Question. I have been using MS apps most of my life. I am on Windows 11, and use Edge and Office 365 etc. Most of your upgrades have been focused on Google Sheets and my questions are 1) are your upgrades to Google sheets integration making it closer to Excel or 2) is Excel application falling further behind what is offered if I used Google Sheets?
If you are telling my experience would be “enhanced” if I used Sheets, I might consider switching. But if you are suggesting that the experience, even if better with Sheets is just a jump ahead until Excel catches up, I will continue using what I have. I have been developing spreadsheets used by others who have either Excel or Numbers (MAC system) and am not anxious to change their experience as well.
But I COULD continue using Excel for everything other than Tiller, if indeed the view is worth the climb
@larry.collings Hi Larry, great question. I’m an Excel user too. I think these new enhancements are bumping up against the inherent constraints of each tool platform’s core architecture approach. The Tiller team typically develops first in Sheets and then when the solution is completed begins to develop the Excel version. The core Tiller templates are available for both Sheets and Excel. The new enhancements being released are more complex and require different technical approaches to be applied within each platform’s development architecture. As stated below in the two posts, Tiller plans to bring these enhancements to Excel, although the implementation may be different given the differences in the available platform development architecture.
I personally choose to wait for these enhancements to be available in Excel. From time to time, I have added a Sheets template and ran parallel Excel and Sheets templates. This allowed me to try out the new enhancements and decide it I want to switch over to Sheets. I have found this a great option that is available..
Thanks for the response. If I were to run a “Sheets” in parallel with Excel (and I as are you waiting for Excel enhancements to happen) because of the length of time most financial institutions keep data on line, what would the mechanism be to migrate older excel data into the new “Sheets” environment so whichever I stayed with, I had access to all historical data? Or when you loaded the Sheets template, did you not concern yourself with older data and just “got what you got”?
I believe if you had the institutions connected and create a new sheet and link the accounts then the old data would load. It won’t have all the categories you’ve created and categorized and the old balances. You might be better off trying to migrate your sheet. All your data will still be in excel and you won’t unlink it but all your work into the categories and balance history will be there.
@valerie1 nothing changes with the handling of individual’s data by Tiller unless you choose to use AI Suggest, we do send the data to an AI tool for processing, but that is completely opt in. The Description Match feature does not send data to a 3rd party AI tool so it stays contained within Tiller’s ecosystem only.
@larry.collings at this time these latest updates to Google Sheets are not closer to making it out for Excel and some of them will likely be handled differently (e.g. Direct Fills) simply because the technology is different between the platforms in ways that we don’t control.
The Excel Add-in is performant and is Microsoft’s standard for interacting with a workbook in these ways. Add-ons for Google Sheets are not performant nor recommended by Google at this point for the heavy lifting Tiller requires. They are not the best option for scalability, sustainability, and reliability, which is necessary for Tiller to continue supporting Google Sheets.
We’re focusing on playing to the strengths of each platform. An add-in is better for Excel, which may mean Excel doesn’t get the exact experience that Google Sheets has. The fills for Excel are lightning fast, with Google they were slow and prone to crashing resulting in data corruption. Direct Fills fixes that for Google Sheets, but no such problem existed for Excel.
There may be ways we can offer automatic fills for Excel, but there are some limitations such as requiring that the workbook be completely closed and stored on OneDrive. These are limitations imposed by Microsoft. We expect to be able to bring the new AutoCat features to Excel, but we don’t have a specific timeline on that just yet. Hopefully that helps.
It’s great to see the AI categorization feature arriving. I actually work at OpenAI on Codex and I was playing with a skill for Codex to do exactly this to my tiller spreadsheet so I’m glad to see an official version.
I suggest adding some mechanism of indicating which transaction rows were categorized by AI for a few reasons:
It will help us tiller users gain confidence in the feature
It allows some HITL (human-in-the-loop) which is optional (human review post-facto)
Importantly, since (I hope) the model’s decisions are based on earlier categorizations of a transactions description (within the same sheet), it allows some way for the bootstrapping signal to help the model when new transaction types start appearing (for example, I just tried it and “Pacific Pipe” was sensibly categorized for me as “Household”, but Pacific Pipe is a climbing gym in Oakland and should be categorized for me as “Kids Activities” – catching the first few instances of a new description will help the model do much better).
Also, reason 4: If the AI-category-assigned rows are marked it’ll help distinguish “ai failed” vs “auto-cat or description match rule needs adjustment” (just noticed multiple that I thought initially were the former, but after looking closer are the later).
Great feedback here! And totally agree. Did you see my recommendation for adding a review column? Even if you don’t want the manual review, the Categorized By column does exactly what you propose, showing which part of the process captured the Transaction. It’s a default column out right in the Foundation Template if you created it today, otherwise you’ll need to add it manually and it will fill going forward.
As for the Pacific Pipe example, the Description Match should catch that one and categorize it the same way next time and continue to use “Kids Activities” going forward so long as you don’t categorize it differently in the future. No model involved in that AI Suggest (that uses an LLM) is only involved if it’s a novel transaction (no prior examples of same full description) or you’ve categorized the same merchant inconsistently. If it’s appeared before and it’s inconsistent the model will use patterns in your prior categorized data to try and choose the right category (i.e. is it a snack at the grocery store vs gas).
No, I didn’t see the guidance about the CategorizedBy column. Wish I had noticed that before the first AI categorized run.
That’s a bit more work than I personally care to put into this problem. I don’t find it worth the time to attempt to arrive at 100% accuracy with this stuff. I find it better to just find ways to surface the errors that move the needle alot in terms of impact. IOW, I don’t worry about whether everything is perfectly categorized. I spot check monthly spend by category (and previously anything left uncategorized) and go check on anything that seems to spike totals one way or another.
The CategorizedBy column is useful.
Since I’m guessing your system is kind of set up to populate columns by name if they exist, I’ll also request CategorizedByAI-Confidence which could be a float [0.0, 1.0] or just (low|med|high).
Presumably the input signals to the model when choosing categories are (at least) description, user categories and as much previously assigned transactions → category, you can give the model guidance to choose a 0-1 confidence score and bias lower if evidence is sparse for a previous categorization for this description. That would help with being able to spot check CategorizedBy == “AI” to the low confidence rows.
AI Suggest (that uses an LLM) is only involved if it’s a novel transaction (no prior examples of same full description) or you’ve categorized the same merchant inconsistently. If it’s appeared before and it’s inconsistent the model will use patterns in your prior categorized data to try and choose the right category (i.e. is it a snack at the grocery store vs gas).