5.10 — Final workshop: ''The pitch'' 🔥

Goal: turn your track’s tool (concept → prompt → agent → secured agent) into a one-page business case, ready to present. It’s the final deliverable of your track.

🟢 In plain terms — a good one-pager answers 5 questions: the problem (with figures), the stack, the ROI (hours saved × hourly cost − agent cost), the guardrail (the sensitive action + the HITL), and the human angle (what the team gains by doing thanks to the agent). Estimate the ROI with demo 5.6.

🎯 The block below frames the pitch for your track (chosen on the home page).

🧠 Your documents copilot

• Problem: time lost searching for info and answering the same questions.
• Measure: sourced-answer rate, refusal quality, review time and actual monthly cost versus your baseline.
• Reassuring guardrail: sourced answers, HITL on sensitive cases (no made-up info). 👉 Deliverable: a one-pager that defends your documents copilot.

📥 Email sorter

• Problem: a generic inbox that swallows hours every week.
• Measure: classification accuracy, human correction time and actual hours saved versus your baseline.
• Reassuring guardrail: nothing goes out without validation (HITL before send). 👉 Deliverable: a one-pager that defends your email sorter.

🎯 Lead machine — the culmination of your track

• Problem: manual prospecting, slow and irregular.
• Measure: accepted qualification rate, correction time and downstream conversion on a limited sample.
• Reassuring guardrail: HITL on sending, GDPR/consent, no false promises.
• Your complete tool: enrich → qualify → reach out → follow up (assembled in Levels 3-4). 👉 Deliverable: the complete “lead machine” + its one-pager (decision-maker pitch).

📄 Document extractor

• Problem: slow manual data entry, prone to errors.
• Measure: field-level accuracy, rejection rate, correction time and total cost versus manual processing.
• Reassuring guardrail: human validation before accounting entry; GDPR on personal data. 👉 Deliverable: a one-pager that defends your extractor.

✍️ Content factory

• Problem: producing enough content, fast enough, without repeating yourself.
• Measure: review time, factual corrections and publishable-output rate versus your baseline.
• Reassuring guardrail: human review before publication (brand voice, accuracy). 👉 Deliverable: a one-pager that defends your content factory.

🎯 Your own project

1. The problem — what business pain does your agent solve? (with figures if possible.)
2. The category — does it automate, transform or create? Which business area?
3. The stack — orchestration (Make/n8n), base (Airtable), assistant (ChatGPT/Claude/Gemini).
4. The ROI — hours saved × hourly cost − agent cost (demo 5.6 + traces from Level 4).
5. The guardrail — the sensitive action and where you place the HITL. 👉 Deliverable: a one-pager that defends your agent in front of a decision-maker.

🚀 Deployment plan (all tracks): pilot on a reduced scope → measure (via the traces) → expand. And keep the human angle: what higher-value task does your team gain thanks to the agent?

✅ Final deliverable: a one-pager that defends your tool in front of a decision-maker — the culmination of your track.