We read 1,000 support tickets. Seventy-one percent were four questions.

Ask any support manager whether their tickets could be handled by a machine and they will say no, sincerely, because our tickets are complicated. And they are right about the ones they remember. Memory is a terrible sampling method: it keeps the hard ones and discards the four hundred identical ones.

So we count instead. Every time, in every industry, the same shape appears.

The shape

Across a thousand consecutive tickets at a mid-size distributor, classified by the action taken rather than the words used:

Four questions. Seventy-one percent. And the long tail is genuinely long — forty-plus intents in the remaining 29% — which is exactly why it feels varied from the inside.

Run this in other industries and the labels change while the curve does not. A clinic: appointments, prescriptions, billing questions, records requests. A SaaS company: login problems, billing, one or two feature confusions, one integration that breaks. A property manager: maintenance requests, rent questions, lease documents, noise complaints. The top four always clear 60%.

Why this is not the same as "just add a chatbot"

Here is the part that matters, and it is why the obvious conclusion is wrong.

Of that 71%, the majority are not questions at all. They are requests for an action phrased as questions.

• "Where is my order" wants a tracking number pulled from a system, matched to the right shipment line, with a real ETA. Not a link to a tracking page.
• "It arrived short" wants a credit issued, or an RMA opened. Not an apology and a form.
• "I need the invoice" wants the specific document retrieved and attached.
• "Change my order" wants a write to the ERP before the pick list is generated.

This is precisely why the FAQ bot generation failed. It correctly identified the top intents and then answered them with information when the customer wanted an outcome. "Your order shipped on Tuesday, you can track it here" is a worse response than a human's, because the human would have looked at the tracking, seen it was stuck in Memphis, and said so.

What the count actually tells you to build

Read the table as an integration list, not a content list. Each of the top four maps to a system and an action:

Two are read-only and can go live almost immediately. Two are reversible writes and belong behind an approval queue until they have earned their way out. That is a thirty-day project with a defined edge, and it is worth more than 71% of your ticket volume because it is the cheap 71%.

Do the count yourself

Five hundred tickets. One column for the action taken. One column for reversible yes/no. One afternoon. It is the highest-return hour of analysis available to any support organisation, and it costs nothing.