xAI launched a marketplace for Grok Bot templates, and the one it built for itself is the interesting one. Haggle Bot is a procurement agent: it connects to internal systems like Slack and Ramp, finds unused SaaS seats, price-checks recurring purchases, and negotiates vendor contracts. In its first week, xAI says it identified over $100,000 in direct savings.
That number is xAI's, about xAI's spend, at xAI's scale — so don't anchor on it. What's worth taking is the list of things it checks, because those checks work the same way in a thirty-person company as in a large one. You can run them this afternoon without an agent.
The four places SaaS money leaks
Every audit I've run comes back to the same four. They're boring, which is exactly why they persist.
1. Seats billed for people not using them
The single largest category almost every time. Someone leaves, switches teams, or quietly stops opening the tool, and the seat keeps billing.
Pull the user list from each subscription and compare last-login dates against what you're charged for. Most tools expose this in an admin or usage report — and if yours doesn't, that's worth knowing too. Anything billed but untouched for 60+ days is where you start.
2. Two teams buying the same capability
Marketing has an email platform, sales has another, neither knows the other's contract exists. This isn't carelessness — it's what happens when purchasing authority is distributed and nobody owns the whole picture.
Compare categories, not vendor names. You're looking for two tools that do the same job, not two instances of one product.
3. Introductory pricing that quietly expired
Year one was discounted. Year two renewed at list. Nobody decided to accept the increase; it just happened.
Pull every contract's renewal date and current price, and flag anything where the price changed without a corresponding change in what you bought.
4. Per-seat plans where most seats only read
A lot of tools charge full price for people who only ever consume. If there's a cheaper viewer or guest tier and you have users who never create anything, that's a downgrade rather than a cancellation — easier to get approved, and often most of the saving.
Why manual audits drift
Most teams have done a SaaS audit at some point, usually when someone in finance got annoyed. It works. Then it decays.
The reason: spend isn't a number you set once. People join, leave, switch tools, start trials on a card. A spreadsheet that was accurate in January is wrong by April, because nothing in it updates itself.
This is the genuine argument for an agent — and it isn't intelligence. It's continuity. A bot that checks seat usage weekly catches a departed employee's licence in week two. An annual audit catches it eleven months late, and you paid for every month in between.
That's the thing worth paying for. The rest — comparison shopping, contract negotiation — is valuable but optional, and negotiation is something plenty of people would rather do themselves.
Run it yourself first
Before you connect anything to anything, do the manual pass once. Two hours, and it gives you the baseline you need to evaluate any tool claim later, including the agent's.
- Build the list from bank records, not memory. Twelve months of card and bank transactions. Memory-based lists reliably miss the two or three tools nobody thinks of as software.
- For each, record four things: monthly cost, seats billed, active users in last 30 days, renewal date. You cannot prioritise without all four.
- Sort by the gap between seats billed and active users. That gap × per-seat cost is your fastest win.
- Put renewal dates in a calendar with a two-month warning. Most savings are lost in the renewal window, not found in the audit.
After that you'll know whether an agent is worth wiring up. Three or four large annual contracts? A calendar reminder beats a bot. Forty subscriptions across distributed teams? The continuity is the point.
One caution on permissions
A procurement agent needs to see billing and usage. It does not need to transact. Grant read-only first, let it produce findings, and only escalate to anything with purchasing authority once you've checked its work against your own numbers.
Same principle as any agent with standing access: value comes from visibility, risk comes from authority. Those are separable, and the default should be the narrow version.
The honest read
The $100,000 is xAI's number about xAI's spend, and a company that size has waste at a scale most readers don't. Don't anchor on it.
The transferable part is the checklist. Unused seats, duplicate capability, expired discounts, over-provisioned tiers — those four show up in almost every software budget, and finding them doesn't require an agent, a consultant, or a new platform. It requires one afternoon and the discipline to read the bank statement instead of the vendor list.
What the agent adds is that it looks again next week. Whether that's worth granting system access is a decision you can only make after doing it by hand once.
Originally published at ChooseSaaS.
Top comments (0)