AI for Event Management: How Hackation Runs Its Operations

Case note: Hackation is our own recurring event series. We run it as a real-world proof of a principle we also implement for clients: AI can carry the operational backbone of recurring events, and the operator can own that stack.

Situation

Recurring events have a structural problem: operations depend on the founder. Invitations, RSVP tracking, reminders, the sponsor pipeline, invoicing, follow-ups and recap content are individually small but collectively a full-time job, and they repeat for every single event. As long as that knowledge lives in one person’s head, the format doesn’t scale, and every event costs the same manual energy as the first.

Hackation sat exactly at that point: a single-operator format (install parties, hackathons, vibe-coding workshops) that should no longer depend on one person’s availability.

Approach

We treated event operations not as a software project but as an operating system: the set of workflows, data structures, SOPs and agent roles that make the format runnable without founder memory.

Concretely, we run Hackation on:

• A lean agent team (around six roles: leadership, engineering, revenue operations, community, finance, orchestration) built on Paperclip. Each role has a tool budget and a monthly budget: total agent operating cost sits in the low double-digit euros per month.
• Codified SOPs instead of ad-hoc instructions. The install-party SOP, for example, is a full checklist from D-7 (set the frame) to D+1 (convert and learn), with ready-made invitation, reminder and follow-up templates.
• Structured data instead of scattered markdown: sponsor pipeline, attendees and lifecycle segments in a queryable database the agents connect to.
• Our own self-hosted GPU node plus staging (1 node / 1 GPU) as the foundation for the workloads we deliberately keep inside our own boundary. We sell the method and the outcome, not borrowed scale.
• Hard HITL (human-in-the-loop) gates: external email sends, publications, contract signatures and financial transactions all require human approval. Agents draft, humans send.

The agents handle drafting (invitations, reminders, LinkedIn, recaps), triage (lead qualification), content packs and attendee ops. Relationship work, day-of facilitation and every approval stay with the human.

Outcome

At a real install party (10–25 people, two parallel tracks), operations ran measurably along the SOP:

• 16 of 21 check-ins, a 76% show rate, tracked without heavy ticketing (below 30 people, deliberately no Pretix; RSVPs tracked manually).
• Follow-ups within 24 hours to every attendee, sent using pre-built, typed templates from the SOP, without the founder writing each message individually.
• D-1 and same-day reminders visibly reduced no-shows.

The decisive point isn’t a single metric. It’s the property of the system: operational load shifts from the founder to a documented, agent-supported system, and the format can repeat without quality hanging on one person.

Transferable lesson

AI doesn’t replace judgment: it replaces repeated manual execution. The leverage sits in three decisions:

1. SOP before automation. Codify the recurring flow as a checklist first, then let agents execute against it. Automation without an SOP is just faster chaos.
2. Structured data as the keystone. Without queryable pipeline and attendee data, no agent can act reliably. It’s the #1 bottleneck, and the first step.
3. HITL in the right places. Clear gates for anything that goes outward or moves money. That makes delegation safe and keeps the operator in control of their own stack.

That’s exactly the pattern we implement for clients in AI Acceleration Sprint and Event Ops with AI Agents: not a borrowed platform, but an operational stack the client owns and controls.

Next step

Running a recurring program (events, onboarding, community, reporting) where operations hang on too few people? Let’s talk about your specific operational bottleneck.

Request a contact / intro call →

Related topics: Event Ops with AI Agents: What Can Actually Be Automated | Self-hosted LLMs for the Mittelstand