Every tool in your stack has shipped an AI feature this year, and a good number have raised their prices alongside it. Some of these genuinely change how the tool works. Most are a text box wired to a general-purpose model, wrapped in your vendor's branding, sold at a premium. Telling them apart matters, because the aggregate cost of saying yes to all of them is substantial.
The question that separates them
Does this feature use data only this vendor has? That's the whole test. A CRM that suggests which accounts to prioritise based on your own closed-won patterns is doing something you genuinely cannot replicate elsewhere. An email tool with a button that writes subject lines is offering you a worse interface to a model you can already access directly. The first is worth paying for. The second is a feature you're renting at a markup.
What's actually worth the money
In practice, the tools that have earned their place in client stacks this year cluster in one area: turning unstructured internal material into something searchable and usable. Call recording tools that surface recurring objections across hundreds of conversations. Support tools that cluster tickets by theme. These do something a human team genuinely cannot do at that scale, using data that's already yours and currently unread.
- Worth it: analysis of your own unstructured data — calls, tickets, product usage
- Worth it: anything removing a genuine bottleneck you can name today
- Not worth it: a generative text box bolted onto a tool you already pay for
- Not worth it: anything whose demo impressed you but whose job you can't describe
The budget discipline that helps
Adopt a rule: any new tool must replace an existing one or eliminate a named recurring task, and someone has to own it by name. Without that, stacks accumulate. I've reviewed a stack this year with three tools performing overlapping jobs, each bought by a different person in a different quarter, none fully rolled out. That's a common outcome, and this year's wave of AI features accelerates it considerably.
Run the trial properly
Don't evaluate on the demo — vendors demo the best case. Take one real recurring task, do it the old way, time it, then do it with the tool for two weeks and time it again. If the saving isn't obvious enough to notice without measuring, it isn't there. Most AI features fail this test, and the ones that pass tend to pass overwhelmingly, which makes the decision easy once you actually run it.
The question isn't whether a tool has AI in it. It's whether it does something with your data that you couldn't do yourself with a general-purpose model and twenty minutes.
Want to apply this to your business?
Book a free 30-minute strategy call and we'll diagnose your specific growth blockers together.
Book a Free Call