Why You Need to Meet Arcwise’s AI Assistant
When we launched our first product, our AI Assistant Chrome extension (shameless plug, but hey, you’re here), we envisioned it as a natural first step towards solving the collaboration problem that plagues spreadsheets, and more importantly, the teams that use them.
As founders and spreadsheet geeks, we’re pumped about LLMs. And there’s a lot to be excited about.
If we had to break it down to three points, here’s what we’d say:
LLMs work best when it creates something you can quickly evaluate for accuracy.
LLMs are great at generating code and spreadsheet code (formulas) in particular.
There are way more spreadsheet coders than other coders. There’s unlimited opportunities to impact actual work being done.
We also know LLMs limitations because we’re constantly testing them. We’ll tell you more about what LLMs really do for Arcwise — and for you as our users.
We Knew Early that LLMs Were Part of Our Story
When the Arcwise team got together to create our MVP, we knew our problem statement on a deeply personal level. Businesses love spreadsheets; data teams hate them. And even though they hate them because spreadsheets are ancient technology that creates busy work for everyone, we’re not ready to abandon them yet. We all need to find a way to work with them and in them, and we need to do it together.
Here’s why LLMs were and are so exciting for our team and our mission by making spreadsheets less painful to work with:
LLMs can take on an annoying time-consuming task (like date parsing) so you don’t have to.
LLMs can give you a first pass at the right formula to use so you don't need to battle through search engine results, read articles, and hope it works.
LLMs can help explain what a sheet is doing, so you understand why what you’ve done and you can repeat it.
We’ve used LLMs to level up novice spreadsheet users and get them on a faster (read: less harrowing) path to top-quality reports.
This is also where we’re most excited about the applications of LLMs: we see LLM as a teacher that shortens the learning curve and levels up users faster.
Improved reporting and shorter learning curves = better outcomes for everyone, including the data team. (The data team still hates your spreadsheet, so naturally, Arcwise couldn’t call our product feature complete with just an AI Assistant.)
What We’re Doing (and Hope to Do) for Our Users with LLMs
Our goal is to make advice reliable enough that a novice spreadsheet user can trust the results. And it does largely that.
With our AI Assistant:
You’ll get small but meaningful assistance directly in your sheet, when you need it.
You’ll automate menial tasks so you can make your work look great with far less effort.
You can get some broader assistance for structuring reports that you might find value in but is still very much “under construction.”
So, how are we going to do this? We’ve already started training LLMs with expert-created spreadsheets and getting feedback on the results that our users find useful. We’ve also internally crafted prompts to serve as the seed for different tasks. But we’re also diving deeper into some use cases where we see LLMs both excelling and pushing its own boundaries:
Automated Analysis
Most of the time (all of the time?) spreadsheet users are trying to make some ad hoc reports. There's no template to follow. They just need to get the work done. Right away. Before they start questioning whether this report is even needed.
LLMs slot in perfectly. It won’t get the full report correct, but it will get the shape of it right, and get the right data. It's like taking over where someone left off instead of starting from scratch.
Our users have been using our automated analysis feature constantly. The result is usually not right, but because they can get the result in the spreadsheet they can (and do) play with it until it is.
Data Cleaning
Unclean spreadsheet data is the definition of chaotic evil. And it’s a popular business truism: “bad data in, bad data out.” But what about the people who have to clean the data? No one ever thinks about them. But LLMs can.
Our team and our users use LLMs for data cleaning all. the. time. It is particularly good for cleaning dates, which you can (and our users do) clean up in a single prompt.
Our users have said things like: “It completely changes the way I use Google Sheets” and “It has literally saved me dozens of hours in the last month alone.”
Formula Assistance
LLMs are particularly good for formula/spreadsheet coding assistance, and not just the simple ones.
If you’re writing a formula and you aren’t exactly sure, an LLM can get you there much, much faster with less guessing. LLMs can even output complex formulas (like string or date parsing).
We use this all the time when building spreadsheets. It’s really good at finding the right string parsing formula, like extracting first names, email domains, and date parts. It even gave the team the right formula for finding a population standard deviation when they would have spent hours trying to figure it out through trial and error.
Spreadsheet Explaining
We’re also here for a use case that LLMs aren’t great at yet, but it’s better than the alternative.
You get handed a spreadsheet that you’ve never seen before. It is the spreadsheet equivalent of a Gorgon with hundreds of beady eyes staring back at you as you’re unsure where it begins and ends.
LLMs can give you a decent first-pass explanation of what the sheet is — sometimes. Someday, this is going to help you understand the spreadsheet without the risk of breaking someone else's valuable work.
LLMs are Just the Beginning
Tools like LLMs fit perfectly into our company’s mission: to make Google Sheets even a little bit less painless to use.
In our view, and in the experience of our 10,000 and counting users, Arcwise AI Assistant does just that.
We see the serious potential for LLMs specifically and AI generally, and we understand there are some limitations. Do we think that LLMs are going to add $2.6 trillion to $4.4 trillion in value annually across 63 use cases like this breathless McKinsey report?
Probably not.
What we do know is that it could have helped the second-year McKinsey researcher who did all the work for that report get it done without staying up til 3 AM every night.
So, give the Arcwise AI Assistant a try. Even if your use case stays limited to data cleansing, we know you’re going to find serious value in our (free!) tool. We’re on a mission to make spreadsheets better, and we’re just getting started.
We’ll have more updates on how we see LLMs are the ultimate teacher and the steps we’ve taken to make it happen.