AI Use for Tiller

Conversation Helpful Tips
1

Here are cleaned-up, stronger prompts based on the video.

1. Subscription Audit Prompt

Search my Gmail for recurring subscriptions, memberships, software services, trials, renewals, invoices, receipts, and billing notices.

Create a table with:

  1. Service name
  2. Estimated monthly cost
  3. Estimated annual cost
  4. Billing frequency
  5. Last billing date found
  6. Evidence from email subject/snippet
  7. Cancellation priority: High / Medium / Low
  8. Reason for cancellation recommendation

Prioritize subscriptions that appear unused, duplicated, trial-based, expensive, or unnecessary.

After the table, give me the top 10 subscriptions I should review or cancel first, ranked by likely savings.

2. Full Subscription Table Prompt

Search my Gmail for all subscriptions, memberships, recurring charges, invoices, receipts, renewals, and free-trial ending notices.

Create a complete table of everything I appear to be paying for.

Include:

  1. Company or service
  2. Category
  3. Monthly cost
  4. Annualized cost
  5. Billing frequency
  6. Last payment or renewal date
  7. Whether it looks active
  8. Any cancellation or account-management link found in the email
  9. Notes

Then summarize my estimated total monthly subscription spending and identify any items that look suspicious, duplicated, forgotten, or worth reviewing.

3. Credit Card Rewards Audit Prompt

Help me audit my credit card rewards and benefits.

Here are the cards I have:

  • [Card 1]
  • [Card 2]
  • [Card 3]

For each card, tell me:

  1. Current point or cash-back value
  2. Best redemption method: cash, travel, gift cards, transfer partners, or statement credit
  3. Annual credits or perks I may not be using
  4. Benefits such as hotel status, lounge access, travel insurance, purchase protection, extended warranty, or streaming/food delivery credits
  5. Any perks or credits that may expire soon
  6. The best card to use for groceries, gas, dining, travel, online shopping, utilities, and everyday purchases

Then tell me where I may be leaving money on the table and what changes I should make going forward.

4. Discount Code / Shopping Prompt

Before I buy from [store name], find the best legitimate ways to save money.

Check:

  1. Promo codes
  2. Newsletter signup discounts
  3. First-time buyer offers
  4. Referral codes
  5. Cash-back portals
  6. Credit card offers
  7. Price-match policies
  8. Student, senior, military, or membership discounts
  9. Free shipping thresholds
  10. Sale timing or upcoming promotion patterns

Tell me what to try first, what is most likely to work, whether discounts can be stacked, and how much I might save on a purchase of about $[amount].

Best practical order: run the subscription audit first, then the credit card audit, then use the discount prompt before any purchase over $50.

2

Intriguing! Which video are you referring to? How do you get it to prompt before you make purchases?

3
  1. Why base your prompts on estimated anything? 1 and 2 prompts do. And they have nothing to do with tiller. Only Gmail scraping.
  2. Are prompts 3 and 4 web chatbot generic queries? Again no tiller template is mentioned.
  3. Even if you meant build these tables (1&2) on generic estimated costs and then query them with prompts 3&4 what has that got to do with the information tiller bank feeds provide? The feeds are deterministic not estimated.
  4. I’m afraid you have me confused. What level of AI subscription are these based on? And do you utilize tiller data?
4

My apologies — I should have tailored these to focus specifically on Tiller, which is what I’m actually doing.

To answer your question: I use both Claude and ChatGPT, and I have subscriptions to both. I use them for different things. That said, both offer Gmail connections you can set up, so this could work as a Gmail-only solution.

However, I do most of my work in Claude’s desktop application, since it lets me connect not only to Gmail but also to my Google Drive — and through that, to my Tiller spreadsheets. That’s the setup the new prompts I’m sharing are built for.

I’ve been experimenting a lot with using AI alongside Tiller to see how it can help. It helped me create my budget, it runs validations, it tracks price increases — quite a few activities. I’m slowly building up a library of prompts for all of this.

Any questions or thoughts, please feel free to ask.

REVISED FOR TILLER

  1. Subscription Audit Prompt (Tiller)

Read my “[Tiller 2025]” sheet. Using the recurring-charge tags in the Transactions tab, find every subscription, membership, and software service.
Build a table with:

Service name (merchant)
Amount per charge
Cadence (monthly / annual)
Annualized cost
Last charge date in the sheet
Category it’s assigned to
Cancellation priority: High / Medium / Low
Reason (unused, duplicated, price increase, former free trial now charging, or expensive)

Flag any charge that rose vs. prior months, any overlap between services covering the same ground, and any trial that’s now billing. After the table, give me the top 10 to review or cancel first, ranked by likely annual savings. Surface them as candidates — don’t cancel anything.

  1. Full Subscription Table Prompt (Tiller)

Read my “[Tiller 2025]” sheet. From the recurring-charge tags in Transactions, build a complete table of everything I’m paying for on a recurring basis.
Include:

Merchant / service
Category
Monthly cost
Annualized cost
Cadence
Last charge date in the sheet
Whether it still looks active (charged within its expected cadence)
Notes

Then total my estimated monthly and annual recurring spend, and flag anything that looks duplicated, forgotten (no recent charge), suspicious, or worth reviewing. Treat the sheet as read-only — surface, don’t change.

  1. Credit Card Spend + Rewards Audit Prompt (hybrid)

Two parts.
Part A — from my “[Tiller 2025]” sheet: using the Transactions and Categories tabs, break my spending down by category (groceries, gas, dining, travel, online shopping, utilities, everyday) and by account/card, so we can see where my money actually goes.
Part B — for the cards I hold below, use current published info to tell me:

[Card 1]
[Card 2]
[Card 3]

Best redemption method (cash, travel, transfer partners, statement credit)
Annual credits or perks I may not be using
Benefits: lounge access, travel/purchase protection, extended warranty, streaming/food-delivery credits
Any perks or credits expiring soon
Best card to use per category — matched to my actual Tiller spend from Part A

Then tell me where I’m leaving money on the table and what to change. Don’t recommend moving investments or opening/closing accounts as advice — just show what the data and card terms say.

  1. Discount Code / Shopping Prompt (hybrid)

Before I buy from [store name] (~$[amount]), find the best legitimate ways to save.
First check my “[Tiller 2025]” sheet: have I bought from this merchant before, at what price, and on which card? Use that to spot the best card to pay with and any price baseline.
Then research: promo codes, newsletter/first-time-buyer offers, referral codes, cash-back portals, current card offers, price-match policy, student/senior/military/membership discounts, free-shipping thresholds, and sale-timing patterns.
Tell me what to try first, what’s most likely to work, whether discounts stack, and estimated savings on ~$[amount].

5

Just updated for you !

6

Added an updated prompt series , How do you get it to prompt before you make purchases? What are you asking here,? purchase what? Glad to discuss

7

Very interesting Post @blaine.atkins Blaine. You have some useful prompts and use cases for managing your budget and financial expenses. Claude sounds pretty powerful for accomplishing your objectives. I could see having a “prompt” for estimating year-end under / over budget (categories) based on YTD expenses. Another “prompt” for suggested reallocation of budget from one category to another.

8

Clint you are correct! I had it schedule a monthly, quarterly and annual review so as to continue the progress and refine it. I think the changes Tiller is making to add AI is fantastic, but I use the AI’s to evaluate my finances 360, so a lot of detail is outside of Tiller, which I think is appropriate.

10

I deleted my first reply to you because it was harsh and I decided it wasn’t constructive.

Based on your experience with AI:

  1. can it hallucinate or timeout or have latency delays?

  2. are you saying that AI prompting outside of Tiller has more value than prompting what Tiller offers?

  3. Do you ever have regrets with sharing your financial data with AI companies? Are the results that good that you welcome it with open arms?

Thanks David

11

Really cool Blaine. Thanks for sharing. I’m not very good at prompting but I’m going to give these a try. I also appreciate that you didn’t focus completely on Tiller. I know that’s why we’re here, but your Gmail prompts are very cool too. You seem to really have a nice grasp on the value of AI right now.

12

David, the 2nd half was harsh. I agree with you, and I thank you. But your 1st was right on. It actually force is forcing me to go back and review my prompt. I Tend to think people are much more knowledgeable than they are. So I need to bring this down to basics and make sure I build in all the guardrails and provide you the correct information to do this properly. Onto your questions.

Hallucinations tend to be overhyped. You need to be conscious of it. I use two AIs to evaluate everything I do.

High timeouts and latencies. I haven’t seen anything like that in anything that I do. I’m not sure how you determine that. I use Claude at the desktop level, and I use ChatGPT web based.

I think prompting out of tiller has value because it gives me more control, and that’s me, over what I’m trying to get. Also, I also look at other data, not just Tiller, which combines a lot of it. But I review a lot of my statements. I review a lot of my information. Any of my stock stuff, all of that is being evaluated on a regular basis using AI using AI.

It’s interesting. I have guardrails built into into Claude and ChatGPT about my data. They know what I allow, what I don’t allow. That said, I feel relatively comfortable because I personally feel that none of your data is safe no matter what you may think. That’s just me.

David, always glad to answer questions, help out in any way I can. That’s why I’m doing this right now. Being retired, have to have something to do.

Blaine

13

Thanks. As I just mentioned to David, I’m going back and looking at my prompts to make sure that the the prompts that I give you will work for you effectively. So I will go back and clean these things up and republish them. My background is IT. I recognize that everything changes over time, and level of resistance to AI is not unusual. But history shows us that resistance is futile.

14

Blaine,

I want to appreciate the time and effort you’re putting into sharing these workflows. It’s clear you’re genuinely looking for creative ways to help the community get more out of their data, and that drive to contribute is awesome to see.

I use AI for a lot of tasks myself, and it’s incredibly powerful for what it’s built to do. But as a database professional, I want to gently share a foundational warning about where the technology stands right now, especially regarding financial modeling.

At their core, Large Language Models are built to predict the next logical token (or word) based on patterns in language. They are probabilistic text engines, not deterministic math engines. Because they are guessing the most likely next word rather than strictly computing relational data, they are fundamentally prone to calculations that are estimated rather than exact, alongside issues like latency or timeouts when reading flat files.

For a financial ledger (like the one Tiller builds) an almost(!) accurate number is structurally a wrong number. Until these models can guarantee 100% mathematical accuracy without the risk of hallucination, they really have to stay on the sidelines when it comes to acting as a reliable budget or financial counselor. AI cannot audit AI accurately. It’s prone to mistakes. My hope is for the future of these AI models to use SQL and DAX with tabular datasets to come to proven accurate results.

Furthermore, because our financial data is so deeply personal, we have to be incredibly protective of sending it out to third-party cloud engines that don’t operate under a strict fiduciary duty to protect us.

I think it’s great to explore new tools, but as an online community, we owe it to each other to highlight these structural pitfalls. I look forward to seeing how we can all keep building more robust, precise, and secure solutions together!

David

15

@blaine.atkins @freshman.david

I fall into this category with AI. Even though I have a technical background, the technology is moving too fast for me to even try to keep up with at even a basic level.

I thought I was hallucinating reading this and other AI threads :laughing: I didn’t even know it was a term being used with AI.

It is great to see the depth of knowledge and experience some Tiller users have and are wiling to share through education and community solutions.

16

Clint, that’s one of the reasons I always loved IT and now enjoy working with AI — the technology keeps moving, and it keeps me challenged.

Hallucinations are real, but I still follow the old principle: trust, but verify. I don’t see AI as something to blindly rely on, especially with financial data, but I do think it can be a very useful assistant when the results are checked carefully.

I also expect Tiller will continue adding more AI-related capabilities over time. Maybe conversations like this can help point toward where users would find real value.

For me, working through these ideas is useful either way. If something eventually helps shape a future feature, great. If not, it still helps me build my own skills and better understand what AI can and can’t do.

17

Sorry for the slow reply.

I watch a lot of YouTube videos on AI ideas and workflows. The video I was referring to was not Tiller-specific, but it gave me the idea to think about how AI could help review or question purchases before they are made.

I’m now working on a more Tiller-focused version of the idea, using Tiller data and some supporting code.

Here’s the video that sparked the idea:

18

Clint FYI Why do AI models hallucinate? | Claude by Anthropic

19

Clint you have some really good business use cases for Blaine to try. It would be a tough to do though based on my understanding of how AI models work. I’m willing to see what he can come up with based on Tiller’s spreadsheet data.

If you ask me though the way to approach these are: I would say you could model them in a relational SQL database. It would be challenging also but doable in my estimation.

Yet like I said I’m willing to see what AI prompt could be engineered by Blaine.

Blaine I just joined in April of 2026 so if you could base these prompts on 2026 Tiller foundation template data I would be much obliged. That way I can test them too.

20

Thanks @freshman.david David for your comments. I appreciate the feedback. I’m not an AI user. @blaine.atkins Blaine’s work with AI and his use cases got me thinking about what I’d like to have within Tiller. I certainly realize some functionality is not easily doable within the spreadsheet tool. For me, my use cases mentioned are a monthly mental exercise for me to review my budgeted categorizes and forecast what the year-end will likely be over or under budget. I focus on the big hitter known expenses and the big unplanned expenses. I’ve worked with SQL db’s and BI tools in the past. I don’t see that working out for me with my personal financial Tiller data but see the power and potential of them. The Budget Plan community solution by @jpfieber is an awesome tool for setting up and managing changes to the budgeted categorizes. It does provide some great options for creating new or changing existing budgeted amounts. The Budget Plan comes the closest to what my specific use cases mention. Maybe Joseph has some ideas :thinking:

21

Hi David, Clint, and JP

I’ve been tinkering with the Claude prompts for Tiller based on David’s Feedback and wanted to open a conversation with the three of you, since you know this data well and I’d rather think them through with you than build in a vacuum.

A few things I’m genuinely unsure about, and where you’d know better than me:

The 2026 Tiller Foundation template — I’m honestly not that familiar with it yet. Is that what you’re each on, or are you running something customized? If anyone can share a copy of their 2026 Foundation sheet, or point me to Tiller’s official version, I’ll get familiar with the structure before I build anything, so what I come up with actually fits your sheets rather than just mine.
Roughly how many rows are in your Transactions tab? That turns out to be the biggest factor in whether a run goes smoothly or bogs down.

Do you use Claude — and if so, free tier or paid? One thing worth flagging up front: the Google Drive connector these prompts rely on is paid-plan-only, so on free Claude you’d attach a CSV of your Transactions tab instead of connecting live. Good to know before anyone spends time on it.

Google Sheets or the Excel version of Tiller? What I’ve built so far is Google-native.
Do you actually use the Monthly/Yearly Budget tabs with real category targets? The projection and reallocation ideas only work if there are budget targets to compare against.

How comfortable are you sending transaction data to a third-party AI? Totally fair either way — it just shapes whether this is even worth it for you.

For context on where I’m coming from: I tested an earlier recurring-spending prompt on Claude Pro, with the connector and code-execution turned on, so results may look different on other setups.

Blaine

Greatly Appreciate your feedback, I am learning a lot from this conversation!