Keeping AI spend predictable with per-team cost controls

The most common question we hear from finance leaders isn’t “is AI worth it?” — it’s “why can’t anyone tell me what next month’s bill will be?”

Why AI spend is volatile by default

Usage-based pricing means your bill is a function of behaviour, and behaviour changes weekly. One team discovers a new workflow, another starts pasting in entire documents, and suddenly the invoice doubles. Nobody did anything wrong — there was just nothing shaping the curve.

The usual reactions make it worse. Hard per-seat caps block people mid-task and push them back to personal accounts; a single shared API key gives you one number on the invoice and no idea who drove it.

Budgets follow teams, not individuals

Harriet’s cost controls work at the team level because that’s where behaviour actually lives:

• Model assignment by team. Give engineering frontier models and give everyone else fast, inexpensive ones. Most work doesn’t need the most expensive model — it needs the right one.
• Spend visibility per team. See exactly which teams drive cost, in real time, before the invoice lands.
• Limits that degrade gracefully. When a team hits its budget, they drop to a cheaper model rather than losing access entirely. Work continues; spend doesn’t spiral.

Run it on Harriet-managed credits or bring your own keys (BYOK) — either way you get the same per-team and per-user visibility.

Predictable beats cheap

The goal isn’t to minimise the AI bill — under-investing in AI is its own cost. The goal is a bill that finance can forecast and defend. One 500-person company we work with replaced a sprawl of per-team subscriptions with Harriet and cut over $500,000 in their first year, simply by routing spend through one plane with the right model assigned to each team.

Setting those tiers up front is part of a smooth rollout. See how plans map to devices on the pricing page, or book a call to model your own spend.

Common questions

Why are AI bills so hard to predict?

Usage-based pricing makes the bill a function of behaviour, which changes week to week. Without per-team visibility and limits, one team adopting a heavy workflow can double an invoice before anyone notices.

How do per-team AI budgets work?

Each team is assigned models and a spend limit. You see consumption per team in real time, and when a team reaches its budget it drops to a cheaper model rather than losing access — so work continues and spend stays bounded.

Is BYOK or managed credits better for cost control?

Both give the same per-team and per-user visibility in Harriet. BYOK suits teams with existing provider contracts; managed credits are simpler to start. The controls that matter — model assignment and budgets — work either way.