Not long ago, AI was a separate purchase. Now it arrives inside the software you already own: a new add-on, a higher edition, a pack of credits, or a price increase explained by "new AI capabilities". If you don't decide how you buy it, your renewal decides for you.
Almost every major software vendor is adding AI features to its products, and most are looking for a way to charge for them. That is fair: the features cost money to build and run. The problem is that AI pricing is new, it changes quickly, and the contract terms around it are often weaker than the terms you negotiated for the core product.
This is how we help clients buy AI inside their software agreements without paying for hype, losing control of their data, or signing up for a bill nobody can predict.
Find out how the AI is priced first
Before you can tell whether an AI offer is good value, you need to know which model the vendor is using. Most fall into one of four:
| How it is sold | What it looks like | What to watch |
|---|---|---|
| Per-seat add-on | A monthly fee per user on top of the licence you already pay | Paying for every user when only some will use it |
| Credits or usage | A pack of credits, tokens or "actions", with a charge for more | Bills that grow with use, and credits that expire unused |
| Bundled edition | AI included only in a higher, more expensive edition | Upgrading everyone to get a feature a few people need |
| Price increase | A higher renewal price justified by the new features | Paying for AI you never asked for and may not use |
Each one needs a different negotiation. A per-seat add-on is about how many seats. Usage pricing is about caps, rates and who watches the meter. A bundled edition is about whether you need the upgrade at all. A price increase is about separating the AI from the product you already pay for.
Start small, on your timeline
Vendors often push for a company-wide commitment from day one. It is rarely the best way to buy something your teams have not tried yet.
- A defined pilot. A small group of named users, a fixed end date and an agreed way to measure whether it helped.
- Pilot pricing. A pilot priced in proportion to its size, not a full-year commitment for the whole company.
- A price for later, agreed now. The price per seat or per unit if you expand, fixed before the pilot starts, so success does not cost you your leverage.
- A way out. If the pilot doesn't deliver, the add-on ends with it and does not renew on its own.
Put limits on usage pricing
Usage pricing can be fair, because you pay for what you use. It can also produce the one invoice nobody planned for. If the AI is priced by credits, tokens, queries or actions, ask for:
- A monthly or yearly cap, with the service pausing or asking before it goes over, not billing automatically.
- Alerts at agreed levels, for example at 50, 75 and 90 percent of the committed amount, sent to a named owner.
- The rate for extra usage written into the contract, not "at the then-current price".
- Unused credits that roll over, or at least a clear statement of when they expire.
- Usage reports you can read, by team or user, so you can see what is driving the cost.
Usage pricing is only fair when someone on your side owns the meter.
Keep AI from raising the price of everything else
Watch for renewal quotes where the core product's price goes up and the reason given is the AI that came with it. Ask the vendor to show the AI as its own line, with its own price. Then you can decide on it separately, and your existing discount on the core product is not lost in the change.
If AI features are switched on by default, find out whether you can switch them off, and whether doing so changes the price. Paying for capabilities your organization has decided not to use, for security or policy reasons, is a cost worth pushing back on.
Get the data terms in writing
AI features often send your content (documents, emails, tickets, customer records) to models that may be run by the vendor or by a third party. The standard terms for that can be vague. Before you agree, make sure the contract answers:
- Training. Is your data, or what your users type, used to train or improve models? The answer should be no, unless you agree in writing.
- Storage and deletion. Where are prompts and outputs stored, for how long, and how are they deleted?
- Subprocessors. Which third parties, including model providers, see your data, and will you be told before that list changes?
- Residency. Does processing stay in the regions your obligations require?
- Confidentiality. Do the same confidentiality and security commitments cover the AI features as the rest of the service?
Plan for change
AI products change faster than almost anything else you buy. Models are replaced, features move between editions, and online terms are updated. Your contract should hold steady while the product moves:
- Terms that don't change on their own. Changes to online terms or AI policies should not reduce your rights during the term without your agreement.
- Notice of model changes. Advance notice when a model or feature you rely on is changed or retired, and the right to leave the add-on if the replacement doesn't work for you.
- Responsibility for outputs. Clear terms on who stands behind the AI's output, including protection if a third party claims it infringes their rights.
- Exit. The right to export anything you created with the AI features, in a usable format, if you leave.
Questions to ask before you sign
- Who needs it? Which teams will use it, and what will they use it for?
- How is it priced? Per seat, per use, bundled or folded into the renewal price, and what does each option cost at your real volumes?
- How will we know it works? What does the pilot measure, and who decides whether to expand?
- What happens to our data? Training, storage, subprocessors and residency, all in writing.
- What if it changes? Notice, the right to leave the add-on, and price protection if you expand later.
Buy AI for the work your teams will actually do, not for the roadmap the vendor is selling.
How MerciIT can help
We negotiate technology contracts for a living, and we never negotiate alone: our sourcing, cyber security, legal and technology specialists work every deal together. That matters with AI, where the price, the data terms and the contract language have to be read side by side.