An AI analyst for your bills, with every number checked

Ask a chatbot where you could save money on cloud and AI, and you get a confident list. Some of it is right. Some of it is made up. And the numbers are whatever sounded good.

That's not something you can put in front of a CFO. So when we built the AI analyst into RootSpend, we started from one rule: the AI can suggest, but it never gets the last word on a number.

What it does

Open Recommendations and click Analyse with AI. You choose:

• the period, from 7 to 90 days, compared with the same number of days before;
• what to look at: everything, or some tools, categories, teams or projects;
• how deep: quick checks only, or a deep dive with the AI analyst.

When it finishes, you get findings, ranked by what they save each month. Each one says what to do, the steps to do it, how risky it is, how sure RootSpend is, and how much work it takes.

Here's a real shape of one, from our sample company:

Move steady Snowflake usage to a capacity commitment. Snowflake runs nearly every day at a steady level. A one year commitment prices that baseline lower, about $1,423 a month, with no change for the people using it.

Two layers: quick checks, then the analyst

Quick checks are plain rules that run on your numbers. They look for spend spikes, idle warehouses, data nobody has read in months, Copilot seats nobody uses, AI work on a bigger model than it needs, and development spend that runs all the time. They are free and they always run.

The AI analyst goes further. It looks across tools at once, which a rule can't do well. It spots steady usage that a commitment would make cheaper, the same job paid for twice, or a team whose growth drives one tool's bill. It also explains the quick check findings in plain words, with steps for your setup.

Why you can trust the numbers

This is the part we care most about.

1. The AI never writes an amount. It writes placeholders like "spend on this service" and RootSpend fills in the real figure from your data.
2. Every finding has to cite its evidence. Each one points at the exact tools, services or resources it's about. If it cites something that isn't in your data, it's dropped.
3. RootSpend computes the saving. The AI only says what share of that spend a fix would cut. RootSpend turns that into money, and caps it at what you really spend.
4. Fixes get measured. Mark a finding as fixed and RootSpend compares the real spend before and after, 14 days later. A saving is something we can see in your bill, not a promise.

What the AI sees, and what it never sees

The analyst gets a compact summary of your spend: totals by tool, service, resource, category, team and project, how steady each cost is from day to day, and how many people have spend on a tool. A count, never who.

It never gets names of people, emails, raw bill lines, your keys, or anything inside your tools. Free text your team writes, like the reason a finding was dismissed, stays in RootSpend too.

Names of tools and resources come from your vendors and your team, so RootSpend treats them as data, never as instructions. Whatever text a resource name contains, the AI is told not to follow it, and its answer is cleaned of links and markup before you see it.

You stay in charge

• RootSpend AI is included, within a monthly allowance.
• Your own key from Anthropic, OpenAI or Gemini works too. It's billed by them, with no RootSpend limit.
• Off is one click. Analyses then use the quick checks only.

And the AI never acts. It can't resize a warehouse, cancel a seat or change a setting. People on your team decide, and RootSpend checks whether it worked.

Run it on a schedule

Save an analysis to run every day or every week, with its own scope and its own recipients. The data team gets the warehouse findings. Finance gets the whole picture.