How to reduce support tickets, and the three you should never deflect

Ask most founders what their customers write in about and you get a shrug and a guess. Ask them to name their top five questions in order and almost nobody can, which is odd, because the answer is sitting in their inbox in full.

Support volume is rarely spread evenly. Go through a month of it and you will usually find five or six questions doing most of the work. The rest is a thin scatter of one-offs. Until you know which six, every hour spent on deflection is aimed at a guess.

Count before you cut

Go through the last sixty messages and put each one in a bucket. Not a clever taxonomy, just a line in a text file per recurring question with a tally beside it.

It takes about forty minutes and it is the best forty minutes you will spend on support all quarter. You will find the question you already knew about. You will also find one you had no idea was so common, usually something small and stupid you could fix in an afternoon and have been walking past for months.

Do this before you buy anything or write a single help page.

The ways out, in the order that works

Each of those six questions has a best answer, and they are not equally good.

Fix the product. If forty people ask where the export button is, the export button is in the wrong place. This is the only option that removes the question instead of answering it, and it is nearly always the cheapest, because a ticket you never receive costs nothing to handle for the rest of the product’s life. Founders skip this step constantly, because a help page looks like work and a moved button looks like a mistake being corrected.

The page that failed is usually the fix. A question asked by people who were already on your pricing page means the pricing page did not say the thing. Go and read it as though you had never seen the product. What is missing is normally one sentence, in the place the reader was already looking.

Some questions deserve answering where they are asked. People arrive mid-thought and will not go hunting, so a legitimate question keeps coming however good the page is. Those are the ones automated customer service is for, and a widget reading your own documentation can take them at the moment they come up.

Anything your documentation does not cover needs writing once. That is mostly a filing job rather than a writing one. The hard part is knowing which answers are missing in the first place.

Where Deacon fits

Every question a visitor asks gets grouped with the others that mean the same thing, so “how do I get my data out”, “is there an export” and “can I download everything” arrive as one item with a number beside it. That is the tally, kept for you.

Anything your documentation cannot answer goes on a second list with a writing box next to it. Fill it in once and the answer joins your knowledge base, so the next person who asks gets it straight away. Until you do, Deacon tells that visitor it does not have the answer and takes their email address instead of improvising, and the conversation waits in your dashboard.

The three you should never deflect

Deflection has a point past which it turns into a wall, and the tools rarely tell you where that line sits. Three kinds of message should always reach a person, so wire that human in the loop before you automate anything.

Anything about money goes first. Refunds, double charges, a card that failed, a plan somebody cannot cancel. Getting one of these slightly wrong costs you a chargeback and a public complaint, and the person writing in is already annoyed.

Bug reports go straight through too. Somebody is doing you an enormous favour and most of them will not do it twice. Send an automated reply telling a bug reporter to check their internet connection and you have taught a helpful person that telling you things is a waste of their evening.

Cancellations are the third, and not because you should fight them. The reason somebody leaves is worth more than the subscription was, and a cancellation that goes through a deflection flow gives you a churned customer and no explanation.

Everything else is fair game.

What to watch afterwards

Two numbers, and neither of them is a satisfaction score.

Watch whether your top question changed. If it has, the fix worked and something else is now the biggest problem. If the same question is still sitting at the top three months later, whatever you did was decoration.

Then watch how many questions your documentation cannot answer at all. That number should fall as you write the missing answers. When it starts climbing again it usually means you shipped something and forgot to document it.

Both of those tell you where the next forty minutes should go.