Solutions Showcase: part 5 of 7

The complete text of the site in 7 parts. Parts:

Part 1 has the contents list of every section and case study.

Case studies in full

AI-Powered Construction Report SaaS (MeinBüromeister)

Turning construction-site voice notes, photos, and messages into structured reports through an automated AI-powered backend for German trade companies.

Section: Custom SaaS Platforms

Overview

MeinBüromeister is a B2B SaaS platform for German construction and trade companies. It allows workers to submit job updates through WhatsApp instead of preparing every report manually. The backend organizes incoming information, generates structured reports, and asks workers to review them before submission.

The Business Problem

Construction workers often send separate voice notes, photos, and text corrections from job sites. Office staff must identify the correct customer and project, organize materials and working hours, and format the information into professional reports. This creates repetitive administrative work and makes report preparation more time-consuming. The goal was to reduce this manual effort while keeping workers involved in reviewing and approving their reports.

The Solution

An n8n-based backend receives WhatsApp messages and groups related inputs into a single session. OpenAI Whisper transcribes voice recordings, while Claude AI extracts relevant report information into structured data. After the worker reviews and approves the report, the backend stores it in Supabase for delivery to the office dashboard. Supporting workflows automate customer onboarding, subscription management, team access, and error handling.

Project Snapshot

A production backend connecting WhatsApp messaging, AI-powered report extraction, worker approval, report storage, customer onboarding, and subscription lifecycle management.

How the System Works

System Architecture

The backend is organized around an n8n automation layer that sits between WhatsApp, the AI services used for transcription and extraction, and the Supabase/Stripe infrastructure that powers the office dashboard.

Supporting Integrations

Core Automation Workflows

The backend is built around interconnected n8n workflows that automate WhatsApp message handling, AI-powered report generation, customer onboarding, and subscription management. Each workflow handles a specific responsibility and integrates with the relevant APIs and backend services to support the complete reporting lifecycle. Some main automation workflows are shown below.

1. Core Onboarding & Workflow Management

This workflow handles company onboarding, setup-code verification, worker registration, subscription checks, report approval, project management, and notifications through interconnected n8n automation nodes.

Image: Core onboarding and workflow management: company setup, worker registration, subscription checks, report approval, project handling, and notifications across 15 interconnected automation groups.

2. AI-Powered Report Processing Workflow

This workflow processes WhatsApp messages using Whisper and Claude AI to extract structured report data, enrich customer and project details, validate reports, and handle database updates and notifications.

Image: AI-powered report processing: message intake, voice transcription, Claude-based data extraction, report generation, and save/notification handling.

Product Experience

Report inbox

Image: The office inbox organizes incoming reports by date, technician, customer/project, category, and status.

Customers and projects

Image: The project view groups customers and highlights open-report counts and recent work.

WhatsApp work history

Image: Workers review and approve each AI-generated report directly in WhatsApp before it goes to the office.

Image: Once approved, the report is saved, confirmed back to the worker on WhatsApp, and delivered to the office as a PDF.

Core Backend Capabilities

Results & Metrics

The project delivered five production workflows connecting WhatsApp, AI-powered report extraction, Supabase, and Stripe. Across five rounds of end-to-end testing, more than 25 bugs were identified and fixed, with over 30 test scenarios executed per round.

Technical Challenges & Solutions

During development and end-to-end testing, several integration and AI reliability issues required targeted solutions.

1. Stripe Subscription Data Handling

Challenge

The Stripe checkout event did not always contain all the subscription information required by the workflow.

Solution

Retrieved the complete subscription details directly through the Stripe API before continuing with subscription processing.

Outcome

Improved the reliability of subscription data handling.

2. Subscription Plan Mapping

Challenge

Mapping subscription plans based on payment amounts could lead to incorrect plan identification.

Solution

Used Stripe product IDs to identify plans instead of relying on monetary amounts.

Outcome

Made plan identification more consistent across subscription events.

3. AI Hallucination Control

Challenge

Claude AI could occasionally introduce materials or details that were not present in the worker’s input.

Solution

Refined the AI prompt with guardrails to discourage unsupported details and keep extraction grounded in the supplied information.

Outcome

Improved the consistency of structured report generation.

4. Supabase PATCH Response Handling

Challenge

Some Supabase PATCH operations returned empty responses, creating difficulties for downstream workflow nodes.

Solution

Explicitly referenced upstream node outputs where required instead of relying on the PATCH response to provide all subsequent data.

Outcome

Made downstream data handling more reliable.

5. End-to-End Integration Testing

Challenge

The system depended on multiple connected services, including WhatsApp, AI models, Supabase, and Stripe, making integration failures difficult to isolate.

Solution

Conducted five rounds of end-to-end testing, covering more than 30 scenarios per round, and fixed over 25 identified bugs.

Outcome

Improved integration reliability across the production workflows.

Key Takeaways & Future Development

Key Takeaways

Closing Summary

MeinBüromeister demonstrates how workflow automation, AI-powered information extraction, and integrated SaaS infrastructure can simplify construction reporting. The primary contribution was engineering the backend workflows and integrations that connect field communication with structured reports, subscription management, and the office dashboard.

Page · Plain text

AI-Powered 24/7 Voice Support Agent for eCommerce Stores

A 24/7 AI voice agent that answers store calls using your product and policy data, escalating only complex calls to humans.

Section: n8n Automations, AI Agents, AI Agents for Ecommerce

Project Overview

The AI-Powered 24/7 Voice Support Agent automates customer support calls for eCommerce stores using conversational AI and a centralized knowledge base. Built around VAPI, n8n, Slack, and Google Sheets, it can answer common questions about products, pricing, shipping, returns, and store policies through natural voice conversations.

The system understands customer intent, retrieves relevant store information, responds in real time, and supports optional escalation for complex or high-risk issues. This reduces repetitive work for support teams while providing customers with access to assistance around the clock.

Technologies: VAPI · n8n · Slack · Google Sheets · AI Voice Agent

01. The Problem: Support Calls Don’t Scale

Before automation, customer support teams handled incoming calls manually.

Common challenges included:

Impact per team

Core challenge: Build a 24/7 support system that can resolve common customer queries using the store’s own knowledge while escalating issues that require human attention.

02. Results After Automation

After implementation, the voice agent can answer calls around the clock and handle common customer questions automatically.

Key improvements

Time and effort savings

Before

After

Estimated monthly impact

Consistency improvements

03. The Solution: AI-Powered Voice Agent

The system acts as a virtual customer support representative capable of handling common eCommerce questions independently.

It:

04. How the System Works

Phase 1: Knowledge Base and Store Data

Relevant store information is collected and organized for retrieval during calls.

This may include:

The knowledge base should be kept up to date to reduce the risk of outdated responses.

Phase 2: Incoming Call and Intent Detection

  1. A customer calls the support number.
  2. VAPI answers with a configured greeting and voice.
  3. The customer's speech is converted into text.
  4. AI identifies the intent, such as a product question, order inquiry, or policy clarification.

Phase 3: Knowledge Base Lookup

The system retrieves relevant information to answer the customer's question.

For example:

Phase 4: Voice Response Generation

Phase 5: Multi-Turn Conversation

The agent maintains context throughout the call, allowing customers to ask follow-up questions about products, orders, and policies without restarting the conversation.

Phase 6: Human Escalation and Call Summary

When configured, complex or high-risk calls can be forwarded to a human agent. The workflow may also support callback requests or ticket creation.

After the call:

05. Business Value

06. Expected Impact

Time Savings

Automating repetitive calls reduces the time support agents spend answering routine questions, allowing them to focus on complex issues.

Error Reduction

Using maintained product and policy data helps reduce inconsistent or improvised answers.

Call Handling Speed

Common questions can potentially be resolved in one or two exchanges, depending on their complexity.

Knowledge Utilization

Centralized store information can be reused across voice interactions, improving consistency and accessibility.

Estimated time savings: Approximately 40–60 hours per month per support team, depending on call volume and automation coverage.

07. Workflow Architecture

The main components work together as follows:

08. How to Replicate the System

Step 1: Configure the VAPI Voice Agent

Step 2: Configure the Call-End Webhook

Step 3: Import the n8n Workflow

  1. Open ShopTalk AI – Call Summary & Lead Capture (1).json.
  2. Import the workflow into n8n.
  3. Connect the VAPI webhook to the workflow.
  4. Configure Slack and select the appropriate summary channel.
  5. Connect Google Sheets and select the spreadsheet and required columns.

Step 4: Test the Flow

Step 5: Activate the Workflow

Once testing is complete, activate the n8n workflow and verify that the VAPI webhook is configured correctly.

The voice agent can then handle customer calls continuously, while n8n processes call summaries and related follow-up actions.

09. Final Outcome

The AI-powered voice support system transforms repetitive customer support calls into a more consistent and scalable operation.

The result: A scalable, 24/7 voice support experience that reduces repetitive work while keeping human agents available for the issues that need them most.

Page · Plain text

ClickUp: Automating Operations for Veterinary Specialist Partners

ClickUp setup used to automate operations for a veterinary specialist partnership.

Section: ClickUp Setups

ClickUp: Automating Operations for Veterinary Specialist Partners

A centralized ClickUp operations system integrating appointment scheduling, patient information, lab results, reminders, and referral coordination for Veterinary Specialist Partners (VSP).

Category: ClickUp Setups Technologies: ClickUp, Calendly, Jotform, Gmail, RingCentral, Slack

1. Project Overview

Veterinary Specialist Partners (VSP) operates veterinary facilities in Louisville and Evansville, providing specialized care for pets with complex medical conditions. Managing appointments, patient information, lab communications, and follow-ups manually created administrative overhead and made coordination across locations more difficult.

The solution used ClickUp as a central operations workspace, connecting scheduling and communication tools through automated workflows. Appointments, related information, reminders, cancellations, and follow-up activities could be managed with greater consistency.

2. The Challenge

Before the implementation, several operational processes depended on manual work.

Fragmented Appointment Management

Appointments were booked through external scheduling tools and needed to be organized across different locations and appointment types. Staff had to maintain accurate records and keep appointment statuses up to date.

Incomplete Patient Information

Required forms, medical details, and laboratory results could arrive through separate systems and email communications. Locating the relevant information alongside an appointment required additional coordination.

Manual Reminders and Follow-Ups

Patient reminders, form-completion checks, and post-appointment follow-ups required staff attention. Inconsistent follow-up processes could result in missed administrative steps.

Cancellation and Record Synchronization

When appointments changed or were cancelled, the corresponding records and lists needed to remain consistent across the scheduling system and ClickUp.

3. Project Objectives

The project focused on the following objectives:

4. The Solution

4.1 Calendly to ClickUp Integration

Calendly bookings are captured and converted into ClickUp tasks.

Image: Calendly-to-ClickUp appointment automation screenshot from Make.com

This setup reduces the need to manually transfer appointment information between systems.

4.2 Jotform Integration and Patient Information

The workflow monitors required Jotform submissions associated with scheduled appointments.

This helps staff identify outstanding administrative requirements before appointments take place.

4.3 Laboratory Email and Attachment Management

Gmail integration brings relevant laboratory communications into ClickUp.

Incoming lab emails and attachments are added to the corresponding patient or appointment task, making related information easier to locate without manually searching through email conversations.

4.4 Appointment Cancellations

Cancellation handling keeps appointment records aligned between Calendly and ClickUp.

Image: ClickUp custom-field automation screenshot.

This provides a clearer distinction between upcoming and cancelled appointments.

4.5 Centralized Appointment Lists and Synchronization

A master Appointments list provides a consolidated view of bookings from Louisville and Evansville.

Custom-field changes are synchronized across the relevant location-specific and central lists, helping maintain consistent appointment information for teams working across departments.

4.6 Automated Appointment Reminders

RingCentral is integrated to send SMS reminders for upcoming appointments.

Image: Ring Central SMS reminder workflow screenshot.

Automated reminders help patients prepare for scheduled visits and reduce the amount of manual communication required from administrative staff.

4.7 Past Appointment Cleanup

Appointments whose dates have passed are automatically removed from the relevant active lists.

This keeps the operational workspace focused on upcoming appointments and outstanding tasks rather than outdated records.

5. Workflow

This workflow connects scheduling, information management, communication, and appointment status updates within the ClickUp operations workspace.

Image: The broader Louisville and Evansville recheck workflow screenshot.

6. Tools and Integrations

ToolPurpose
ClickUpCentral appointment lists, tasks, custom fields, and workflow automation
CalendlyAppointment booking and cancellation events
JotformPatient forms and medical information collection
GmailLaboratory communications and attachments
RingCentralAutomated SMS appointment reminders
SlackSupporting team communication and coordination

7. Outcomes and Business Value

Improved Operational Efficiency

Appointment creation, cancellation handling, record synchronization, and past-appointment cleanup reduce repetitive administrative work.

Better Information Organization

Centralized appointment lists connect scheduling records with relevant patient information and laboratory communications, making information easier for staff to find.

More Consistent Patient Communication

Automated reminders and form-completion notifications support more consistent communication before scheduled visits.

Improved Cross-Location Visibility

The central Appointments list provides a consolidated view of bookings across Louisville and Evansville, supporting teams that need visibility across both facilities.

More Maintainable Operations

Custom fields, dedicated lists, and defined automation rules provide a structured foundation that can be refined as operational requirements change.

8. Technical Highlights

9. Future Improvements

Potential next steps include refining automation rules based on staff feedback, expanding reporting on appointment statuses and follow-up completion, and reviewing notification workflows to ensure they continue to meet the needs of both locations.

10. Conclusion

The ClickUp implementation brings key operational processes for Veterinary Specialist Partners into a more connected system. By integrating appointment scheduling, patient forms, laboratory communications, SMS reminders, cancellations, and centralized tracking, the solution reduces repetitive administrative work and helps teams maintain more consistent records across Louisville and Evansville.

The result is a structured operations workspace designed to support appointment coordination, information accessibility, and ongoing workflow improvement.

Page · Plain text

PeakNIL Athlete Content Automation

Two n8n workflows that find athlete news, draft the post, and publish it to Instagram once a person approves.

Section: n8n Automations

Project Overview

PeakNIL connects college athletes with brands for Name, Image, and Likeness (NIL) opportunities. To attract brand partnerships and maintain a consistent social media presence, the team needs a steady stream of posts highlighting athlete achievements across 28 sports.

This project uses two connected n8n workflows to automate the process. The first monitors athlete news, evaluates stories, drafts captions, and sends content for review. The second detects the manager’s email response, generates a branded graphic, and publishes the post to Instagram after approval.

Human approval remains part of the publishing process, helping ensure that the final content aligns with PeakNIL’s brand and content standards.

Technology stack: n8n, Claude, Google Sheets, Gmail, Placid, Instagram.

The Business Problem: Scaling Athlete Coverage

PeakNIL manages hundreds of athletes across 28 sports. Finding relevant stories and turning them into social media content requires ongoing research, writing, image selection, graphic design, and publishing.

The main challenges include:

The objective was to reduce repetitive work while improving consistency and retaining human control over publication.

The Solution

Two connected automations handle the work, and a person approves the result.

Workflow 1: Athlete Monitor (finds and drafts)

Image: Workflow 1, Athlete Monitor: a schedule trigger reads the roster from Google Sheets, searches for news, has Claude draft the post, and emails it to the content manager through Gmail.

Workflow 2: Reply Detection (approves and publishes)

Image: Workflow 2, Reply Detection: a Gmail trigger reads the manager’s reply, then one branch builds the graphic with Placid and another publishes to Instagram after approval

How the System Works

Step 1: Read the Athlete Roster

The Athlete Monitor workflow retrieves athlete records from Google Sheets and checks whether each record contains the information needed for monitoring.

Athletes with incomplete data or who were scanned within the previous three days are skipped.

Step 2: Discover Recent News

For eligible athletes, the workflow searches trusted sports sources for relevant news published within the last three days.

This creates a pool of recent stories that can be evaluated for potential social media content.

Step 3: Evaluate and Draft Content

Claude evaluates each story for suitability, assigns a content category, and generates a short headline and caption.

The content must follow PeakNIL’s positive, brand-safe approach, avoiding injuries, losses, and controversies.

Step 4: Send the Draft for Review

The workflow finds a suitable athlete photo and sends the proposed content to the content manager by email.

The manager can review the story and select an image or provide an approval response.

Step 5: Generate the Branded Graphic

The Reply Detection workflow reads the manager’s response and uses Placid to create a branded graphic from the selected photo and headline.

The graphic is sent back for a final check before publishing proceeds.

Step 6: Publish and Update Tracking

After approval, the workflow publishes the post to Instagram and updates the athlete content tracking sheet in Google Sheets.

This closes the loop between discovery, content creation, review, and publication.

05. Workflow Architecture

Workflow 1: Athlete Monitor

Weekday schedule → Google Sheets athlete roster → Eligibility checks → Recent news search → Claude story evaluation and caption writing → Athlete photo selection → Gmail review email

Workflow 2: Reply Detection

The first workflow prepares content for review. The second uses the manager’s response to continue the process and publish only after approval.

Content Rules and Human Approval

The automation is designed to maintain a positive and brand-safe social media presence.

Story Selection

Claude evaluates whether a news story is worth posting and assigns it a content category before generating the draft.

Brand Safety

The content rules exclude stories focused on injuries, losses, or controversies. This helps keep the proposed content aligned with the project's positive athlete-celebration approach.

Human Review

The content manager receives the draft by email and provides a response. The workflow then generates the branded graphic and sends it for a final check.

Publishing remains approval-gated: the workflow is designed to publish to Instagram only after the required human approval.

07. Integrations and Their Roles

IntegrationResponsibility
n8nOrchestrates both workflows and connects the services.
Google SheetsStores the athlete roster and tracks content activity.
ClaudeEvaluates news stories and drafts headlines and captions.
GmailDelivers draft emails and detects manager responses.
PlacidGenerates branded social media graphics.
InstagramReceives approved social media posts.

Each service has a defined role, allowing the workflow to move content between research, writing, review, design, and publishing.

08. Results and Business Value

The project is designed to reduce the manual effort required to create athlete-focused social media content while improving consistency across the roster.

Delivered Capabilities

Expected Business Value

The project description does not provide measured time savings, publishing volume, engagement improvements, or brand-partnership conversion figures. These should be added after reviewing actual workflow and Instagram performance data.

Key Takeaways

Future Development

Final Outcome

PeakNIL Athlete Content Automation connects athlete news discovery, AI-assisted drafting, email-based review, branded graphic creation, and Instagram publishing in two n8n workflows.

The system is designed to help PeakNIL maintain a consistent flow of athlete-focused content while reducing repetitive production work. By keeping human approval in the publishing process, it balances automation with editorial oversight and brand safety.

Page · Plain text

WidowWise Platform Automation, Data Integrity & Dashboard Modernization

Repairing and expanding a connected learning platform with automated enrollment synchronization, accurate reporting, and certificate delivery.

Section: Make Automations

1. Overview

Wings for Widows is a Minnesota-based nonprofit that provides free one-to-one financial coaching to widows through a network of approximately 150–170 volunteer CFP advisors. Its digital learning operations run through two distinct programs: WidowWise University (WWU), which provides free and scholarship-funded courses, and Advising Through Widowhood (ATW), a separate paid financial advisor certificate program.

The platform uses Teachable for course delivery, Airtable for operational data and dashboards, Make.com for email and certificate workflows, and Mailchimp for student and advisor communications.

The engagement focused on reverse-engineering and repairing an existing automation system, resolving recurring data and dashboard problems, improving enrollment synchronization, and expanding reporting and certificate automation.

Client: Wings for Widows Platform: WidowWise / WidowWise University Additional Program: Advising Through Widowhood (ATW) Industry: Nonprofit · Financial Education · Professional Training

2. The Problem

The existing system had recurring synchronization and reporting problems that affected enrollment records, dashboard accuracy, and student completion workflows.

The main challenges included:

3. How It Works

The solution connects the learning platform, operational database, automation workflows, and communication system.

Step 1 — Teachable to Airtable Synchronization

Teachable provides the course, student, enrollment, completion, and transaction information used by the operational system.

Airtable serves as the central data layer, containing 20 tables covering people, enrollments, completions, products, lessons, modules, transactions, prices, coupons, graduates, and automation errors.

Scheduled JavaScript automations synchronize Teachable information into Airtable.

Step 2 — Enrollment Matching and Data Integrity

The original enrollment-matching logic used a combination of course and user identifiers. This could cause a new enrollment cycle to overwrite an earlier record.

The matching logic was rebuilt around each enrollment's actual Person and Product relationships.

Additional JavaScript and Airtable field-formatting issues were identified during testing and corrected before production.

Screenshot to add: Teachable enrollment records alongside their corresponding Airtable enrollment records. Highlight the Person and Product relationships and the corrected matching logic.

Step 3 — Engagement Status and Dashboard Accuracy

Dashboard totals were not reconciling because some incoming Teachable records had blank Engagement Status fields.

The synchronization process was updated so Engagement Status is automatically derived from Percent Complete on every sync run.

This removed the need for repeated manual corrections to dashboard records.

Screenshot to add: Airtable dashboard showing Total, Completed, In Progress, and Not Started, alongside the Engagement Status and Percent Complete fields.

Step 4 — Graduation and Certificate Automation

A certificate automation pipeline was built around the student's completion status.

When Engagement Status changes to Completed:

  1. A sequential Certificate ID is generated.
  2. Certificate and mailing details are sent to C&M Printing.
  3. Chris receives the CFP continuing-education filing information.
  4. The process is gated to prevent certificates from being issued when registration data is incomplete or mismatched.

The workflow also addressed real completion-record issues, including students stuck at 99% and certificate records requiring manual intervention.

Screenshot to add: Make.com certificate automation scenario, the generated Certificate ID in Airtable, and an example of the certificate notification workflow.

Step 5 — Dashboard Expansion

Several dashboards were built or improved to provide more reliable operational reporting.

Screenshot to add: The ATW Enrollments dashboard, All Courses dashboard with its course selector, and ATW Progression dashboard.

Step 6 — Automation Ownership and Reliability

The Teachable API credential previously belonged to the original developer, creating an operational dependency.

A new independently owned API secret was created, and the synchronization automation was repointed to use it.

This restored the client's control over the integration and reduced dependency on the previous developer.

Screenshot to add: Teachable API integration configuration and the relevant Make.com or Airtable automation scenario. Do not expose API keys or credentials.

4. The Result

The engagement transformed a recurring maintenance problem into a more reliable learning and reporting system.

Enrollment synchronization was rebuilt around actual student and course relationships, preserving enrollment history and reducing the risk of duplicate records. Dashboard calculations were corrected, and Engagement Status is now automatically derived from completion data during synchronization.

Certificate issuance was automated through sequential Certificate IDs, print-partner notifications, and completion checks. Additional dashboards improved visibility into course enrollment, advisor progression, and student completion.

The platform's API ownership was also transferred away from the previous developer, giving the client independent control over the integration.

5. Results & Metrics

The source documents specific implementation outcomes rather than a complete before-and-after performance study.

AreaDocumented outcome
Airtable operational database20 tables
Enrollment synchronizationRebuilt around Person and Product relationships
Engagement StatusAutomatically derived on every sync
Dashboard accuracyMetric-card calculations corrected
Certificate workflowSequential Certificate ID and automated notifications
Dashboard expansionATW Enrollments, All Courses, and ATW Progression
API ownershipNew independently owned Teachable API secret

6. Tech Stack

7. Where This Pattern Applies

This pattern is useful for organizations that rely on multiple connected platforms to manage people, enrollments, learning progress, and communications. By improving synchronization, data integrity, and workflow automation, they can build a more dependable operational system around their existing tools.

Page · Plain text

From Manual Calendar Checking to Automatic Tutor Matching

Finding the right tutor for a family used to mean checking the tutor sheet and then each tutor's calendar, one by one. Now a staff member posts the request in Slack, and a ranked list of tutors who are really free arrives in the same thread within minutes.

Section: n8n Automations

From Manual Calendar Checking to Automatic Tutor Matching

Works with: n8n, Slack, Skyvern, Google Sheets, TutorBird, Claude

Before A tutor sheet and 56 separate calendars to check for every request

Now One ranked report in Slack, in about 2 to 4 minutes

Project overview

A tutoring company places children with tutors. For every family request, one question matters most: which tutors are free at the times this family needs, and right for this child?

Answering it means combining two sources: a tutor sheet (who is available now, where they work, which grades, what experience, how many students) and a separate calendar in TutorBird for each of 56 tutors.

Growlyze built a Slack bot that does this check. A staff member posts the request in a fixed numbered format. The bot shortlists the right tutors from the sheet, has a browser automation read their live TutorBird calendars, matches their free time to the family’s windows and session length, and replies in the same thread with a ranked report.

The bot also handles the family’s time zone, applies the company’s rule for how many sessions a week a tutor must cover, and says why a tutor is only a partial match.

Key results at a glance

MetricBeforeAfter
Time to a match report6–8 minutes2–4 minutes
Calendars read per requestAll 56 tutorsOnly the shortlisted tutors (24 in the first live test)
Tutor informationA sheet plus 56 separate calendarsOne ranked report in Slack
Name matchingTwo name formats: “First Last” in the sheet, “Last, First” in TutorBirdAll 34 tutors marked available matched to their TutorBird names
Temporary page errorsRe-run by handRetried automatically, up to 3 attempts

Report times come from measured live runs (2:00 and 3:54). The calendar figure comes from the first live test, a remote request with 24 shortlisted tutors. No other numbers are claimed.

01. The business challenge

When a family asks for a tutor, speed and accuracy both matter. A family left waiting, or promised a time that can’t be kept, can lose trust quickly. Matching a request to the right tutor involves several connected problems.

1.1 Every request has many conditions

A match has to fit the child’s grade, any experience the family needs (such as ADHD or autism), a city or online, the session length (for example 45 or 60 minutes), how many sessions a week, and the family’s free windows in their own time zone. All of these have to be true at once.

1.2 The answers live in two places

The tutor sheet says who is available now, where they work, which grades they teach, what experience they have and how many students they already have. Their actual free time sits in a separate TutorBird calendar for each of 56 tutors, which has to be opened through the website.

Image: The TutorBird calendar, where each tutor’s lessons and free time live. Student and tutor names are blurred.

1.3 Time zones and names don’t line up

Families write their free times in their own zone (ET, MT, PT and others), while TutorBird shows times in Eastern. The sheet writes names as “First Last”, while TutorBird writes “Last, First”, and some names have a middle name or a two-word surname.

1.4 Free time isn’t always usable time

A tutor can be free for 30 minutes inside a family’s window when the session needs 45 or 60. A request for two sessions a week also needs two separate days, not one long gap.

1.5 A wrong match is costly

If a tutor looks free and isn’t, the family is promised a time that can’t be kept. A child waits longer for help, and the team has to start the search again.

02. The solution: a Slack tutor-matching bot

Growlyze built the bot around the company’s own tutor sheet, TutorBird calendars and Slack channel. It replaces none of them. It connects them, so one Slack message sets the whole check in motion.

2.1 A request in Slack

A staff member posts the request as eight numbered lines: student, remote or address, backup option, session length, how often, the family’s free times, grade, and any experience needed. The bot adds a “working on it” reaction and checks every line; if one is missing, it replies in the thread and says which. AI turns the days named in the request, even a range like Monday to Friday, into real dates for the upcoming Sunday-to-Saturday week.

Image: An example request posted in Slack. The sender, student name and address are blurred.

2.2 A shortlist from the tutor sheet

The bot reads the tutor sheet and keeps tutors who are marked available now, work in the right city or remote, and teach the child’s grade. AI then checks which of them have the experience the family asked for, matching on meaning (“Autism” and “Autism Spectrum Disorder” count as the same condition) and requiring every condition named. For in-person requests, the bot also works out how far each tutor is from the family.

2.3 A live calendar read with Skyvern

The bot sends the dates and the shortlisted tutors, written the way TutorBird writes names, to Skyvern. Skyvern logs into TutorBird, selects those tutors and reads each tutor’s free times for the requested days from that tutor’s own calendar column. The workflow checks every 30 seconds until the read is done, and if TutorBird has a temporary page problem it retries up to two more times.

2.4 Matching free time to the family

AI turns the family’s free-text availability into clear day and time windows, including “either day” options. The bot matches sheet names to TutorBird names, converts times from TutorBird’s Eastern time into the family’s time zone, and keeps only the free time that fits a window and is long enough for the session length.

2.5 A ranked report in the same thread

The bot then checks how many sessions a week each tutor can cover. Tutors who cover every requested session are full matches; those who cover fewer are partial matches, with the reason and any too-short slots listed; the rest are named as having no availability. The top pick is chosen from the full matches (closest for in-person requests, then fewest students), and remote backups are added when too few in-person tutors fit. The report is posted as a reply to the original message, with a check-mark reaction.

Image: An example report. Student and tutor names, and the logo, are blurred.

2.6 Clear failure messages

If a step fails, the bot replies in the thread with “Match failed for this request” and the reason, and marks the message with an X. Temporary calendar errors are retried first, so the team only sees an error when every attempt has failed.

03. End-to-end workflow

The whole process runs as one n8n workflow, started by the Slack message.

Image: The n8n workflow, from the Slack message to the posted report.

  1. Request posted in Slack
  2. Dates and fields read from the request
  3. Shortlist built from the tutor sheet
  4. Calendars read by Skyvern, with automatic retry
  5. Free time matched to the family’s windows and session length
  6. Weekly frequency checked and tutors sorted
  7. Top pick chosen
  8. Report posted in the same thread

There is one path from the Slack message to the posted report, and any failure is reported back to the same thread.

04. What makes this solution different?

Built around the company’s own rules

The bot uses the company’s grade bands, experience matching and student counts. When a family accepts “1x or 2x a week”, a full match must cover the higher number, so tutors who can only do one session show as partial.

Reads only the tutors who could match

The sheet filters run first, and only the shortlisted tutors’ calendars are read. In the first live test that was 24 tutors instead of all 56.

AI where judgment is needed, fixed rules where precision is needed

AI reads the family’s free-text availability and matches experience and names. Time-zone conversion, overlap, session-length checks and sorting follow fixed rules.

Partial matches explain themselves

Every partial match says why it falls short (for example, one session of two) and lists any slots that were too short for the session.

Each tutor is read from their own column

Skyvern takes each tutor’s availability from that tutor’s own calendar column, and the read is set up to check that the number of tutors matches the number of columns.

Temporary errors fix themselves

If TutorBird has a temporary page problem, the bot retries up to two more times before showing an error.

05. Business impact

Match reports in 2–4 minutes

Measured live runs finished in 2:00 and 3:54. Before tuning, matches had slowed to 6–8 minutes; removing an unused calendar-read step and checking for results every 30 seconds instead of every 2 minutes brought them back down.

Fewer calendars to read

Only the shortlisted tutors’ calendars are read. In the first live test, that was 24 tutors instead of all 56.

Less manual re-running

Temporary TutorBird page errors used to mean re-running the request by hand. The bot now retries up to two more times on its own. This has been tested with simulated failures, and the first live retries are still to be observed.

Clearer shortlists

Each report names the full matches, the partial matches with the reason, and the tutors with no availability, so the team can see why someone was or wasn’t suggested.

Accuracy checked by hand

In two test requests, all 21 and all 24 tutors in the reports were checked against the raw calendar data and the tutor sheet, and every time range and experience label matched. All 34 tutors marked available in the sheet matched their TutorBird names.

06. Technology stack

TechnologyRole in the solution
n8nWorkflow automation: runs every step, from request to report
SlackWhere requests are posted and where the report is replied
SkyvernBrowser automation: logs into TutorBird and reads the tutors’ free times
TutorBirdThe calendar system the team already uses
Google SheetsThe tutor roster: availability, location, grades, experience, student count
Claude (Anthropic)AI for dates, free-text availability, experience matching, name matching and ZIP estimates
Zippopotam.usTurns ZIP codes into locations for distance checks

The technologies support the bot’s main functions: request intake, shortlisting, calendar reading, matching and reporting. The team keeps working in Slack, TutorBird and their own tutor sheet.

07. Future enhancements

The following items were identified as future improvements rather than completed capabilities.

Warn when a tutor can’t be found in TutorBird

Today a tutor the calendar reader can’t find shows up as having no availability. The report should say “not found” instead.

Refresh and retry inside the calendar reader

Retrying inside the Skyvern run itself would let a temporary page error recover in seconds, instead of starting the whole read again.

An error message for every failed step

Some steps, such as the AI steps, don’t report a failure to Slack. Adding that would make sure no request is ever left without a reply.

08. Where this pattern applies

This isn’t really about tutoring.

Worked example: a home care agency. A family asks for a caregiver who speaks Spanish, lives nearby, and is free Tuesday and Thursday afternoons. The sheet shortlists caregivers who are active, certified and in the right area. A browser bot reads only those caregivers’ schedules in the scheduling system, keeps the ones free at the family’s times, and posts a ranked list in Slack.

Where else it fits:

The shape: a request comes in, a sheet shortlists who qualifies, a browser bot reads the live calendars, and a ranked answer lands where the team already works.

What changes from case to case: the filters in the sheet, the calendar system being read, and the rules for what counts as a match.

Conclusion

Matching a family to a tutor depends on a tutor sheet and 56 separate TutorBird calendars, each with its own times, time zones and name formats.

Growlyze connected them with a Slack bot that shortlists tutors from the sheet, reads the shortlisted tutors’ live calendars, converts times to the family’s zone, and replies with a ranked report that explains every partial match.

Measured reports arrive in 2–4 minutes, and temporary calendar errors are retried automatically. The team keeps using Slack, TutorBird and their own tutor sheet.

Page · Plain text

Automatic Cleanup of Scheduled Posts When a Card Is Unapproved

An n8n workflow that checks every 15 minutes for cards that are no longer approved and removes their still-scheduled posts from Metricool, without touching anything already published or about to go out.

Section: n8n Automations

Project Overview

This project automates the cleanup of scheduled social media posts using n8n, Airtable, and Metricool. When a content card is denied, returned to review, or has a scheduling failure, its previously scheduled posts may still be waiting to publish.

The workflow checks Airtable every 15 minutes for cards with statuses such as Pending Review, Denied, Schedule Failed, or a blank status. It retrieves the associated Metricool post IDs, checks each post's current state, and applies safety rules to determine whether it should be deleted, retained, cleared from the card, or checked again later.

The workflow also updates Airtable with the results, clears obsolete post IDs, records timestamped audit entries, and removes the scheduled time when a post is deleted so the content can be scheduled again after approval.

Technology stack: n8n, Airtable, Metricool, X, Facebook, LinkedIn, Threads.

01. The Business Problem: Unapproved Content Can Still Be Published

A content team manages social media scheduling through Airtable cards and Metricool. Each card stores the Metricool post IDs for its scheduled posts across supported social platforms.

If a card's approval status changes after scheduling, the posts already queued in Metricool may still be published. Without an automated cleanup process, the team must manually locate and unschedule each post.

The main challenges include:

The goal was to make scheduled-post cleanup automatic while maintaining strict safeguards around publishing status and timing.

02. The Solution: Automated Unscheduling with Safety Checks

A single n8n workflow runs every 15 minutes and searches Airtable for cards that may require cleanup.

The workflow separates each eligible card into individual Metricool post IDs and checks each post sequentially, with a short pause between API calls. A decision step determines whether each post should be deleted, kept, cleared, or retried during a later run.

Only future, unpublished posts that meet the deletion criteria are sent to Metricool's delete endpoint. Published posts and posts due within the next five minutes are protected. Posts that are already overdue but are not marked as published are also left untouched for manual checking.

After processing, n8n updates the corresponding Airtable card, clears IDs for deleted or already-missing posts, records timestamped decisions in the Unschedule Log, and clears the Scheduled Time when a post is deleted.

If a lookup or deletion fails, the workflow leaves the relevant record untouched so that a later run can retry it.

03. Project Snapshot

ItemDetail
Platformn8n
ScheduleEvery 15 minutes
Source of approval statusAirtable
Social scheduling platformMetricool
Supported networksX, Facebook, LinkedIn, Threads
InputAirtable cards and Metricool post IDs
Decision logicStatus, publishing state, and scheduled time
Processing methodOne post ID at a time
Audit trailTimestamped Unschedule Log in Airtable
Error handlingFailed lookups and deletions remain available for retry
StatusLive and running on an automatic schedule

04. How the System Works

Step 1: Find Cards Requiring Cleanup

Every 15 minutes, the workflow searches Airtable for cards with a status of Pending Review, Denied, Schedule Failed, or blank, provided they still contain a Metricool post ID.

This identifies content that may no longer be eligible for publication but still has scheduled posts associated with it.

Step 2: Split Cards into Individual Posts

A code node converts each eligible card into one item per Metricool post ID.

The workflow checks its processing log and skips IDs that have already been handled and recorded. This reduces unnecessary repeated processing of settled posts.

Step 3: Retrieve Each Post from Metricool

The workflow requests the current details of each post from Metricool, processing one post at a time with a short pause between requests.

If an API lookup fails, the error does not stop the entire workflow. The affected post is left untouched so the workflow can try again during a subsequent run.

Step 4: Apply the Safety Rules

A decision step evaluates the retrieved post information and assigns an action:

An overdue post that is not marked as published is not automatically deleted. Instead, the workflow records that someone should check its status in Metricool.

Step 5: Delete Eligible Posts

The Delete? branch sends a deletion request to Metricool only for posts marked for deletion.

Posts assigned any other action bypass the deletion request and continue through the merge step. This keeps the delete operation separate from the logic that decides whether deletion is safe.

Step 6: Update Airtable and Record the Outcome

After the deletion and non-deletion branches merge, a code node prepares one update per Airtable card.

The workflow:

Airtable stores the resulting updates, giving the team a record of the cleanup decisions.

05. Workflow Architecture

The workflow has four main functional layers:

This separation helps ensure that a post is not deleted simply because its Airtable card has an unsuitable status. Its individual publishing state and scheduled time must also satisfy the safety rules.

n8n Workflow

Image: The workflow is live and includes protections for published content, posts about to go out, and operations that fail. Its next improvement opportunity is to measure cleanup performance and monitor unresolved cases over time.

06. Safety Rules and Deletion Logic

The workflow applies the following rules before deleting a scheduled post.

Post conditionActionReason
Future post, not published, and more than five minutes from its scheduled timeDeleteEligible for automatic cleanup.
Already publishedKeepPublished content must not be deleted by this workflow.
Scheduled within the next five minutesKeepAvoid interfering with content about to go live.
Past its scheduled time but not marked as publishedKeep and flag for manual checkingAvoid making an unsafe assumption about its status.
Metricool no longer has the postClear the stale post IDThe post is already gone, so the ID no longer needs to remain on the card.
Lookup or deletion request failsLeave unchanged and retry laterPreserve the record for a subsequent attempt.

The same safety rules apply across all supported networks, including X.

07. Airtable Synchronization and Audit Logging

Airtable remains the central record of approval status and scheduled-post identifiers.

After processing, the workflow synchronizes the card with the actual cleanup outcome.

Airtable field or recordPurpose
Card statusIdentifies content that may require unscheduling.
Metricool post IDsTracks the scheduled posts associated with the card.
Unschedule LogRecords timestamped actions and decisions.
Scheduled TimeCleared when a post is deleted so it can be scheduled afresh.

The log provides visibility into what was deleted, retained, or cleared, along with the time of the action. Failed operations remain available for retry rather than being treated as successful cleanup.

08. Reliability and Error Handling

The workflow is designed to continue processing other posts when an individual operation encounters an error.

Sequential API Requests

Metricool is called one post at a time, with a short pause between requests. This reduces the risk of overwhelming the service when many cards require cleanup.

Failure Isolation

A failed lookup or deletion does not stop the entire workflow. The affected post is left untouched, allowing a later run to retry it.

Previously Handled IDs

IDs that have already been handled and logged are skipped, reducing repeated checks of settled posts.

Separate Decision and Deletion Paths

The workflow sends only posts explicitly marked for deletion to Metricool's delete operation. All other decisions bypass the delete request and continue through the merge path.

These controls reduce unnecessary API calls and help protect posts that should remain scheduled. They do not imply that API failures or every possible scheduling edge case are eliminated.

09. Results and Operational Capabilities

The project description confirms that the workflow is live and runs automatically. It does not provide measured figures for the number of posts cleaned up, API success rate, time saved, or prevented accidental publications.

Implemented Capabilities

Metrics to Measure Next

A measured production review could track:

These figures would help quantify the operational value and verify the effectiveness of the safety rules.

10. Key Takeaways

11. Future Development

The next stage should focus on measuring reliability and improving operational visibility.

12. Final Outcome

The Automatic Cleanup of Scheduled Posts workflow provides an automated safeguard between Airtable approval decisions and Metricool's publishing queue.

By checking cards every 15 minutes, evaluating each associated post, applying time- and status-based deletion rules, and recording the outcome in Airtable, it helps prevent obsolete scheduled content from being left in the publishing queue.

Page · Plain text

AI-Powered Customer Support Agent

An AI agent that reads support tickets, applies your SOPs and product manuals, and drafts accurate replies in the customer’s language.

Section: AI Agents, AI Agents for Ecommerce

AI-Powered Customer Support Agent

An AI agent that reads support tickets, applies company SOPs and product manuals, and drafts accurate responses in the customer's language.

Technologies: n8n, Zapier, PostgreSQL, Supabase, Google Sheets, OpenAI

Project Overview

Customer support teams often spend significant time reading long ticket conversations, searching through standard operating procedures (SOPs), reviewing product manuals, and preparing responses to routine customer questions.

This project introduces an AI-powered customer support agent designed to automate these repetitive tasks. It classifies incoming tickets, retrieves relevant policies and product information, applies business rules, and generates customer-ready response drafts alongside internal support notes.

The system also supports ongoing knowledge synchronization so that responses can reference current SOPs and product documentation. Complex or exceptional cases can be escalated for human review.

01. The Problem: Manual Customer Support Doesn't Scale

Before automation, support teams faced several operational challenges.

Time-consuming manual research

Agents had to read lengthy tickets and conversation threads, then switch between multiple sources to find the information needed to respond correctly.

Inconsistent decision-making

Different agents could interpret the same policy differently, particularly when handling returns, refunds, warranty claims, missing items, defective products, and payment issues.

Outdated knowledge

When product manuals or SOPs changed, agents could unknowingly rely on outdated information, increasing the risk of incorrect responses.

Operational impact

These challenges resulted in:

Core challenge: The business already had the necessary knowledge, but needed a system to retrieve, apply, and use it consistently across support tickets.

02. The Solution: AI-Powered Customer Support Agent

The solution acts as a virtual senior support agent that combines ticket understanding, policy-based decision-making, product knowledge retrieval, and response drafting.

What the system does

The objective is to automate routine research and drafting while preserving human oversight where needed.

03. How the System Works: Six-Phase Workflow

Phase 1 — Ticket Intake and Understanding

The process begins when a new customer support ticket arrives.

The AI system:

Output: A ticket prepared for classification, with its conversation context available for subsequent processing.

Phase 2 — Smart Ticket Classification

The system identifies the type of support request and extracts relevant details.

Ticket categories include:

Relevant details may include:

Based on the identified category and available details, the workflow determines which SOP or product manual is relevant.

Output: A categorized ticket with extracted information and a defined knowledge-retrieval path.

Phase 3 — SOP-Driven Decision-Making

The system applies the company's documented procedures and configured business rules to the ticket.

Rules may cover:

Instead of generating decisions solely from general AI knowledge, the workflow uses the relevant company rules to guide its response.

For example, a refund-related ticket can be evaluated against the applicable refund conditions before a response draft is generated.

Output: A proposed support decision grounded in the applicable SOPs and business rules.

Phase 4 — Product Manual Intelligence (RAG)

For product-specific questions, the system retrieves relevant information from product manuals and the knowledge base using retrieval-augmented generation (RAG).

The retrieval process supports questions involving:

The retrieved information provides context for the AI response, helping it address the customer's actual question rather than relying only on a general model response.

The knowledge-update component is intended to keep product information and SOPs synchronized as documentation changes.

Output: Relevant product information and policy references for the response-generation stage.

Phase 5 — Response Drafting

After the ticket has been classified and the relevant policies and product information have been retrieved, the AI generates two outputs.

A. Customer reply draft

B. Internal support notes

These two outputs serve different purposes: the customer receives a clear response, while the support team receives context to help review the proposed answer.

Phase 6 — Continuous Knowledge Updates

The system includes a knowledge-maintenance component designed to keep support information current.

The intended process includes:

This helps reduce the risk of responses relying on obsolete policies or product instructions.

Output: A knowledge base that can be maintained as product documentation and company procedures evolve.

Image: Main workflow from n8n

04. Complete System Workflow

Main ticket-processing workflow

Knowledge retrieval workflow

Exception-handling workflow

Knowledge maintenance workflow

05. Business Value

Reduced manual research

Agents spend less time searching through lengthy SOPs, manuals, and conversation histories to prepare routine responses.

Consistent policy application

Configured business rules and relevant SOPs guide responses across return, refund, warranty, and other support categories.

Faster response preparation

The system automates ticket understanding, knowledge retrieval, and initial response drafting, helping support teams prepare replies more quickly.

Multilingual support

Customer response drafts can be generated in the language detected in the incoming ticket.

Better knowledge reuse

Existing product documentation and company procedures become reusable sources of context for support requests rather than requiring agents to locate and interpret the same information repeatedly.

Improved agent confidence

Internal support notes explain the basis of a proposed response and identify the relevant policy or manual, giving agents useful context for review.

Support scalability

Automating repetitive work can help teams handle larger volumes of routine requests without increasing manual research effort at the same rate.

06. Results and Expected Impact

The original case study reports the following figures and expected benefits.

Time and effort savings

MetricReported figure
Routine tickets handled before automation50–100 per day
Time spent reading manuals and drafting replies before automationApproximately 2–4 hours per day
Human time required after automationApproximately 10–15 minutes
Monthly time saved per teamApproximately 50–60 hours
Equivalent support capacity freedApproximately 1–1.5 full-time support agents

Accuracy and consistency goals

The case study describes the intended benefits as:

Cost efficiency

The original case study estimates AI costs at approximately $30–40 per month per agent workflow. Actual costs depend on model usage, ticket volume, retrieval infrastructure, and the deployed integrations.

Important: These metrics are reported in the supplied project description and have not been independently verified. The expected benefits should not be interpreted as guaranteed outcomes for every deployment.

07. How to Replicate This System

The supplied case study describes a workflow-based implementation using an automation platform, an AI model, and a knowledge source.

Step 1 — Prepare the workflow

Step 2 — Connect the ticket platform

Configure the relevant ticket-platform integration.

The exact configuration depends on the ticket platform and the available API or webhook integration.

Step 3 — Configure the knowledge base

Choose the knowledge source used for SOPs, product manuals, and supporting documents.

The original case study lists:

Configure the required connection and organize the relevant support documentation. If the implementation uses RAG, ensure the retrieval and indexing components are configured for the selected knowledge source.

Step 4 — Connect the AI model

Configure the OpenAI model credentials and the workflow components responsible for:

Use clear instructions that require responses to follow the retrieved information and configured business rules.

Step 5 — Configure business rules and escalation

Define the policies that determine how common support categories should be handled.

Examples include:

Confirm that exceptional cases are routed appropriately rather than being treated as routine requests.

Step 6 — Test the workflow

Submit test tickets covering different scenarios.

Verify that:

Test return, refund, warranty, product troubleshooting, and other relevant scenarios before enabling live processing.

Step 7 — Activate and monitor the workflow

Once the workflow has passed testing:

From that point, supported tickets can be processed automatically according to the configured workflow, with human review or sending behavior determined by the deployment.

08. Final Outcome

The AI-Powered Customer Support Agent brings ticket understanding, policy retrieval, business-rule evaluation, product knowledge, and response drafting into a unified workflow.

By automating routine research and preparing customer-ready drafts, the system is designed to reduce repetitive work, improve consistency, and help support teams respond more efficiently. Its knowledge-maintenance and escalation components also support ongoing documentation updates and human handling of complex cases.

The result is a more structured approach to customer support in which AI handles repetitive processing while the support team retains visibility into the policies, information, and next steps behind each proposed response.

Page · Plain text

End of part 5 of 7. Continue with part 6.