Skip to content
Live
Vibe Coding Cologne · 2 Sep 2026 · 56 min

Build the ad dashboard the platform cannot give you

Connect ad spend to real sales, turn the result into a daily decision, and understand when AI-generated content actually needs a label.

Ben Sufiani, Gianna, Paul, Marcus, Alexa Scheffler · The article: Vibe Code Your Ad Reporting Stack

A useful dashboard ends in a decision

Paul and Marcus combine Meta campaign costs with the sales that reached Shopify, including the conversions browser tracking misses. Their dashboard calculates a blended return and adds an action beside each campaign.

The output is deliberately plain: scale, watch, or reduce. The dashboard earns its place by shortening the morning decision rather than by showing more charts.

“This column called action basically tells us what to do now – increase the spend or decrease the spend based on these two numbers.”
Paul

The full answer crosses four sources

Ben extends the pattern to a subscription product. Campaign IDs arrive in the first-party database, PostHog describes behavior, Stripe holds revenue, and Meta holds cost.

The agent can join those sources without pretending a cookie captured every sale. That is what lets a B2B product build the same operating view Paul and Marcus use for ecommerce.

“In B2B, you would usually have four sources: the ad engine, your database, your web analytics, and your revenue source like Stripe.”
Ben Sufiani

The automation is trained through daily correction

The reporting loop was not correct on day one. For months, Paul asked for the data, rejected bad joins, and explained what the numbers should mean until the routine became trustworthy.

That history matters. The polished morning action is the result of repeated supervision, not a single prompt that happened to produce a convincing table.

“Every morning I would say the data is bad because of this, and it would fix itself. This is the result of two or three months of work.”
Paul

AI labeling depends on who did what

Alexa separates model providers from people using a finished model, and a stylized generated image from a realistic deepfake of a real person. The obligation changes with the role and the use case.

ReguLite turns that classification into a guided conversation backed by a curated knowledge base. It stores the reasoning so a company or lawyer can inspect how the recommendation was reached instead of trusting a confident chat answer.

“I was getting a wall of text. I just wanted to know: can I use it?”
Alexa Scheffler

Dive deeper

Vibe Code Your Ad Reporting Stack

Read the full reporting method and see how Marcus and Paul supervise the system instead of operating it.

Read the insight
Next Vibe Coding Cologne

Meet builders turning domain expertise into products, dashboards, and working AI systems.

Register free
Every event recording

Watch the full Insights library whenever you need the implementation details again.

See the Pirate Pass