if available)

  • Notes: what worked / failed before​

​

Expected output

  • 12 creative angles

  • 4 angle buckets

  • Hook lines

  • Suggested formats

  • Risk notes

 

QA check

  • Are the ideas specific to the product and audience?

  • Are they varied (not the same angle repeated)?

  • Any direct copying or generic filler?

​

Next action

Select top 3-5 angles and pass them into creative production.

Prompt (copy-paste)

Agent Recipe 3 – Creative Test Planner Agent

Task

Turn creative ideas into a structured test matrix.


When to use

Once you have angles/ideas but need a clean testing plan.


Inputs

  • 10-12 creative ideas

  • Budget range

  • Channel (Meta / TikTok / etc.)

  • KPI target

  • Test duration

  • Constraints (creative volume, team bandwidth, launch dates)

​

Expected output

  • Test matrix

  • Decision criteria (stop/ iterate/ scale)

  • Priorities and timeline

  • Top risks

 

QA check

  • Are too many variables mixed in one test?

  • Are success criteria clearly defined?

  • Is the KPI realistic for the channel and budget?

​

Next action

Approve the sprint test plan and assign owners.

Prompt (copy-paste)

Agent Recipe 4 - Trafficking QA Agent

Task

Check campaign setup before launch (naming, UTM, tracking, event mapping).


When to use

Before every launch, especially during scale when setup errors get expensive.


Inputs

  • Naming convention rules

  • UTM template

  • Tracking links

  • Event mapping

  • Campaign launch list

  • Channel requirements (if needed)

​

Expected output

  • QA summary

  • Critical/ Medium/ low issues

  • Fix list

  • Launch readiness status

 

QA check

  • Are event names and mappings correct?

  • Are all UTMs and tracking params present?

  • Any legacy templates or broken links? 

​

Next action

Fix critical issues before launch approval.

Prompt (copy-paste)

Agent Recipe 5 - Weekly Reporting Summary Agent

Task

Create a clear, decision-ready weekly performance summary.


When to use

Weekly team reporting, leadership updates, client-facing summaries.


Inputs

  • MMP export

  • Ad platform exports

  • KPI definitions

  • Current week + previous week

  • Business objective (efficiency / scale / testing)

​

Expected output

  • Weekly snapshot

  • Top wins

  • Top issues

  • Opportunities

  • Recommended next steps

 

QA check

  • Are there conflicts across sources?

  • Is the summary actionable (not just descriptive)?

  • Are the KPI definitions consistent?

​

Next action

Send to team/leadership and use it in the next optimization meeting.

Prompt (copy-paste)

Agent Recipe 6 - Fraud / Traffic Quality Triage Agent

Task

Flag suspicious traffic-quality patterns for faster investigation.


When to use

When testing new sources, scaling, or investigating abnormal performance.


Inputs

  • Source-level performance data

  • CTR/CVR/CTIT

  • Geo/placement breakdown

  • Conversion quality signals

  • Benchmarks (if available)

​

Expected output

  • Suspected issues

  • Evidence by source

  • Confidence labels

  • Recommended next steps

  • Missing data list

 

QA check

  • Is there enough evidence?

  • Is low-volume noise being overinterpreted?

  • Are conclusions benchmarked against normal patterns?

​

Next action

Escalate to analyst / AM for validation and action.

Prompt (copy-paste)

Agent Recipe 7 - Budget Reallocation
Suggestion Agent

Task

Suggest conservative budget shifts based on performance.


When to use

During optimization cycles when you need faster budget recommendations.


Inputs

  • Campaign performance data

  • Budget limits

  • KPI targets

  • Business priority (efficiency vs volume)

  • Constraints (minimum spends, partner commitments)

​

Expected output

  • Reallocation suggestions

  • Reasoning with metrics

  • Expected impact

  • Risks/trade-offs

  • Validation checklist before action

 

QA check

  • Is there enough volume to justify a change?

  • Are downstream quality signals considered?

  • Is seasonality or promo context missing?

​

Next action

Approve small shifts first, then re-check performance.

Prompt (copy-paste)

Agent Recipe 8 - Post-Test Insights Agent

Task

Turn test results into reusable learnings for the next sprint.


When to use

After each test cycle (creative tests, audience tests, offer tests).


Inputs

  • Test matrix

  • Results by test cell

  • KPI outcomes

  • Team notes (optional)

​

Expected output

  • Outcome summary

  • Key learnings

  • Reusable patterns

  • Next hypotheses

  • Confidence notes

 

QA check

  • Are conclusions too strong for the sample size?

  • Are “winners" actually statistically meaningful enough for your process?

  • Are next hypotheses practical?

​

Next action

Use the output in sprint planning and update your learnings library.

Prompt (copy-paste)

Starter Setup Checklist

Before using any Al agent, make sure these basics are in place.

1. KPI Rules

  • Primary KPI is defined

  • Thresholds for alerts/actions are set

  • Evaluation window is clear

​

2. Data Rules

  • Source of truth is defined

  • Naming/taxonomy is standardized

  • Reporting cadence is clear

​

3. Guardrails

  • What Al can do is defined (monitor / summarize / suggest)

  • What Al cannot do is defined (launch / pause / move budget)

  • Approval owner is assigned

 

4. QA Process

  • Human reviewer is assigned

  • QA checklist exists

  • Outputs are logged for review

Prompt Adaptation Guide

Use these placeholders to adapt prompts to your stack:
 

  • [KPI] → CPA/ROAS/ volume/ retention proxy

  • [Channel] → Meta / TikTok/ Google/DSP/etc.

  • [Geo] → country/region/ tier

  • [Attribution window]→ 1d/7d/ etc.

  • [Risk threshold] → CPA drift %, overspend %, volume drop %

  • [Goal] → install/ trial/ purchase/ deposit

​

Tip: Keep your prompts consistent across the team.
Small prompt differences create inconsistent outputs.

What Not to Automate Yet

Do not automate these without strong controls and mature QA:


• Final budget approvals
• Source blocking decisions
• Tracking changes
• Campaign launch/pause actions
• Any execution based on incomplete data

Final Takeaway

The fastest wins come from repeatable workflows, not full autonomy. Start with one of these first:

​

1. Daily Pacing Monitor
2. Trafficking QA Agent
3. Weekly Reporting Summary Agent

 

Deploy one workflow, validate it, then expand.

If your team is building these workflows around real campaign execution - especially targeted in-app advertising and performance optimization - BidMatrix can help adapt them to your UA setup. Send us an email to bd@bid-matrix.com

bottom of page