AI formulas in Excel that return a number
Hunch adds four formulas to Excel. =HUNCH.ASK returns a probability, =HUNCH.PICK returns one option from a list, =HUNCH.SCORE returns a position on a scale, and =HUNCH.MULTI returns several probabilities from one row at once. Every answer is a number or a short label you can sort, never a paragraph.
The four formulas
| Row | A | B | C |
|---|---|---|---|
| 1 | Formula | Returns | Use it for |
| 2 | =HUNCH.ASK(text, question) | Probability 0 to 1 | One yes/no judgment per row |
| 3 | =HUNCH.PICK(text, options) | One option, exactly as written | Sorting rows into categories |
| 4 | =HUNCH.SCORE(text, question, levels) | Position across your levels | Ordered scales like sentiment or fit |
| 5 | =HUNCH.MULTI(text, questions) | One probability per question | Several flags on one row at once |
All four read a cell or a range of text and spill an answer of the same shape back. A range in, a range out; no separate step to run and no output to copy back into the sheet.
Which one to use
- One yes/no question with a clear definition: HUNCH.ASK. "Is this a decision maker" becomes a definition inside the question, and the answer is a probability you can threshold.
- One category out of a fixed list: HUNCH.PICK. Add other so nothing gets forced into the wrong bucket.
- An ordered scale, such as sentiment, urgency or fit: HUNCH.SCORE. The result is a position, not only a label, so it sorts and averages.
- Several independent flags on the same row, such as refund, churn risk and blocked all at once: HUNCH.MULTI, one text read, several columns out.
How this differs from a chat window
Copilot in Excel and any chat-style AI formula take a prompt and write text back: a summary, a draft, a rewrite, a category name worded slightly differently every time you ask. That is the right tool for writing. It is a poor tool for a column you plan to sort, because the format drifts between rows, and a chat model's "0.8" is a token it generated, not a number computed to be a probability.
| Row | A | B | C |
|---|---|---|---|
| 1 | Copilot or a chat formula | Hunch | |
| 2 | Output | Generated text | Probability, option, or level |
| 3 | Format between rows | Varies | Identical |
| 4 | Price | Included in a Microsoft 365 plan, or per token | Per row, $29 for 5,000 |
| 5 | Install | Built into eligible Microsoft 365 plans | Sideload a manifest once |
| 6 | Writes prose | Yes | No |
| 7 | Sorts, filters, pivots | After cleanup | Directly |
What none of the four formulas do
- Write anything. No summaries, no drafts, no rewrites; that stays a chat model's job.
- Arithmetic, counting, or comparing dates. Keep amounts and dates in normal formulas next to the Hunch column.
- Read images, PDFs, or attached files. Text in a cell only.
- Match English-level accuracy in every language. Test a sample before relying on another language.
Use both. Let Copilot or a chat formula draft the reply, and let a Hunch column decide which rows need one before anyone starts typing.
One row, four kinds of answer
=HUNCH.ASK(A2, "Does this message need a reply from a human?")
=HUNCH.PICK(A2, "billing|bug|feedback|other")
=HUNCH.SCORE(A2, "How urgent is this message?", "low|medium|high|critical")
=HUNCH.MULTI(A2, "Should this go to a human instead of an automated reply?")In Google Sheets: =HUNCH(A2, "Does this message need a reply from a human?")
=HUNCHPICK(A2, "billing|bug|feedback|other")
=HUNCHSCORE(A2, "How urgent is this message?", "low|medium|high|critical")
=HUNCHMULTI(A2, "Should this go to a human instead of an automated reply?")
Six customer messages, run through all four formulas above, with the model's answers as returned on 28 September 2026.
| Row | A | B | C | D |
|---|---|---|---|---|
| 1 | Message | Needs a reply? | Topic | Urgency |
| 2 | Why was I charged $49 when the plan page says $39? | 0.74 | billing | 1.08 |
| 3 | Everything crashed during our demo in front of the client. | 0.87 | bug | 2.91 |
| 4 | Just wanted to say thanks, the new update is great. | 0.32 | feedback | 0.00 |
| 5 | Is there a way to export to PDF automatically every Friday? | 0.56 | other | 0.14 |
| 6 | This is the fourth time support has closed my ticket without fixing it. | 0.90 | feedback | 2.04 |
| 7 | No rush, but eventually could you add dark mode? | 0.64 | feedback | 0.00 |
| Row | A | B |
|---|---|---|
| 1 | Message | Escalate to a human? |
| 2 | Why was I charged $49 when the plan page says $39? | 0.61 |
| 3 | Everything crashed during our demo in front of the client. | 0.91 |
| 4 | Just wanted to say thanks, the new update is great. | 0.29 |
| 5 | Is there a way to export to PDF automatically every Friday? | 0.44 |
| 6 | This is the fourth time support has closed my ticket without fixing it. | 0.92 |
| 7 | No rush, but eventually could you add dark mode? | 0.38 |
Row 6 makes the case for using more than one formula on the same row: "no rush, but eventually could you add dark mode" needs a reply at 0.64, since a feature request usually deserves at least an acknowledgment, while urgency lands at exactly 0, the floor of the scale, because the customer said outright there is no rush. Same message, two different questions, two unrelated numbers. Row 5 is the surprising pick: "the fourth time support has closed my ticket without fixing it" is filed under feedback rather than bug, because the sentence is a complaint about the support process, not a description of what is broken, and the four-option list had nothing closer. Escalation tracks urgency closely (0.92 here, 0.91 on the crashed demo) but not exactly: the billing question sits at 0.61 to escalate against only 1.08 on the urgency scale, because a small charge discrepancy is not urgent, yet still not something an automated reply should try to resolve alone.
Setup in three steps
- Sideload the add-in from hunchsheet.app/excel: Excel for Windows (Microsoft 365), Mac 2016 or later, and the web.
- Save your key once:
=HUNCH.SETKEY("hunch_..."), then delete the cell. The same key and credits work in Google Sheets too. - Type whichever of the four formulas fits the column, and fill it down.
A guide for each job
- Sentiment analysis in Excel: HUNCH.SCORE on reviews, plus a complaint flag.
- Text classification in Excel: HUNCH.PICK with a confidence column to find rows to check.
- Categorize bank transactions in Excel: raw card descriptions into budget categories.
- Lead scoring in Excel: decision-maker probability and ICP fit, ranked with SORTBY.
- Categorize support tickets in Excel: category, urgency, and several flags at once.
- Analyze open-ended survey answers in Excel: one column per theme, counted with COUNTIF.
Questions
Does Hunch replace Copilot in Excel?
Do the formulas work offline?
Can I mix a Hunch formula into a normal calculation?
Is there a free way to try all four?
Do the four formulas cost different amounts?
Run it on your own column
Install the Excel add-in (two minutes, Windows, Mac and the web), save your key with =HUNCH.SETKEY, and point the formula at your data. 100 rows free to start, then $29 for 5,000 rows. Credits never expire and there is no subscription.
Reference: HUNCH · HUNCHPICK · HUNCHSCORE · HUNCHMULTI
More guides
- AI formulas for Google Sheets: what returns a number, what returns text AI formulas compared
- Categorize support tickets in Excel Ticket triage
- Sentiment analysis in Excel with one formula Sentiment
- Lead scoring in Excel without a CRM Lead scoring
- Text classification in Excel with a formula Classification