By Sagar Shankaran, Founder of CallSphere
Spend limits, 75% and 90% alerts and entitlements arrived in July 2026. A per-role AI budget for an imaging center, plus cost per authorized study.
Key takeaways
Most imaging center owners reading this will say the same thing: nobody here is spending real money on AI, so there is nothing to govern. Then they open the business card statement and find four separate charges. A twenty-dollar personal plan the front desk lead expensed in February. A team plan the billing manager bought for herself and the two accounts-receivable specialists. Something the referral liaison signed up for to write outreach letters to orthopedic practices. And the one you actually approved, for the prior-authorization desk.
Four subscriptions is not the problem. Four subscriptions with no cap, no visibility into who is spending what, and — this is the one that should worry you — no clarity about which of them has a signed business associate agreement while somebody is pasting in a patient's clinical history, is the problem.
On 2 July 2026 that became a solvable problem rather than a policy memo. Claude Enterprise shipped a governance update: a dashboard that shows usage and cost by person, spend limits you set at the organization level and at the individual level, automatic alerts when you hit seventy-five percent and ninety percent of a budget, default model choices, and entitlements — control over which people can use which capabilities at all. In plain terms, AI stopped being a mystery charge and became a line item you budget, cap, watch and review like your contrast media spend or your courier contract.
Worth separating two things that get lumped together. The clinical software you already buy — the triage software that flags a suspected large-vessel occlusion on a head CT, computer-aided detection on screening mammography, structured reporting inside your dictation system — is billed per study or per seat, negotiated at contract time, and is not what this post is about. That spend is predictable because it moves with volume.
The unpredictable spend is the general-purpose assistant, and it is unpredictable for one reason: usage is not tied to patient volume. Your authorization coordinator working through Thursday's advanced-imaging cases costs a steady amount per case. Your billing manager deciding on a Tuesday to feed fourteen months of remittance data into an assistant and ask it to find every denial pattern by payer can spend more in an afternoon than the coordinator spends in three weeks. Neither is wrong. But only one of them was in your plan.
This is also where the accreditation and privacy exposure sits. There is a meaningful difference between a scheduler drafting a reminder message and a lab manager pasting proficiency-testing results, or a technologist pasting a report, into a tool with no agreement behind it. Entitlements are the setting that stops the second thing without banning the first.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for healthcare in your browser — 60 seconds, no signup.
flowchart TD
A["Monthly AI budget set by owner"] --> B["Prior-auth desk allocation"]
A --> C["Billing and AR allocation"]
A --> D["Front desk pool"]
A --> E["Lab manager allocation"]
A --> F["Referral marketing allocation"]
B --> G["Alerts fire at 75% and 90%"]
C --> G
D --> G
E --> G
F --> G
G --> H["Monthly review: cost per authorized study, cost per clean claim"]
Do not set one pot and let the fastest typist drain it. Split by role, size each share by what that role is supposed to produce, and set the alert thresholds so you hear about a problem on the eighteenth of the month, not on the first of the next one.
A workable starting split for a single-site center with three modalities and a small in-house lab, on an illustrative $900 monthly budget:
| Role | Monthly cap | What it is for |
|---|---|---|
| Prior-authorization coordinator | $300 | Assembling and checking authorization packets |
| Billing / AR specialists (2) | $250 | Denial analysis, appeal letters, remittance review |
| Front desk pool (4 users) | $120 | Patient messages, prep instructions, rescheduling |
| Lab manager | $80 | Procedure manual updates, inspection prep drafts |
| Referral liaison | $50 | Outreach letters, referring-practice reports |
| Owner / administrator | $100 | Analysis, board packets, payer contract review |
Alerts at seventy-five percent land at $675 and at ninety percent at $810. The important part is not the numbers — yours will differ — it is that each line has a named owner and a stated purpose, so when a cap gets hit you have a specific conversation instead of a general worry.
Budgeting a technology by the invoice tells you nothing. Budget it by unit cost against the work it replaces, and check it monthly. Two ratios are enough for a diagnostic business.
Cost per authorized study: take the authorization desk's monthly spend and divide by the number of advanced-imaging cases authorized before the day of service. On the $300 line above, against, say, 268 authorized cases a month, that is $1.12 per case. Compare it against the coordinator time it displaced — thirteen minutes at $32 an hour fully loaded is $6.93. The spend is defensible at roughly six times over, and now you can say so in one sentence to whoever asks.
Cost per clean claim: take the billing line, divide by claims that went out without a rework touch. If that number drifts upward two months running while the clean-claim rate stays flat, the tool is being used for something other than what you funded. That is not a reason to cut it — it is a reason to ask the billing manager what she found, because sometimes the answer is that she found $40,000 of underpaid contrast-enhanced studies and the spend was the best money you have ever spent.
Here is the failure this whole approach can cause, and you should plan for it before it happens. Your busiest authorization week of the year is not spread evenly. December brings the deductible rush — patients who have finally met their deductible and want the MRI they postponed in March — and your authorization volume can run half again over a normal week for six or seven weeks. If your coordinator's monthly cap is sized for an average month, it runs out around 20 December, at exactly the moment where an unauthorized study becomes a write-off.
So do two things. Size the caps against your busiest month, not your average one. And decide in advance who can raise a cap, how fast, and by how much — a named person, a same-day decision, a stated ceiling. A cap with no fast release valve will eventually cost you more than no cap at all.
Still reading? Stop comparing — try CallSphere live.
See the healthcare AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
A spend dashboard tells you how much was spent, not whether it was worth spending. Nothing in the July release will tell you that the referral liaison's outreach letters produced eleven new orthopedic referrals and the billing manager's afternoon of denial analysis produced nothing. Only you can pair the spend with an outcome, and that is a fifteen-minute monthly review with the numbers in front of you, not a setting.
Entitlements are also not a privacy program. Controlling which people can use which capabilities is useful, but the thing that makes patient information safe in a vendor's hands is a signed business associate agreement covering that specific product, a written confirmation that your data is not used to train anything, and an audit trail you can actually pull. If the agreement is not signed, no cap in the world makes the use appropriate — and both Texas and California AI statutes took effect on 1 January 2026, so the state you operate in likely has its own opinion on disclosure and use as well.
Last, do not confuse a cap with training. Roughly seven in ten small-business owners say their people need more training on these tools. A coordinator who has never been shown how to hand over a whole day's authorization queue as a single goal will burn her budget asking one question at a time and conclude the tool is useless.
The dashboard and the per-person caps are worth it at eleven people precisely because at eleven people nobody is watching. What you should not do is buy an enterprise agreement you do not need — start by consolidating four scattered subscriptions into one billing relationship with per-user limits, which is most of the value for a fraction of the effort.
Cancel and re-issue, in that order, and do it in one week rather than one at a time. Send a short note saying which tool is now the approved one, that it is covered by an agreement that permits patient information, and that anything else is out of policy from a stated date. Expect one person to be annoyed. That is cheaper than an unlogged disclosure.
Frontier AI pricing dropped roughly tenfold from 2025, so the honest answer is: less than most owners assume, and the risk is not the rate, it is the absence of a ceiling. Set a first-quarter budget you would not mind losing entirely, watch the seventy-five percent alerts for two months, then re-size against the ratios above. Do not commit to an annual contract before you have three months of your own usage data.
The practice administrator or office manager, with a fifteen-minute standing slot in the monthly close. It sits naturally next to the courier invoice and the contrast media reconciliation. It does not need a technical person; it needs somebody who reads invoices carefully.
Worth noting where a separate line item usually pays for itself first: the phones. CallSphere builds AI voice and chat agents that answer the main line and web chat around the clock, book and reschedule scans, and capture referral details — spend that maps cleanly to a countable outcome, which makes it one of the easier lines to defend when you sit down with the dashboard at month end.

Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
Claude's 2 July 2026 governance update added spend limits and alerts at 75% and 90%. How a heavy-truck shop owner budgets AI per repair order, not month.
The EU AI Act's August 2 date, Texas TRAIGA and California SB 53 all landed. What a US rehab clinic must document, disclose and log, and what it can skip.
Freezer logs, eyewash tags, reagent expiry and MRI Zone III: what autonomous inspection rounds actually cover in a diagnostic building, and the honest math.
Seasonal spend ceilings, per-person limits, 75% alerts and cost per quote packet: budgeting AI at a contract assembly shop without stalling the quoting desk.
Work backwards from the bill rate to an AI allowance per productive hour, split caps by program, and put alerts at 75% and 90% where finance will see them.
How a US city sets per-department AI spend caps, alerts at 75% and 90%, and a single budget line — using the 2 July 2026 Claude Enterprise governance update.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI