Find budget savings with AI in 4-steps (steal my method)

Marketing sent one file. Procurement sent another. Engineering wrote theirs in French.

In a recent ChatGPT workshop, I played a controller with 3 questions.

Which software licenses do we have?
Are we using them?
Are two departments paying for the same one?

So I asked each team for their data. Marketing sent one file. Procurement sent another. Engineering sent something else again, and one person wrote theirs in French.

Every tab was different.

Now imagine 30, 40 or 50 of those tabs. You can't review your spending like this. You can't even start a pivot table by department until you have everything together on one list.

That is where software waste – or any type of overspending – is hidden.

Plus, it’s not just overspending on software and goods – Not long ago, a CFO secretly told me they had fired their consultants.

Before, they used to pay thousands of dollars for a market entry analysis. Now this CFO gets the same result 10 times faster, for $20 per month. They just AI to do the market analysis, competitor analysis and benchmark analysis for them.

So today, in Part 2 of my budgeting mini-series, we’ll look at how to find budget saving opportunities using AI.

From 30 messy tabs to clear savings

Step 1 – Put every team's spending on to one list

Ask each team for the same things: tools, users, seats, fees and contract terms. Then ask AI to merge all the tabs into one list.

My prompt starts with "Consolidate all the tabs into one with the following information". Then I list the columns I want: department, vendor and product, licenses used, price per unit, total cost and utilization ratio.

What is a utilization ratio? Say you pay for 25 seats and 10 people use them. The ratio is 10 out of 25, or 40%. Just ask AI to calculate it. This column shows you where you pay for something that nobody uses.

Pro Tip – Don't just say "consolidate". Ask AI "What is the best way to do this quickly?" In my workshop it gave me three options: Power Query, VBA / OfficeScript (it even wrote the code, and the code was short and super simple to implement) or copy and paste if you only have a few sheets. Nobody in the session had thought of Power Query 😉

Plus, don’t forget that you can now use AI in Excel. Either using Microsoft Copilot, or by using the Claude or ChatGPT plugins. Like this, you can use AI to consolidate your data, and ask the AI to produce the formulas so that you are producing the same outputs and calculations in your spreadsheets every time.

Step 2 – Ask for 4 types of analysis

If you ask AI "where can I cut?" you will just get one shallow answer. But, ask for four types of analysis and you force it to go much deeper.

• Descriptive: tells you what happened. This is what usually comes back when you don’t push for more detail.
• Diagnostic: tells you why. This is where long tails can show up: E.g vendors outside the top 80% of your spend, which could be ones where you might pay for something you don't use.
• Predictive: Tells you what comes next, which you can then ask AI to help you turn into a forecast
• Prescriptive: Tells you what to do to bring your spending down. Again, you can ask AI for recommendations here too, that you can then add your own expertise and knowledge to.

I then ask for 3 things on top:

1. The calculations, shown in detail, so you can verify them.
2. Graphs, to show the main patterns.
3. And commentary, because reading it helps you understand your data and gives you insights for you and your team.

Here is one prompt example that you can copy, and change to fit your own business.

Step 3 – Test a better deal on your biggest contracts

Ask AI to model your yearly fees against month-to-month deals.

We all know that an annual deal gets a much better price than a quarterly or monthly one. We also know that suppliers have a habit of increasing their fees on renewal.

So get AI to model this based on your current contracts.

This could be as simple as taking your current contract terms, adding them all to an AI project (like a ChatGPT or Claude Project) so that the AI can reference all of them without you having to upload them as attachments all the time.

The use a prompt like:

Pro Tip – Ask AI to give you a detailed negotiation plan that you can use during renewal conversations to get favorable terms.

Speaking of renewals…

Step 4 – Set a review rule before every renewal

If you’ve created a project for all of your contracts, you can also use that data to set up renewal reviews.

Track when each deal ends, and set a rule for the bigger ones, like a review process for every contract over $10K.

You can ask AI to create the renewal calendar. Plus, you can get AI to:

1. Set this up this up as a skill so it runs the same review analysis in the same way every time you ask
2. Set it up as a scheduled task to automatically run for example once per month.

Just take the conversation you had in Step 3, and In ChatGPT or Claude ask:

3 additional tips:

1. Make sure to put the review date before the notice period ends, not on the renewal date.
2. If you’re creating a renewal calendar, ask AI to produce a calendar visual with a list of your upcoming renewals. Sometimes you will miss a renewal if it’s a line of text in a contract, but you are unlikely to miss a renewal if you can clearly see when it’s coming up on a calendar.
3. If you considering an alternative supplier for the service, get AI to search the web, or conduct some Deep Research (more on this shortly) on alternative suppliers that offer the same service.

4 other places to check for overspending

1. Unused software seats: Compare the seats you pay for with the seats people use. Most software tools will have an admin portal where you can see what licenses are assigned to who. Plus, some also have usage metrics so you can see not only who has a seat, but their consumption as well.

2. Same tool bought by two teams: Check whether two departments pay for the same license, where one joint purchase would have been better. For example, a lot of AI tools have both individual plans and team plans. You don’t want a few members from one team paying for individual licenses, and another team on a shared team license.

3. Double bills and claims: There are times where you will have two genuine $500 invoices to the same contractor on the same day for two different projects. But what if you have amounts that are very similar that could be duplicates? You want to be able to identify both. So download your transactions, and ask AI to identify:
​
A) Exact duplicates from the same supplier / teams
B) Potential “Fuzzy Match” duplicates where the transaction data looks mostly the same, but the amounts are slightly different.

4. Outside research fees: Remember the CFO from the start? AI Deep Research now covers the research they used to pay consultants for: market analysis, competitor analysis and benchmark analysis. Test it on one report you would normally outsource, then compare compare the two.

AI Deep Research is a mode where an AI assistant works like a research analyst instead of answering in one shot. It breaks your question into sub-questions, searches and reads dozens or even hundreds of web pages (and your own files if you give it access), then writes a structured report with citations. A run takes a few minutes to about half an hour, so it fits big questions like vendor benchmarking or market pricing better than quick AI responses or internet searches.

ChatGPT Deep Research, Gemini Deep Research and Claude Research all offer a tool like this.

When I do research, I often ask the same question to all 3 tools so that I can compare.

For me, Gemini Deep Research produces really good research. But, you’ll have to decide which tool works the best for you.

Whichever tool you use, check the cited sources before relying on the output, because the reports can still contain errors.