The Taxonomy of Customer Questions

When we built the intent classification model, we started by categorizing hundreds of thousands of real support messages into distinct intents. The final taxonomy has 315 intents organized into 13 categories.

Here is what the distribution looks like across industries, and what it tells us about customer behavior.

The Universal Top 5

Regardless of industry, these 5 intents consistently appear in the top 10 for every business:

1. Account access issues (password reset, login problems, locked accounts). People forget their passwords constantly. This is the most automatable support category.

1. Billing questions (charges, invoices, payment methods). Customers want to understand what they are paying for. Clear billing pages reduce this, but it never goes to zero.

1. Product/feature questions (how does X work, does it do Y). Pre-purchase and post-purchase curiosity. Better onboarding and documentation reduces volume but does not eliminate it.

1. Order/delivery status (where is my order, tracking, shipping time). E-commerce specific but universally common. The answer is always in the fulfillment system.

1. Cancellation/refund requests (I want to cancel, I want my money back). These are emotionally charged but procedurally simple. Fast, respectful handling matters here.

Industry-Specific Patterns

SaaS companies see high volumes of:

• api_error and integration_issue (technical users hitting problems)
• feature_request (customers want the product to do more)
• onboarding_help (new users getting stuck)
• data_export (customers want to get their data out)

E-commerce stores see high volumes of:

• order_tracking (the single biggest category)
• return_request and exchange_request
• product_availability (is this in stock)
• shipping_cost and delivery_time
• discount_code_issue (codes not working at checkout)

Service businesses see high volumes of:

• appointment_scheduling (booking, rescheduling, canceling)
• pricing_inquiry (custom quotes, package details)
• availability_check (do you serve my area)
• complaint (service did not meet expectations)

What the Data Tells Us

Insight 1: 60 to 70% of all support is informational. The customer wants a piece of information that already exists somewhere (in your database, on your website, in your docs). They are asking because they could not find it or it was faster to ask. This is entirely automatable.

Insight 2: 15 to 20% is procedural. The customer wants an action performed (cancel, refund, change plan, update address). The action is standard, but many businesses still require manual handling. With the right integrations, these are automatable too.

Insight 3: 10 to 15% is genuinely complex. Multi-step technical issues, complaints that need empathy, situations that require judgment. This is where humans add real value.

Insight 4: Product gaps drive support volume. When a specific intent suddenly spikes, it almost always points to a product issue. A confusing UI element. A missing feature. A bug. The best long-term support strategy is not faster responses; it is fixing the product so the questions stop.

Using Intents to Improve Your Product

Here is a simple exercise: look at your top 10 intents. For each one, ask:

1. Can we automate the response? (Quick win)
2. Can we improve the product so this question does not arise? (Long-term win)
3. Can we improve our docs or onboarding to address this proactively? (Medium-term win)

Some examples:

• Top intent is onboarding_help: Your onboarding flow needs work
• Top intent is pricing_inquiry: Your pricing page is not clear enough
• Top intent is integration_issue: Your integration docs or UI needs improvement
• Top intent is bug_report for a specific feature: That feature needs fixing

The 315 intents in the classification model are not just labels for routing. They are a map of what customers need, organized by category. Use that map to improve your product, and your support volume drops naturally.

The Evolving Picture

Intent distribution changes over time:

• After a product launch: new_feature_question and onboarding_help spike
• After a pricing change: pricing_inquiry and subscription_change spike
• After an outage: bug_report and complaint spike
• During holiday season (e-commerce): order_tracking and shipping_delay spike

Tracking these shifts weekly helps you anticipate support needs and adjust automation rules proactively rather than reactively.