EBITDAI Blog
    Product Update
    September 22, 2026
    6 min read

    ASKJEV: ask a yes or no question in a cell, get a probability back

    EBITDAI now has a custom function. You type it into a cell, it asks a yes or no question of your data, and it returns a number between 0 and 1: the probability that the answer is yes. It is a typed, repeatable answer sitting in a cell, not a chat reply you have to read and retype.

    The formula

    =EBITDAI.ASKJEV(question, [format], [context])
    • question, a yes or no question in plain words.
    • format, a code: 1 = probability (the default, a number from 0 to 1), 2 = TRUE/FALSE, with TRUE at 0.5 and above. Leave it blank for a probability. The older text values "probability" and "binary" still work.
    • context, an optional range of cells the function reads as evidence. Any shape works, and headers help.

    For example:

    =EBITDAI.ASKJEV("Will revenue beat plan?", 1, A1:A10)

    That is the whole interface. The rest of this post is what comes back, what happens when the question cannot be answered, and how to get the function today.

    Watch it run

    ASKJEV in a workbook: the formula, the answer, and what a changed input does to it.

    What answers the question

    The answer comes from Jev, TypeSafe AI's System One model. Jev does not write a paragraph that something else then reads a yes or a no out of. It returns a typed judgment, and that judgment is calibrated: across many questions of this kind, the ones Jev scores 0.8 come back yes about 80 percent of the time. So a 0.6 and a 0.9 are different claims, and the difference is something you can sort on, filter on and set a threshold across.

    It answers in about half a second. Fill the formula down a column and the column goes out as one batched request rather than one request per row.

    Unchanged inputs are not re-sent. The add-in keeps a client-side cache, so recalculating a sheet is free: the same question over the same cells gives back the answer it already has. Change a cell inside the context range and that answer recalculates.

    When it will not answer

    Three kinds of question return #N/A instead of a number, with the reason in the cell's tooltip: a question that cannot be answered from the evidence you gave it, a question that asks for a forecast of an unknowable future event with no data behind it, and a question that is nonsense.

    A model that always returns a number is easy to build and hard to use, because nothing in the output separates the answers it knows from the ones it invented. Jev declines instead, and the cell says so. A format argument other than 1 or 2 (or the older "probability" and "binary") returns #VALUE!.

    Three examples, with the numbers we measured

    These are from our test runs on 22 September 2026, in a live workbook.

    1. A fact it knows

    =EBITDAI.ASKJEV("Is Paris the capital of France?")

    returns 0.99. The same question with format 2:

    =EBITDAI.ASKJEV("Is Paris the capital of France?", 2)

    returns TRUE.

    2. A judgment over your own cells

    Put a small block of figures in A1:B4: net debt 900, LTM EBITDA 300, EBITDA falling 12 percent year on year for two quarters, undrawn revolver 50. Then ask:

    =EBITDAI.ASKJEV("Is this company likely to breach its net debt to EBITDA
      covenant of 3.5x within the next four quarters?", 1, A1:B4)

    The cell returns 0.82. Change LTM EBITDA in that range from 300 to 600 and the cell recalculates to 0.78. The answer is wired to the range, not to the text of the question, which is the behaviour to check for before you trust a column of these.

    3. Two it refuses

    =EBITDAI.ASKJEV("Will the US Federal Reserve cut its policy rate at its
      November 2026 meeting?")

    with no data attached returns #N/A. Jev rated it unanswerable. This one returns #N/A as well:

    =EBITDAI.ASKJEV("Is the number seven happier than the colour blue?")

    A number in either of those cells would have been made up, and a made up number is the thing you cannot catch once it is sitting in a column with 200 real ones.

    What to point it at

    Anything you would otherwise eyeball row by row. In finance work, the questions we expect it to earn its place on:

    • Has a covenant been breached, given the debt schedule.
    • Is this line item capex or opex.
    • Does this contract clause allow early termination, with the clause pasted into a cell.
    • Is this customer likely to churn, given its usage row.
    • Is this invoice a duplicate of the one above.

    Nothing about the function is finance-specific, so the same shape works on any list of rows you need classified: is this review positive, does this email need a reply today, does this product description mention a warranty, is this address in the EU.

    How to get it

    ASKJEV is now part of the EBITDAI add-in on Microsoft AppSource.

    If you already use EBITDAI, you get ASKJEV automatically when Office picks up the update. That can take up to a day, and closing and reopening Excel helps.

    If you are new to EBITDAI, install the add-in from Microsoft AppSource. Open the pane, sign in, and type the formula into a cell. It works in Excel on Windows, Mac and the web.

    What each plan gets

    ASKJEV is on every plan. A call is one answered cell, counted per calendar week.

    • Free: 3 calls a week.
    • Lite: 100 calls a week.
    • Pro: 2,000 calls a week.

    The cache matters here: unchanged inputs are not re-sent, so recalculating a sheet you have already answered does not go back to the model. See pricing for what else sits on each plan.

    Questions

    Is it deterministic?

    The same question over the same data returns the same answer, served from the cache. Underneath the cache, what Jev returns is a calibrated judgment rather than a lookup, so treat the number as a probability, not as a fact the model retrieved.

    Is my data sent anywhere?

    Yes, and it is worth being precise about where. The question text and the contents of the context cells go to EBITDAI's server and on to TypeSafe, which is what produces the answer. TypeSafe does not train on customer requests.

    Can it forecast?

    No, and it says so rather than pretending otherwise. Ask it what a central bank will do at a meeting months away with nothing attached and you get #N/A, not a number that looks like knowledge. Give it evidence and ask it to judge what the evidence implies, which is what the covenant example above does, and it will answer.

    Try it

    Install or update EBITDAI, open a sheet with a column of rows you keep reading one at a time, and put a question next to the first one. The add-in page covers what the rest of EBITDAI does, and the free plan needs no card.

    Related Articles