Aditya Sriram recently shared something that caught my attention: "I built an AI that grades my Google Ads campaigns every morning." He followed it with a problem every paid marketer recognizes: "Logging into Google Ads daily to check campaign health is time consuming. And by the time you spot an underperformer, you've already wasted budget."

I like this post because it is not vague "AI will optimize everything" hype. It is a clear operating principle: reduce time-to-detection, make triage automatic, and keep the decision rules deterministic. In other words, wake up to a prioritized to-do list instead of a tab jungle.

Below, I want to expand on what Aditya built, why it works, and how you can adapt the same approach to your own accounts without introducing hallucinations or over-automation.

The real goal: shrink the gap between signal and action

When Aditya says the system runs every day at 9 AM, the point is not the specific time. The point is cadence. Most Google Ads waste does not come from a single catastrophic change. It comes from small drifts: CTR slides, CPA creeps up, conversion volume drops, a budget gets allocated to a campaign that no longer deserves it.

A daily health check is the simplest countermeasure, but it is also the easiest to skip when you are busy. So Aditya automated the habit.

Key insight: you do not need an "AI optimizer" first. You need an "AI auditor" that surfaces what deserves attention.

What Aditya actually built (and why the architecture is smart)

Aditya described a system in n8n that runs automatically each morning. The workflow is modular and that is the right way to think about it.

1) Data fetch: pull the last 7 days of real performance

As Aditya explained, his agent "pull[s] the last 7 days of campaign data - clicks, impressions, conversions, spend" using GoMarble MCP.

The 7-day window is a practical default:

It is recent enough to reflect current performance.
It is long enough to reduce noise for most accounts.

If your account has lower volume, you might shift to 14 days. If you run very high volume and change creative frequently, you might use 3 to 5 days plus day-of-week comparisons.

2) Scoring engine: deterministic math, not vibes

One line in the post should be printed and taped above every marketing dashboard:

"No AI hallucinations here - pure math."

Aditya scores each campaign from 0 to 100 using CTR, conversion rate, CPA, and volume. That combination is powerful because it balances:

Efficiency (CPA)
Effectiveness (conversion rate)
Ability to win auctions and attract clicks (CTR)
Statistical confidence and business impact (volume)

If you want to replicate this, the trick is not picking the perfect formula. The trick is creating a formula that is consistent, explainable, and aligned with how you actually make decisions.

A practical approach:

Define acceptable ranges per metric (for example, CPA thresholds by campaign type).
Normalize each metric to a 0-1 score (cap outliers so one weird day does not dominate).
Apply weights (for example, CPA 40%, conversion rate 30%, volume 20%, CTR 10%).
Convert to a 0-100 grade.

The grade does not need to be "true." It needs to be useful and stable.

3) Smart routing: turn grades into next actions

Aditya then classifies campaigns into tiers:

Excellent (75+) - Ready to scale
Good (55-74) - Test incremental growth
Fair (35-54) - Needs optimization
Underperforming (<35) - Urgent review

This tiering step is more important than it looks. It converts a numeric score into an operating system. Your brain does not want 42 numbers. It wants a small set of decisions.

If you adopt this framework, define what each tier means in your organization:

"Ready to scale" might mean +10% daily budgets with guardrails.
"Test incremental growth" might mean expand keywords, audiences, or add new creatives.
"Needs optimization" might mean tighten targeting, fix landing pages, or refresh ad copy.
"Urgent review" might mean pause, investigate tracking, or isolate the failing segment.

4) Multi-channel alerting: meet people where they work

Aditya routes results into:

Slack and Gmail alerts
Google Sheets logging for historical tracking

This is a subtle content strategy lesson too: the system is designed for distribution. The output is not trapped in a dashboard that nobody opens.

For teams, I recommend:

Slack for daily triage (fast and visible)
Email for stakeholders who want a record
A sheet or database for trending and retrospectives

What to watch out for (so the system does not backfire)

A scoring bot can create false urgency if you do not add a few guardrails.

Low volume can distort grades

A campaign with 1 conversion on $30 spend can look amazing, until it goes to 0 conversions tomorrow. Volume should act as a confidence multiplier. If conversions are below a minimum threshold, consider labeling the campaign "Insufficient data" instead of forcing a grade.

Attribution and tracking issues will masquerade as performance issues

If GA4, Shopify, or your conversion tags break, your best campaign may instantly look like an underperformer. Add a separate check for conversion tracking health (for example, sudden account-wide conversion drops).

One score is not enough for every campaign type

Brand campaigns, remarketing, and prospecting campaigns can have very different CTR and CPA profiles. Consider separate scoring profiles by campaign objective.

A practical blueprint you can copy (without rebuilding everything)

Aditya mentioned sharing a "Ready to import n8n workflow" and a "step by step guide" for connecting n8n, MCP, Google Ads, Slack, and Gmail. Even if you do not use his exact stack, the blueprint is universal:

Step 1: Define your metrics and thresholds

Pick 3 to 5 metrics that reflect success. For most lead gen or ecommerce accounts:

Spend
Conversions
CPA (or ROAS if you have reliable revenue)
Conversion rate
Clicks or impressions (volume proxy)

Step 2: Decide your grading rules

Write them down in plain English first. Example:

If CPA is 20% below target and conversions >= 10, grade increases.
If conversions drop week-over-week and spend is flat, grade decreases.
If CTR drops and impression share is stable, investigate creative.

Then translate into math.

Step 3: Automate the daily run

Schedule the workflow. Pull data. Compute scores. Assign tiers.

Step 4: Alert with context, not just alarms

A good alert includes:

Campaign name
Grade and tier
The 2-3 metrics that drove the grade change
A suggested next action tied to your SOPs

Step 5: Log everything for learning

Aditya logs each run to Google Sheets. That turns a daily check into a dataset. After a month, you can answer:

Which campaigns frequently oscillate between Good and Fair?
Do your optimizations actually move grades?
What is the average time between Underperforming and intervention?

Why this post went viral (and what it says about modern paid marketing)

Aditya's post resonated because it hits three trends at once:

Operators want leverage, not more dashboards.
Teams want reliable automation, not black-box AI.
Everyone is trying to protect budgets in a world where waste compounds fast.

Also, the call to action is concrete: comment "ALERT" and get the workflow, the scoring system, and the templates. That is a strong example of packaging operational know-how into something shareable.

Where to take it next

If you build your own version, a few extensions are worth considering:

Add anomaly detection (spend spikes, conversion cliffs) alongside the deterministic grade.
Track changes (budgets, bids, creatives) so you can tie grade movement to actions.
Add a "scale safely" module that suggests budget increases with caps.
Create weekly rollups that show trend lines, not just daily snapshots.

The main point stays the same: as Aditya Sriram showed, you can use agents to grade and route work, while keeping the scoring logic grounded in math and business rules.

This blog post expands on a viral LinkedIn post by Aditya Sriram, Building GoMarble || AI Agent for paid media marketers; built on your Meta Ads, Google Ads, Shopify, and GA4.. View the original LinkedIn post →