What a repeated question tells you
When yet another customer asks whether a case fits their phone model, it is easy to conclude that the description lacks a list of models. Sometimes it does. But sometimes the list is there and on a phone it hides under a closed tab, and sometimes the customer has read it and wants confirmation, because they are buying for an expensive device and prefer to ask twice.
So treat every repeated question as a hypothesis. Before you add another paragraph, retrace the customer's path on a phone to see what was really in front of their eyes. Missing information and information that cannot be reached are two different fixes. A table hidden under a tab needs a better place. If the table lacks the dimension people ask about, add a new column.
With questions about delivery, separate two matters. Some people ask about dispatch time because they did not find it on the product page. Others write because their parcel is late, and no description will fix that.
There is also someone who is missing from these records. A conversation tells you about the person who wrote. For those who left without a word, your site analytics and your own test purchase will tell you more than guesswork.
How do you analyse repeated customer questions?
Collect questions from every place where customers reach you, which means email and chat, but also social media messages and notes from phone calls. Once a week is a sensible rhythm to start with, and with a small number of conversations once every two weeks will do. Begin with the questions you answer by hand even though the answer is already somewhere in the shop.
Count cases. Three emails from the same person about the same case are one case. The raw number of questions is not enough either, because a product sold in hundreds will collect more questions than a niche one, even with a better description, so next to the number of questions always put the number of orders.
A simple spreadsheet is all you need. You do not need the customer's name or address in it, because knowing who asked will not help you correct a size table. The columns can look like this.
- the question in the customer's words
- the product or topic it concerns
- the number of cases in the period and the number of orders for that product
- the place to check, for example the product page, the delivery page or the bot's knowledge
- the date of the fix and the result of rechecking
Chat conversations and what remains of them
Chat is simpler. Conversations save themselves, and in CartBot you browse them in the panel together with a list of questions the bot found no answer to in your materials.
Read the question together with the bot's answer. If the bot showed a different product from the one the customer meant, read the whole conversation from the start, because the mistake may have begun earlier, when the need was being established. The list of topics to fix is still put together by a person who knows the shop.
History does not last forever. A conversation record in CartBot disappears 60 days after the session started, and unanswered questions stay for 180 days from the last update. If you want to compare problems across the whole season, move your conclusions into the spreadsheet as you go, without copying personal data.
When you correct a description, correct the knowledge the bot uses as well, because the page and the source of the answer are two separate places and it is easy to update only one. Then put the customer's question to the bot again. The answer should match the new description.
How do you turn conversations into a list of shop fixes?
From the spreadsheet, pick first the questions that concern many products at once, or those that block the purchase of a specific variant, and with each fix note the day of the change and what is supposed to happen afterwards. A goal like “the customer sees the dimensions right after opening the product page on a phone” you can check in a minute.
Then watch whether the questions return. Say that before the fix you had 8 questions about dimensions per 100 orders in a category, and after it 6 questions per 200 orders. The rate fell from 8 to 3 questions per 100 orders. The numbers are made up for the sum, and even in a real shop such a drop would not prove that the description change alone caused it, because traffic may have changed in the meantime.
If questions keep coming, read what they say before you decide the fix failed, because customers may now be asking about a different dimension or a new phone model. The old problem may long since be solved.
Customer conversation analysis at the end of the year
At the end of the year you can build a season summary from the same notes, but first check what period your data covers and from which channels. If your chat records reach back two months and you started taking email notes in September, call it an autumn review and write that limitation next to the results. Do not fill in missing months from memory.
Four labels are enough to start with: missing information on the site, product outside the range, trouble with the purchase itself, and a question after which the customer placed an order. One conversation can get several labels, so stick to the labelling method you chose.
Then a few numbers. Count what share of conversations began outside your support hours. If your sample has 100 conversations and 25 began after closing, that is 25 percent of conversations in your sample and no more than that follows from it. The number says nothing about all the visitors who needed help in the evening.
What from the summary goes into the January plan
If your chat attributes orders to conversations, look at them together with their value and with how your measurement counts them, because attribution alone does not prove the customer bought because of the conversation. Questions about products you do not stock are a signal of interest. Do not add them to your sales forecast yet.
From the list pick a few tasks for January, for each one write a person and a deadline, and next to it how you will know the fix worked. Missing dimensions are a job for the product page. A payment error someone first has to reproduce, and a new range will wait until you have checked the interest and counted the costs. In January add the end of December to the review, plus returns and cancellations that you do not yet know about on the day of the summary.
Since a conversation record disappears after 60 days, the December summary rests on notes made every month. Make the first one today, from the conversations still in the panel.