Skip to content

Dissatisfaction Tracker with AI Builder and a Model-Driven App

International learning and development company

In short

How can AI Builder sort customer complaints by sentiment and category?

When a customer complains, the email trail passes through several inboxes before anyone logs it in a spreadsheet. Each hand-off adds delay, and the monthly picture arrives late. We built a proof of concept that turns a complaint email into a tracked record in one click, with AI Builder suggesting the sentiment and category.

Results in numbers

saved per complaint
10 min
training emails
100

At a glance

  • Spreadsheet tracker replaced by a model-driven app
  • AI Builder sorting complaints by sentiment and category
  • A complaint logged from the email in one click

The client

The client is a learning and development company with operations in many countries. Customer complaints reach sales and customer service staff by email, with Dynamics 365 handling customer email.

What it was costing them

  • Wasted motion: the email trail is forwarded to a distribution group, then again to the workflow manager.
  • Manual logging: each complaint is typed into a spreadsheet tracker and updated after calls with staff.
  • Late insight: findings are discussed monthly, so patterns surface weeks after the events.
  • Heavy root cause work: the 5 Whys step depends on chasing staff for information.

What we built

We built a model-driven app and flows that capture complaints from email and classify them with AI Builder.

  • Spreadsheet retired: a Tickets table of about 30 columns and a Departments table replace the tracker.
  • One click logging: a button on the email ribbon opens a custom page and starts a flow that creates the ticket and links the email to it.
  • Customer acknowledged: a flow sends an automatic reply that the complaint is being processed.
  • Sentiment at a glance: the AI Builder prebuilt model flags positive, negative or neutral text.
  • Category suggested: a custom AI Builder model, trained on tagged emails, proposes the category for a categoriser to confirm.
  • Guided handling: a business process flow with three stages leads each ticket through the process.
  • The right people see the right tickets: two security roles, Support and Categorizer, and views filtered by team.
  • A monthly picture without effort: four charts and one dashboard, such as problem categories and categories by status.

Results

  • The scope was 2 tables, 2 security roles, 1 business process flow, 4 charts and 1 dashboard.
  • Model training needs 100 sample emails with a category, at least 10 per category.
  • Built in 8 to 12 hours for the first app and 4 to 6 hours for the email button.
  • Result: logging a complaint drops from several forwards and a manual entry to one click, saving 10 minutes per complaint.

The risk most teams miss

At month three the risks are model drift and thin training data. With about 10 examples per category the model will mislabel unusual complaints, and categories change over time. The solution includes a confirm step, a retraining routine fed by corrections stored in Dataverse and a named owner for the category list. The email button also depends on a Dynamics 365 setup that another team owns.

What we would add today

Improvements we would add on top of what is live.

  • Use an AI Builder prompt to summarise the email trail and draft the 5 Whys, and to pick from a fixed category list instead of a custom model.
  • Add a Copilot Studio agent for the customer service leadership team that answers what the top causes were this month.
  • Show trends in Power BI or Fabric over Dataverse, with monthly category and department views.
  • Apply Dataverse row security by department and field security on customer details.
  • Deploy through Power Platform pipelines with connection references, so the flows do not rely on a person's account.

Technology

Power Apps model-driven app, Dataverse, AI Builder, Power Automate, Dynamics 365 email, Outlook.

Tell us how a complaint travels through your teams today, and we will tell you which steps a tracked, classified record can replace.

Products and capabilities

  • AI Builder
  • Power Apps
  • Dataverse
  • Power Automate
  • Dynamics 365
  • sentiment analysis
  • complaint tracking
  • model-driven app

Have a Microsoft project in mind?

Tell us what you need built, fixed or supported. A senior consultant replies within one business day.