VeraConnect Marketing Pipeline

The problem

VeraConnect's customer success team was running marketing campaigns manually for each new client: interpreting analytics dashboards, deciding what to send, drafting emails, scheduling sends, and then waiting to see if it worked. The loop was slow and the quality varied by which team member picked it up.

What I built

A closed-loop agent system, end-to-end inside VeraConnect's platform:

• Ingestion. Pulled user activity data from the platform's existing event store and normalized it into a per-account activity timeline.
• Reasoning. A planner agent that takes the timeline plus the client's stated KPI and proposes a campaign strategy: segments to target, messaging angles, send cadence.
• Generation. A copy agent that turns the strategy into concrete email variants, with subject lines and body copy tuned per segment.
• Execution. Integration with VeraConnect's existing email service provider to schedule and send the campaigns.
• Feedback. The pipeline reads back its own send metrics and uses them to inform the next round of strategy. The "closed loop" is the part that actually moves KPIs.

What was hard

The hardest part was not the agent orchestration. It was defining a clean boundary between what the planner was allowed to decide (strategy) and what the human-in-the-loop still had to approve (final copy, send timing for sensitive accounts). I went through three iterations of this contract. The version that worked framed the planner as a recommender with explicit confidence scores, and treated approval as a fast path for low-confidence cases rather than a checkpoint for every send.

The second hardest part was making the agent's reasoning inspectable. Marketing teams do not trust a black box that "did something." Every decision the pipeline makes is logged with the inputs, the alternatives it considered, and the reasoning trace, so a human reviewer can audit any individual send in under a minute.

What's next

Marketing is a much better proving ground for agent systems than software engineering, because the loop is faster and the metric is unambiguous. A campaign runs in days; you find out whether the agent worked. Code refactors run in months; nobody knows. If you are building an agent system and you want to know whether the architecture is right, start with marketing-shaped problems.

The other lesson was about scoping a summer internship. Three months is enough to ship something a real team uses daily, but only if you spend the first two weeks watching the team use the current tool before writing a line of code. I did, and it saved me from building a feature that would have been technically neat but operationally unused.