TL;DR
Most AI assistants read a spreadsheet like a document, flatten its structure, and answer from a sample of rows.
Kotae finds each table separately and calculates over every row, every time.
Answers trace back to the source rows, and your file is parsed on your own systems.
In 2011, Dr. Dre put out "I Need a Doctor" with Eminem and Skylar Grey - a double-platinum hit whose title says everything: when something you depend on stops working, sympathy doesn't help. You need someone who can fix it.
If Excel and CSV files are your team's pain, Kotae is the doctor you're looking for.
Kotae is our AI chatbot for your business: connect your documents and spreadsheets, ask a question in plain language, and get an answer back.
But no good doctor skips the diagnosis, so let's start with where it hurts.
The symptoms: simple Excel questions, wrong answers
"How many customers signed up last quarter?"
"What was the margin for our Europe region?"
Simple questions. But when the answer lives in a spreadsheet, most AI assistants give you an answer that's confident, fluent, and completely wrong.
Here's the unsettling part. There's no pain, no error message, no red flag. The number arrives looking perfectly healthy, and you take it into a meeting. Symptoms this quiet are the ones worth worrying about.
It isn't that AI is bad at math. It's about how it reads - and spreadsheets break the way most tools read in two different ways.
Misdiagnosis one: treating a spreadsheet like a document
AI tools do a decent job on Word documents and PDFs. Excel files and CSV exports are different.
In the real world, a single sheet is rarely one neat table. It often holds several tables side by side, a summary box in the corner, section headings, footnotes, a signature block, and formulas - all on the same page. A person glances at it and instantly knows: that's the pricing table, that's the summary, that's just a note at the bottom.
Most tools don't. They photocopy the whole sheet into one big blob and hand it to the AI. The AI then can't tell where one table ends and the next begins - so it mixes up the numbers, or quotes from the wrong table entirely.
It's the equivalent of reading through a patient's file, stopping at last year's sprained ankle, and treating that instead of the chest pain they came in with.
Misdiagnosis two: the answer isn't in any one row
To answer a question, an AI assistant skims your file for the passages that look most relevant, then writes an answer from what it found. For a policy document, that works - the answer usually sits in a paragraph or two.
Spreadsheets don't behave like that. The answer to "how many" isn't in any one row. It's in all of them.
So when an AI assistant grabs the thirty most relevant-looking rows and counts those, you get a number that sounds authoritative and isn't. It's diagnosis by glance: a look at the patient from the doorway, no tests, and a firm answer anyway. The dangerous part is that a wrong number reads exactly like a right one. There's no hedge, nothing to warn you it came from a fraction of your data.
Kotae's treatment plan: how it reads Excel and CSV files
Our approach comes down to two ideas.
We read the whole file properly. Kotae reads a spreadsheet the way a person does - finding each separate table and keeping it apart, keeping every number tied to the label and section it belongs to, and holding on to the formulas and links that most tools throw away. The result is clean, well-labeled information the AI can actually trust.
We run the test instead of eyeballing it. When a question needs a calculation, Kotae does the calculation - over every row, every time. A question like "what's our total spend by region?" is answered by totaling the column, not by reading a sample and estimating.
That single difference is what separates an answer that's usually about right from one that's exactly right.
We take patients as they arrive: real, messy spreadsheets
Anyone who works with real business spreadsheets knows they look nothing like the tidy grids in a tutorial. There's a totals row at the bottom, a title and a date above the table, blank cells that say "N/A" or "TBD," product codes with leading zeros, dates written five different ways. CSV exports are usually the worst of all: the formatting that told a human what each column meant is gone, and only the raw text is left.
Every one of those quietly corrupts an answer if it's read wrong. So we don't ask you to tidy the file up first, any more than a doctor asks you to feel better before your appointment. Handling that mess is the unglamorous work that turns a demo - one that only works on a clean sample file - into something you can trust with the spreadsheet someone actually exported last Tuesday. It's a big part of what we've built into Kotae, and it's the part that's hardest to get right.
Every diagnosis should be checkable
There's one more thing we cared about, and we'd argue it matters most.
When Kotae answers a question about your data, that answer traces back to the rows it came from. Ask to see the working and it's there - the equivalent of a doctor putting the scan on the screen rather than telling you to take their word for it.
And if it can't answer something reliably, it says so, rather than inventing a plausible number.
A chatbot that's occasionally unsure but always tells you so is far more useful than one that's usually right and never flinches. A doctor who is never uncertain isn't confident; they're just not looking. With numbers people make decisions from, confidence without accuracy is the worst possible combination.
House calls: private, fast, and yours
Beyond accuracy, three things matter for real businesses:
Privacy. This is the big one. Kotae comes to you: your spreadsheet is read on your own systems, not shipped out to an outside service. Most AI tools work the other way round - they cut your file into fragments and send those to a third party to be indexed before you can ask anything. Kotae doesn't. When you upload a spreadsheet, no outside model sees it at all. (For the technically minded: no chunking, no third-party embedding provider.)
Speed. We process a file in a fraction of a second. No upload to a parsing service, no queue, no waiting on someone else's API. Drop the file in and ask.
Cost. No per-page or per-document parsing fee - the kind that quietly compounds across thousands of files.
The general-purpose tools out there are good at what they're built for - reading scanned paperwork and faithfully reproducing how a sheet looks. Think of them as generalists. Kotae is the specialist: understanding native Excel and CSV files - which tables exist, what the columns mean, and getting the number right.
Your prescription for spreadsheet relief
Ask in plain language. "How many open tickets are assigned to support?" "What's the average order value by month?" No formulas, no pivot tables.
Get answers over your whole dataset - whether that's fifty rows or five thousand - not a sample of it.
Bring the file you already have. The totals row, the title at the top, the half-empty columns, the raw CSV straight out of your system: we expect all of it.
Mix data and documents in one question. Your policies and your numbers can live in the same chatbot, and the same question can draw on both.
The bottom line
Most AI assistants treat a spreadsheet like a document and skim it. Kotae treats it like a spreadsheet and calculates with it.
The title still holds. When the thing you depend on stops working, you don't need reassurance - you need someone who can fix it.
If your team keeps having to answer the same questions off the same spreadsheet, let's run Kotae on one of your own files. Book a demo at https://kotae.ai and we'll walk through it with your spreadsheet, not a tidy sample of ours.
Consider this your appointment.
The team behind the treatment
Kotae's spreadsheet capability was built by:
Duy Cung - AI Engineer
Dr. K.P. - AI Engineer
Alice L. - QA Tester
Tony V. - Product Manager
And yes, there really is a doctor on the team.