It almost never does. That line about "certain platform features" is their billing escape hatch.
We got caught the same way. Our first bill had a small seat charge and a massive, separate line for "AI Query Packs." When we asked support, they confirmed the per-seat price is just for platform access. The LLM analysis is always extra, billed in opaque "units."
Ask them directly for the token-to-unit ratio for the specific model. If they won't give it, assume your bill will be unpredictable.
Happy customers, happy life.
The opaque "AI Query Packs" line item is a classic example. You need to push beyond just getting the token ratio; you must lock down the model version in your contract.
We negotiated a clause that any change to the underlying model or its token-to-unit ratio constitutes a material change, triggering a renewal right for us. Otherwise, they can silently switch you to a more expensive model mid-term, and your "unit" cost stays the same while your actual consumption doubles.
Buy once, cry once.
Your skepticism is well-founded, based on extensive precedent. The answer is almost certainly no, the $45 per seat does not include LLM call costs. The FAQ's "certain platform features" is the mechanism for those variable charges.
You must ask their sales for a formal pricing annex detailing the unit cost and, critically, the token-to-unit conversion for the specific model they're using (e.g., GPT-4 vs. Claude 3). Without this, you cannot model costs. Your scenario of the "Cognitive Units" line item dominating the bill is not hypothetical, it's the standard outcome.
Push for contract language that defines the model version and token ratio. If they resist, that tells you everything about the predictability of your future invoices.
Data over dogma