Welcome back
It's renewal season, and some enterprise customers are pushing back on bigger commitments. That makes this a good moment to take a closer look at your own renewal.
A bigger commitment usually means a bigger discount, but that discount only pays off if your team uses most of what you bought. With cheaper models launching every few months, that's harder to predict than it used to be.
So, I want to walk you through how to choose the right size for your next AI commitment:
what you're actually paying for
when a commitment saves money and when it costs more
which contract terms to read closely
what to ask your vendor before you sign
Before we get into it, I'm hosting free AI at Work office hours this Friday. In 45 minutes, I'll work through the questions leaders are stuck on live. Save your spot here.
IN THIS EDITION
Byte-size news: API credits, ServiceNow in ChatGPT, and SharePoint agent changes
Main read: everything you need to know before your next AI renewal
■ Byte-size News
Eligible ChatGPT Pro and Business plans now get free API credits for building apps and agents. Business credits depend on your paid seats, up to $500 a month or $6,000 a year. Ask your workspace owner to claim them before you buy more.
ChatGPT · OCT 29
ServiceNow's plugin for ChatGPT lets your IT team look into a problem and update its record in one conversation, within their existing permissions. More features are coming by the end of 2026.
ChatGPT · OCT 28
From November, new SharePoint agents move to Microsoft Copilot Agent Builder. Existing .agent files stop working after March 2027, so ask IT to list yours and plan the upgrade.
SharePoint · OCT 30
Free AI at Work Office Hours with Haroon
Stuck on an AI question at work that nobody on your team can answer? Bring it to me.
This Friday, Nov 6 at 2 PM ET, I'm holding free office hours for leaders making AI work in their teams. Whether you're picking a tool, getting colleagues on board, or trying to make a workflow run reliably, I'll work through real questions live.
You'll leave with a clear next step, plus the trade-offs I'd weigh in your seat. It's 45 minutes, relaxed and free, and you don't need any technical background. Spots are limited, so grab yours early.
■ Haroon-in-the-loop
Everything you need to know before your next AI renewal
Renewal season is here, and some AI buyers are pushing back. The Information reported that some enterprise customers are resisting larger commitments.
That's a handful of buyers, so it isn't a sign that companies everywhere are walking away. But it's a good reason to take a closer look at your own renewal.
Here's the question to ask: would this deal still make sense if your team spent less on AI next year? Cheaper models keep coming out, so last year's bill may not tell you much.
Here's everything that affects the answer.
Which situation are you in?
Employee seats (Claude Team, Claude Enterprise, ChatGPT Business): your decision is how many seats you need, which tier, and whether to pay monthly or annually.
Usage inside a workflow (your product or internal tools call a model directly): you decide how much to prepay or commit to.
Reserved capacity (paying to guarantee processing power for one model): decide how much capacity you need and for how long.
Most teams pay for more than one of these, and each one deserves its own decision.
Seats
Here's what to check on each plan:
Claude Team: each seat has its own weekly limit, and usage credits can extend it. Ask who on your team actually hits their limits.
ChatGPT Business: seats can be reassigned, workspace credits cover extra usage, and API usage is billed separately. Ask whether you can drop seats mid-term.
Claude Enterprise: the seat fee covers access only, and all usage is billed on top. Ask what metered usage will add to the seat cost.
Paying annually saves $5 a seat each month. I'd only do it for people you're sure will keep using it all year.
Model usage
A model charges by the token. A token is a small chunk of text, roughly part of a word, and you pay for the tokens it reads and the tokens it writes.
Smaller models charge much less per token. At list price:
Haiku 5.5: $0.10 per million tokens read and $0.50 per million written, for prompts up to 100,000 tokens.
Haiku 4.5: $1.00 and $5.00.
Anthropic estimates Haiku 5.5 costs about 75% less per task on average, even though it uses slightly more tokens. That's the company's own estimate, so your bill may look different.
A cheaper model only counts as an option once it passes your own checks. I'd test it on a few of your real tasks first.
The commitment math
Here's an illustrative example. You pay $100,000 for the year and get $125,000 of usage at list prices. Anything you don't use expires.
The discount looks like 20%, but it only pays off once you use 80% of the pool. If your realistic usage sits near the top row, where the commitment costs you more, the commitment is too big.
What to check in your contract
These are the terms I'd read most closely:
Unused value. OpenAI's standard credit terms say credits expire after one year unless otherwise specified. Anthropic's API credits also expire after one year and are nonrefundable.
Overage. At OpenAI and Google, usage above a reservation is billed at standard pay-as-you-go rates.
New models and products. With Google provisioned throughput, you can move between Gemini models but not to Claude.
Price drops. OpenAI's Reserved Tier is counted against list prices, and any discount is applied when you buy.
Notice and renewal. Vendor pages rarely state notice periods, so check your contract.
Standard terms can differ from negotiated contracts, so read your own.
To put it simply:
Reserved capacity
If you're reserving capacity, check exactly what you're buying:
OpenAI Reserved Tier: only available to Enterprise customers. You reserve capacity for one specific model, priced in dollars per minute, and anything above your reservation is billed at normal rates.
AWS Bedrock: billed hourly, with options for no commitment, one month, or six months. Billing keeps running until you delete the reservation.
Google: terms run from one week to one year. You can't cancel mid-term, and the fees still apply if the model is discontinued.
Reserved capacity pays off when you have steady, predictable traffic. An average monthly bill can hide hours you'd still be paying for.
Platforms in the middle
If you buy AI through a platform, it adds its own costs on top:
OpenRouter: you pay the model provider's price, plus a 5.5% fee on the Standard plan or 8% on Business when you buy credits. Check which providers your data actually goes to.
Snowflake: AI Credits are separate from Platform Credits, and capacity discounts don't apply. Watch for other Snowflake charges on top of your AI Credits.
The cost of switching
Moving a workflow means moving more than the model name. You also carry over its instructions, the tests that prove it works, its tool connections, and its permissions. Before you switch, rehearse it once on one real workflow.
When committing makes sense
A commitment is a good deal when usage is proven and steady, when you really need capacity at peak times, or when the vendor's support is worth paying for. Keeping alternatives ready costs time too. For most teams, one well-supported default and one credible backup for important work is enough. Nobody needs to switch every time a new model launches.
Before you approve the renewal
Ask the vendor to price the same work two ways, paying as you go and with a smaller commitment, next to the current offer. Put your realistic low usage estimate beside all three. If the large commitment only wins when everything goes well, ask for a lower minimum or a shorter term.
Ask Haroon for Advice on AI at Work!
I’ve spent over a decade in AI, helped more than 300 teams put it to work, and taught 70,000+ people through AI For Anyone. Along the way, I’ve picked up a few lessons, and I want to use them to help you where you’re stuck.
So hit reply and ask me anything. Each week, I’ll pick one question and answer it here. I’ll keep your name and company out of it unless you tell me otherwise :)
■ Signing off
That’s it for this week. Before you approve another year of AI spending, ask what happens if your team needs less than you expect.
If you’re stuck on a renewal question, hit reply. I’d love to hear what you’re weighing up.
Until next time,
Haroon
