AI Marketing Systems

Build tools around the recurring work

As Blaxel's first marketing hire, I owned the strategy and budget while also doing the work. I built tools for the recurring reviews: what was happening in paid acquisition, what was changing in the market, and which sources were shaping answers about the category.

Each tool had a specific input and a decision to support. Paid review brought together delivery and product outcomes. Market research connected a market read to its sources. Answer-engine citation analysis helped identify conversations worth reviewing.

Review paid performance daily

The paid-growth system combined Google Ads delivery, PostHog product activity and Stripe payment outcomes. AI-agent workflows synthesized conversion data and queued proposed optimization changes, moving review from monthly to daily.

I kept campaign changes behind human approval. A proposed change had to pass validation, match the approved change set and have a rollback or compensating action. Low-volume evidence remained inconclusive.

Paid growth / Review workflow

From paid signals to proposed changes

Review cadence

Monthly → Daily

Delivery

Google Ads

Spend, clicks and ad conversions

Product

PostHog

Product activity and conversion behaviour

Payments

Stripe

Payment outcomes

AI synthesis

Findings + proposed changes

Bring the evidence into review. Keep low-volume findings inconclusive.

Human decision

Approval before campaign changes

Validation, a valid approval matching the change set, and a rollback or compensating action.

That made the automation useful for preparation and review while keeping responsibility for spend explicit. The paid-growth page covers the program's results and measurement boundaries.

Make market research traceable

The Market Signal dashboard kept each assessment linked to its sources. The workflow collected and scored material, then checked the sources behind the claims before including them in the dashboard.

Find the conversations behind AI answers

I also built a citation-review tool around Profound's answer-engine data. It identified Reddit conversations for review using three signals: frequently cited threads, topics where Blaxel was less visible and threads gaining citation share.

The tool turned those signals into a weekly queue. It helped direct attention to the source material influencing answers, with human review before any participation. It didn't automatically post into the conversations.

Community Citation Radar / Queue design

Three signals shape the weekly queue

Profound citation data, with Reddit thread context.

Highest cited

Up to 3 review slots

What already shapes the answers?

Threads with strong citation activity in the category.

Underrepresented

Up to 3 review slots

Where is Blaxel less visible?

Cited conversations in topics with a visibility or representation gap.

Newly climbing

Up to 3 review slots

What is gaining attention?

Threads gaining citation share across the latest complete week.

9 threads at most

One weekly review queue

Each selected thread appears once. Read the source and decide whether there's a useful contribution to make.

Participation stays human-reviewed.

This was part of the answer engine optimization work: understand what the engines were citing, inspect the source and decide whether there was a useful contribution to make.

Give the tools shared context

The go-to-market (GTM) knowledge graph held positioning, customer context, messaging, proof and writing guidance in versioned files. It distinguished source-backed claims from inferences and drafts, with review rules for deciding what was ready to use.

That gave the work a common reference. A launch draft, a paid review and a positioning decision could draw on the same product and customer context while preserving the evidence each task needed.