Speaker 1
Hey, everyone. Welcome to taking the BS out of buying SaaS. I'm Jacob Leitman, head of Tropics procurement services team. We're here to make software buying and renewals a little less painful, one quick tip at a time. Today, we're talking about how to combat scaling AI costs. Let's get into it.
If you're seeing your AI costs by or out of control, you're not alone. Here's what's happening. We just got comfortable with cloud pricing, and now everything's shifting again from traditional SaaS to AI. And this time, the cost risk is different.
Your suppliers are moving away from predictable seat based pricing to consumption models, credit systems, and outcome based pricing. This means your costs are becoming unpredictable. A Salesforce license that was a hundred fifty dollars per user per month is now a hundred fifty dollars per user per month plus whatever AI features your team consumes. HubSpot on the other hand gives you AI credits that you might burn through in two months or you might never use.
The problem isn't that costs are just higher, it's that you can't forecast them as accurately. When you're scaling AI adoption, that's unpredictability that can kill your budget. Here's how great operators are managing this. I typically segment AI spend into four buckets, LLM licenses, LLM APIs, AI native suppliers, and AI features from legacy SaaS suppliers.
Let's start with LLM licenses.
For these, negotiate what you can. Everyone thinks LLM pricing is nonnegotiable, but that's not true. With OpenAI's new pooled credit model, there is variability, especially at the four hundred pooled credits per user tier. Push hard, especially in the two hundred, five hundred user range.
OpenAI will say it's impossible to negotiate, but be If you're a Google Workspace user, threaten to switch to Gemini. That's a very compelling threat, and that's real leverage now. Second, optimize your licenses. Don't pay for six hundred credits per user with OpenAI when your team only needs five hundred or four hundred.
With Claude on the other hand, don't give everyone Claude code licenses. They're ten dollars more per month than chat licenses. Only developers need code licenses, so make sure these are allocated appropriately across your user base. Third, set hard spend limits.
This one is critical. In ChatGPT Enterprise, users can exceed their credits. They'll get hit with seven cents per credit and overages. Set permissions up to prevent this.
In Claude's new pricing model, configure spending caps at the organization level, seat tier level, or individual user level. This will prevent unexpected bills and keep usage low and keep costs low as well.
Fourth, train your users on efficient token consumption. With ChatGPT, make sure your prompts are detailed and comprehensive instead of back and forth. No more thank you messages eating up credits. If a designer asked ChatGPT to iterate on an image twenty times at five credits each, that's a hundred credits burned.
Train them to write one detailed prompt instead. This will be much more efficient and cost effective. With Clawd, it works a little differently. You're charged at current API rates. Think of tokens as the units AI uses to process text. A hundred tokens equals roughly seventy five words. To keep costs down, use less expensive models for simple tasks, be specific and concise in your prompts to avoid back and forth, and use features like prompt caching and projects to reduce repeated token consumption.
Let's talk about API costs. This is where your cost can really scale. Here are a couple strategies to combat this. First, diversify your suppliers.
Don't put all of your API usage into one supplier like OpenAI. Test different use cases across providers. For example, Gemini excels at image extraction. Different providers are better at different things.
This gives you negotiation leverage and protects you from price hikes long term. Next, evaluate purchasing channels. You can buy OpenAI directly from OpenAI, but you can also purchase through Azure. You can purchase Anthropic through AWS Bedrock.
Models can be surprising and have better latency when you purchase through a hyperscaler, which can allow you to use less expensive models to satisfy your use case. Last, commit small and renew early. We recommend just doing one year deals with API providers and only committing to a small percentage of your forecast. The discounting bands are so wide on these API providers, so there's rarely an advantage to committing to seven hundred fifty thousand versus five hundred thousand.
Both are gonna get the same if not or at least very similar discounts. If you burn through your commitment early, you can renew early.
This keeps you flexible as new models emerge as well. The next bucket of AI spend is native AI tools. Think Play or Cursor or Glean. Many of these companies are hungry for market share.
Make sure you negotiate hard upfront. Their pricing models are still evolving. If you don't like their model or don't think it makes sense for your business, you can suggest alternatives. We've seen a lot of AI native tools offer different pricing models to different customers.
Next, you're gonna wanna get protection from commercial model changes in your contract and not just price increases. Everyone thinks it's a success if you negotiate a a three percent renewal cap or something like that, but these AI native tools are going to completely overhaul their their pricing models year over year. This is a constantly evolving space, So you're gonna wanna get commercial model protection in your agreement as well. The last AI spend cost center is legacy SaaS companies adding AI.
Salesforce, Slack, HubSpot, Zendesk, they're all bolting on AI features and trying to charge more. Do not accept these uplifts. Make sure the provider shows you ROI. Just because they spent more on r and d doesn't mean you should pay more if you're not getting the value.
And lastly, use consolidation as leverage here. These companies are desperate to show AI revenue. The reps are incentivized to sell AI. Use that to negotiate costs down in other areas of your agreement or use the threat of moving to a newer AI native tool as a way to keep costs reasonable with your existing SaaS provider.
So here's your action plan. Set spending limits on all your LLM tools today, audit your licenses, cut over provision credit types and seat types, try to test one use case with an alternative API provider this month, you might be surprised.
If you do have an API commit, make sure you're only committing to one year agreements at sixty to seventy percent of your forecast, not a hundred percent. And lastly, train users on token efficient prompting. And one last thing, if you've built AI into your product, as AI adoption scales among your customer base, your cost will scale as well, but your revenue might not. Get ahead of this now by being intentional about how you go to market and price your own AI offering. Thank you.
