AI Functions in Excel: Sort, Score and Summarise Text with a Formula

Most of the text in a spreadsheet never gets used. A column of customer comments, ticket descriptions or survey answers sits next to the numbers, and nobody reads all of it because nobody has an afternoon to spare. Excel's own functions can't help, because LEFT, FIND and SEARCH don't know what a sentence means.

AI functions do. With the SheetAI Excel add-in installed, you get formulas that read a cell and answer a question about it: which category is this, how angry is the customer, what's the order number. We tried them in Excel on the web, on twelve support tickets from a made-up furniture shop. Every screenshot below is from that session.

The functions

The SHEETAI. prefix catches people out. Excel requires every add-in function to carry the add-in's name, so the functions you might know from Google Sheets as =AI_CLASSIFY are =SHEETAI.AI_CLASSIFY here. Start typing =SHEETAI. and Excel's autocomplete lists them all.

The tickets

Twelve messages, the kind any small online shop gets in a week: a desk stuck in "processing" for three weeks, a double charge, a lamp that flickers, a delighted customer, a question about shipping to Ireland. Column A has the ticket number, column B the message. We wanted four columns filled in: a category, a sentiment score, the order number if there was one, and a one-line summary.

Step 1: Write the formulas for the first row

One formula per column, all pointing at the message in B2:

The first ticket came back in a few seconds: Delivery, -0.75, 58213, and a summary that kept the detail that matters, that the customer is considering asking for a refund. Then we selected C2:F13 and pressed Ctrl+D to fill all four formulas down.

Step 2: Read what came back

Going through it row by row, here's how each function did.

• AI_EXTRACT: 12 out of 12. Every order number came out right, whether it was written "order #58213", "Order 58190", "Order no. 58102" or just "#58301". The four messages with no order number stayed blank instead of inventing one. This is the function we'd trust first.
• AI_SENTIMENT: sensible throughout. The double-billed repeat customer and the rain-soaked desk both scored -0.85, and the thank-you to Priya scored 1. The mixed "good value, but step 6 is unclear" got 0.55, which is about right. Sort by this column and the tickets to answer first are at the top.
• AI_SUMMARIZE: kept what matters. Every summary held on to the detail a support lead needs: the refund threat, "this is the second time". For an even shorter version, use AI_ASK with a word limit.
• AI_CLASSIFY: every ticket in a bucket. Billing errors, damaged items, product faults, praise and questions all landed where you'd expect, which makes the column ready to filter and count.

Getting the most from AI_CLASSIFY

AI_CLASSIFY always returns one of the categories you give it, so the list is where you shape the result. Make each category distinct, and add a separate one for anything you'll want to count on its own. For a support inbox, splitting "Delivery" into late orders, transit damage and wrong items gives your team three clear queues:

Step 3: Draft the replies with AI_ASK

AI_ASK is the general-purpose one: any instruction, any text. We used it to draft a first reply to each ticket, with the rules a support lead would give a new hire:

Both replies are ready to edit and send. Add your own details, such as the refund policy or the courier's tracking link, and a reply that used to take five minutes takes thirty seconds.

How the functions work, and what they cost

Cost is the thing to plan for. Each cell with new text is a separate question to the model, at least one credit each, so our four columns across twelve rows were about 48 credits. That's fine for a weekly batch of tickets. For a 5,000-row export, start with a 50-row sample, check it, and only then fill down the rest.

The good news is that answers are remembered. Recalculating, re-opening the workbook or sorting the sheet doesn't send the same question twice, so you aren't charged again for text that hasn't changed.

Functions or the SheetAI pane?

The add-in also has a chat pane, and it's worth knowing which to reach for. Functions are for one answer per row that you want to keep: a category, a score, an extracted value. They live in the cells, they fill down, and anyone who opens the workbook with the add-in sees the same results.

The pane is for work across the sheet: "build a summary of tickets by category with a chart", "highlight every ticket below -0.5", "which product gets the most complaints?". You describe the result and it does the steps. With the four columns above in place, that summary is one sentence in the pane.

Formulas you can paste

What we'd tell someone trying it tomorrow

Start with AI_EXTRACT and AI_SENTIMENT, which need no setup and were right on every row we checked. Then add AI_CLASSIFY, and spend five minutes on the category list, because that's where the quality comes from. Read a handful of rows before you sort or count by any AI column. And run it on a sample before filling down ten thousand rows, so the first thing you learn isn't the bill.