Day 5 presentation1 / 16

Zero to MVP AI Bootcamp

Workflow & quality

Connect the business workflow, test the AI systematically, and make failures visible before launch.

Today you ship

An automated lead journey with test checkout, measured AI quality, and live alerts.

Scroll to move through the presentation.

Learning objectives2 / 16

By the end of today

  1. 1Trigger a reliable n8n follow-up from a verified product event.
  2. 2Add a clearly labeled Stripe test deposit without moving real money.
  3. 3Make repeated events safe so actions are not duplicated.
  4. 4Run a repeatable ten-case AI evaluation and improve the weakest result.
  5. 5Confirm uptime and application errors reach the instructor-approved alert channel.
Run of show3 / 16

Today’s learning path

  1. Module 140 min

    From product event to follow-up

  2. Module 235 min

    Test checkout

  3. Module 345 min

    Evaluate the AI

  4. Module 435 min

    Monitoring and recovery states

  5. Guided build

    AI Lead Assistant · Stage 5

    Build, test, checkpoint, and ship.

Module 1 · Learn · 40 min4 / 16

From product event to follow-up

Automation begins when something meaningful happens in the product. A qualified lead event can notify the business, reply to the visitor, and record whether each action succeeded.

A real example

A lead receives a website-audit recommendation. n8n emails the visitor with next steps, alerts the agency, and records that both messages were sent.

Technical terms

Event
A meaningful occurrence, such as a completed qualification.
Workflow
A sequence of automated steps triggered by an event.
Module 1 · Apply it5 / 16

Apply the concept

Trace one qualified lead through business notification and customer follow-up.

  1. 1Define the completed-lead event and minimum safe data.
  2. 2Receive and validate the event in n8n.
  3. 3Map the fields into business and visitor messages.
  4. 4Record success or failure and test safe replay.

Use this prompt

Design an n8n workflow for a qualified lead. Show trigger, validation, field mapping, business notification, visitor reply, status recording, and failure path. Minimize the personal data sent between steps.

Success looks like

One completed lead creates one visible workflow execution and the two expected follow-up messages.

Watch for

  • If fields are missing, inspect the trigger payload before changing downstream steps.
  • If a retry sends duplicates, add an event identifier and check whether it was processed already.
Module 2 · Learn · 35 min6 / 16

Test checkout

Stripe test mode lets us build the real payment journey using fake cards and zero real money. The server creates checkout, and a verified Stripe event confirms completion.

A real example

A qualified visitor pays a test deposit to book the recommended consultation. The product marks the lead as deposit completed only after Stripe confirms it.

Technical terms

Test mode
A safe payment environment that uses fake money and test card details.
Webhook
A signed notification sent from one service to another after an event.
Module 2 · Apply it7 / 16

Apply the concept

Map the shared test deposit's safe checkout, success, and cancellation paths.

  1. 1Create the test product and price in Stripe.
  2. 2Create checkout from protected server code.
  3. 3Build clear success and cancellation screens.
  4. 4Verify the Stripe event and update the lead only once.

Use this prompt

Add a clearly labeled Stripe test deposit to the existing lead journey. Create checkout server-side, verify the completion webhook, prevent duplicate processing, and do not collect card details in our own form.

Success looks like

A Stripe test card completes the checkout, one verified event updates the correct lead, and no real money moves.

Watch for

  • If the success page appears but the lead is unchanged, inspect webhook delivery rather than trusting the browser return.
  • If one payment creates repeated actions, store and reject already processed event IDs.
Module 3 · Learn · 45 min8 / 16

Evaluate the AI

An AI feature needs repeatable product tests. The same inputs should be scored before and after a change so improvement is measured rather than guessed.

A real example

A strong result recommends an approved service, cites relevant answers, uses the required structure, and safely redirects unsupported requests.

Technical terms

Evaluation
A repeatable test that compares AI behavior with expected behavior.
Rubric
The rules used to score quality consistently.
Module 3 · Apply it9 / 16

Apply the concept

Run ten cases, find the most damaging pattern, make one fix, and compare scores.

  1. 1Prepare ten normal, unclear, unusual, and off-scope lead cases.
  2. 2Record the expected behavior before running the AI.
  3. 3Score relevance, grounding, structure, and fallback behavior.
  4. 4Fix one failure pattern and rerun the unchanged test set.

Use this prompt

Score these AI Lead Assistant results from 0–2 for relevance, grounding in approved services, required structure, and safe fallback. Identify the single repeated failure pattern; do not suggest multiple changes.

Success looks like

The second run shows a documented score improvement without creating regressions in previously passing cases.

Watch for

  • Do not change the test cases after seeing poor results; that hides the failure.
  • If scores depend on personal opinion, make the rubric more observable and specific.
Module 4 · Learn · 35 min10 / 16

Monitoring and recovery states

Monitoring tells you when the product fails even when you are not watching. Recovery states tell the visitor what happened and what safe action they can take next.

A real example

If the AI provider times out, the visitor keeps their answers, sees a retry option, and the owner receives an error alert without private lead content.

Technical terms

Uptime
Whether the public product is reachable and responding.
Error monitoring
A service that records and groups application failures.
Module 4 · Apply it11 / 16

Apply the concept

Trigger one harmless test error and confirm the alert reaches the chosen phone channel.

  1. 1Monitor the public product address on a schedule.
  2. 2Capture useful application errors without secrets or private content.
  3. 3Define which failures deserve a phone alert.
  4. 4Trigger a harmless failure and verify both alert and visitor recovery.

Use this prompt

Review this lead journey for loading, timeout, service failure, empty result, and retry behavior. Propose messages that preserve the visitor's answers and explain one safe next action.

Success looks like

A test failure reaches the chosen alert channel, and the visitor can retry without re-entering all information.

Watch for

  • If alerts contain private answers or keys, remove that data before continuing.
  • If every small error triggers a phone alert, reserve urgent alerts for failures that block the core journey.
Guided project12 / 16

Build brief

AI Lead Assistant · Stage 5

Complete the business follow-up loop and produce the evidence needed to trust the product before deployment.

User story

As a business owner, I receive a qualified lead automatically and can trust the AI result and operational status.

Guided build · Part 113 / 16

Build it step by step

1

Trigger n8n

Send the minimum qualified-lead event into the shared workflow.

2

Complete follow-up

Notify the business, reply to the lead, and record workflow status.

3

Add test checkout

Connect a clearly labeled Stripe test deposit and verified completion event.

Guided build · Part 214 / 16

Build it step by step

4

Run the AI evaluation

Execute all ten cases and score relevance, grounding, format, and safety.

5

Fix the top failure

Change the prompt, business context, validation, or interface once and rerun the cases.

6

Prove monitoring

Trigger safe uptime or application failures and verify useful alerts and recovery states.

Ship checkpoint15 / 16

Do not ship until

  • One completed lead triggers the expected business and customer follow-ups.
  • Repeating the same event does not duplicate the workflow or test payment outcome.
  • No secret or private lead content appears in source code or unsafe logs.
  • The ten-case evaluation records improved before-and-after scores.
  • A test outage or application error reaches the selected alert channel.
Wrap and prepare16 / 16

Ship it. Show it. Prepare the next move.

Before the next day

  • Confirm Google Cloud access and the instructor-provided deployment path.
  • Write a two-minute demo script around the complete lead journey.
  • Prepare one backup recording or screenshot for every risky live-demo step.
All presentationsOpen Day 6