Medical Practice AI Budget: What to Fund First and What It Costs

A small medical practice should fund the job that loses the most measurable time or income first: AI phone answering if calls go unanswered, a scribe if clinicians finish notes after hours, billing tools if claims are rejected, recalls if the diary has gaps. Pick one job, set a stop date, and run it for four to six weeks.

The order matters more than the tool. Practices that buy four tools in a quarter cannot say which one worked, and they pay for features nobody opened. Below is a decision rule by job, cost bands for each class of tool from vendor pages, a worked budget with numbers, and the inputs you can measure to decide whether it paid back. Nothing here promises a saving; it shows how to find out whether you have one.

Follow me on Instagram@sagnikteaches

The rule: spend where the pain is measurable

Put the four candidate jobs through the same test. You need one number for each, taken from your own records over the last four weeks.

Connect on LinkedInSagnik Bhattacharya

The thresholds are rules of thumb, not standards. If two jobs tie, choose the one whose result is easiest to see in the first month. Phones are the easiest, because a missed-call count falls or it does not.

Subscribe on YouTube@sagnikteaches

Cost bands by tool class

These are list prices read on the vendors' own pages in early November 2026. They move, so check the vendor page the day you decide. A band is not a quote for your practice.

Two points shape the budget. First, scribes charge per clinician and phone tools charge per call, minute or customer, so the same practice can see very different bills as volume changes. Second, the plan price is not the whole cost: add set-up time, staff review time and any integration add-on. Where a vendor says the integration with your records system is extra, ask the price now.

A worked budget for a four-clinician practice

Illustration. A primary-care practice has four clinicians and two front-desk staff. The manager measures four weeks and finds:

• Calls: 1,000 inbound in the month. The phone log shows 140 abandoned or unanswered, 90 of them in the 8:00 to 9:30 peak and after 17:30.
• Notes: clinicians report an average of 35 minutes of note-writing after clinic each day, over about 20 clinic days.
• Claims: 6% of claims are rejected at first submission; fixing them takes the billing clerk about 5 hours a week.
• Recalls: 210 patients are overdue for a chronic-condition review.

Now the sums, all illustrative. Overflow phone cover: 140 missed calls at about two minutes each is 280 minutes, so a plan with 500 minutes (the $199 band above, with extra minutes at $0.40 on that plan) leaves headroom. Say $199 a month, plus an hour of set-up and fifteen minutes a week of transcript review, so call it $230 a month all in at a $20 hour for staff time.

Scribe: 35 minutes a day times 20 days is 11.7 hours a month per clinician, so about 47 hours across four. If the scribe halved that (an assumption you must test, not a promise), the saving is 23 hours. At $59 a seat for Freed Core (read the checkout, since the page shows $79 on one toggle), four seats is $236 a month; at Premier annual, $104 times four is $416. Whether 23 hours is worth $236 depends on what those hours are worth to the clinicians, and whether they are given back to patient time or to going home on time.

Billing: 5 hours a week is about 22 hours a month for the clerk. The tool cost is unknown until quoted. Recalls: 210 overdue patients; if a text and a booking link brought back 15% that is 31 appointments. What a review appointment is worth to you decides this one.

The decision for this practice: phones first, for three reasons. The pain is the clearest (140 calls), it is the cheapest to trial ($199 a month with an exit), and the result shows in the first week. Scribe second, as a one-clinician pilot. Billing and recalls wait for quotes.

Payback inputs you can measure

A payback number is only as good as its inputs. For each job, record the same small set before and during the trial. Write them on one sheet.

Phones

• Calls received, answered, abandoned, per week.
• Calls handled by the AI, and the share transferred to a human.
• Appointments booked by the AI, checked against the diary.
• Complaints or corrections from patients, per week.

Scribe

• Minutes of notes per consultation, before and with the tool, sampled over five days.
• Time spent editing each AI note.
• Notes where an error needed correcting (count them).
• Finish time of the last clinician each day.

Billing

• First-submission rejection rate.
• Hours spent on rework.
• Average days from visit to payment.

Recalls

• Messages sent, replies, bookings made and attended.
• Opt-outs and complaints.

Keep the sheet to one page. If you cannot get an input in under ten minutes, pick another measure.

A payback sum with the inputs filled in

Using the practice above after a four-week phone trial (illustrative figures). The AI handled 150 calls; 110 were answered fully, 40 were passed on. 28 appointments were booked that the diary confirms. The manager does not count all 28 as new income, because some patients would have rung back. She decides to count half, 14, at an average appointment value of $70, which is $980. Cost: $199 plan plus $31 of staff time, $230. Net before any other effect: $750 for the month.

That is not a forecast. If the numbers next month drop, she stops. Notice that she discounted her own result, wrote the assumption on the sheet, and left room for it to fail. Compare this with a method in the 90-day receptionist return tutorial, which uses a longer window.

What to leave out of the first budget

• Anything that needs a custom integration before you have proved the plain version works.
• Annual contracts before a trial has finished. Monthly is dearer but lets you leave.
• Bundles that add tools you did not pick. Two lines on an invoice you never use are costs.
• Clinical decision features beyond note-taking, until a clinician has read the vendor's evidence and your insurer's view.

Checks that belong in every first budget

Add a line for these even if they cost nothing in cash.

1. Data terms. Read the vendor's privacy page, and ask whether a business associate agreement or equivalent data-processing contract is offered on your plan. See the BAA check by plan.
2. Consent and notice. Callers need to know a call may be recorded or answered by AI; patients should know a scribe is in the room. Write the wording before the trial.
3. Human back-up. For every tool, name who takes over when it fails.
4. Exit. Can you export your data and cancel in a month?
5. Owner. One named person runs each trial.

The vendor questions in what to ask before buying any AI tool for a medical practice cover these in more depth.

Three smaller examples of the order in action

A solo clinician with a part-time receptionist

Calls are manageable; the clinician writes notes every evening. First spend: a scribe, trial on the Free or entry plan for a month, measuring finish time. Phones wait.

A busy practice with a front desk overwhelmed at 8:30

Notes are fine, but the phone queue at opening is long and 15% of callers hang up. First spend: AI phone answering as overflow only, so the AI picks up when the queue passes a threshold, with transfers for anything urgent. The tutorial on adding AI answering to existing lines shows how.

A specialist clinic with long gaps between reviews

Phones and notes are in order; patients overdue for a review never hear from the practice. First spend: recalls, with a monthly list from the diary and one reminder message. See how dental practices run recalls for the mechanics, which carry over.

A one-page decision record

Keep a record of every spending decision. It stops the same debate returning and gives the next person your reasoning.

AI SPEND DECISION RECORD
Job chosen: AI phone answering (overflow and after-hours)
Why: 140 of 1,000 calls unanswered in 4 weeks; the largest measurable gap
Tool and plan: [plan], $199 a month, monthly term
Set-up time: 1 hour; weekly review 15 minutes by practice manager
Measure: unanswered calls per week; bookings confirmed in diary; patient complaints
Start: [date]. Stop-or-decide date: [date + 6 weeks]
Stop if: unanswered calls do not fall, or more than 2 complaints a week
Data terms checked: yes / no (attach vendor page printout)
Fallback if it fails: calls revert to the front desk and voicemail
Next job in line: scribe pilot with one clinician

Using an assistant to tidy your tally

The measuring step is where most practices stall, because the data sits in an exported spreadsheet nobody has time to read. An AI assistant can summarise a de-identified export. Remove patient names, numbers and dates of birth first, keep only the columns you need, and check the assistant's data terms. Here is a prompt for a phone log.

Below is a de-identified call log from a small medical practice for four weeks. Columns: date, time, answered (yes/no), duration in seconds, reason category.
Give me: total calls; calls answered and unanswered by week; the three hours of the day with the most unanswered calls; the five most common reasons; and the average duration of answered calls. Show your arithmetic for the totals. If anything is unclear in the data, ask me instead of guessing.

[PASTE ROWS]

Illustrative output: "Total 1,000. Unanswered 140 (14%). Worst hours: 08:00 to 08:59 (41 unanswered), 17:30 to 18:29 (33), 12:30 to 13:29 (22). Top reasons: booking or changing an appointment (310), prescriptions (180), results (120), opening hours and directions (95), referrals (60). Average answered call 2 minutes 10 seconds." What you would fix: re-add the totals yourself with a spreadsheet formula, because language models can slip on sums. The pattern is what you want, not the decimals. Note that 95 calls were fixed-answer questions, which an FAQ bot or a phone greeting may handle more cheaply than an AI voice agent; the clinical search or patient FAQ bot tutorial covers that choice.

Designing the trial so it can fail cleanly

A four-week phone trial

Week one: set up and test with staff calling in, including a caller who asks for medical advice, one who is in distress, and one who speaks quickly. Week two: switch on for after-hours only. Week three: add peak-hour overflow. Week four: review transcripts daily and compare with the baseline sheet. At the end, one meeting, one decision: keep, change or stop.

A scribe pilot with one clinician

Choose the clinician who is most keen and the one who is most doubtful, in turn rather than together. Ask them to record the time it takes to finish notes on five typical days before, then five days with the tool. Ask them to count how many notes needed a correction and what kind. The doubter's view is the more valuable, because it shows what the sceptics on your team will worry about. Patients must know a scribe is in use; agree the wording at the front desk first.

What counts as failure

Write the failure line before you start. Examples: more than two patient complaints a week about the phone assistant; a clinician reporting that editing takes as long as writing; transcripts showing the tool giving advice it should not. A written line means you stop on the day, not after a debate.

Telling the team

Staff hear "AI" and think "job cuts". Say what is true in plain words. A short note you can adapt:

Team,

From [date] we are trialling [tool] for [one job] for six weeks. It is a trial: we are measuring [one number] and we will stop it if it does not help. It does not replace anyone. [Name] owns it and will review it with you every Friday. If you see it do something wrong, tell [name] the same day.

[your name]

Then keep the promise. If the trial succeeds, say what the time saved will be used for before you are asked.

When the budget is tiny

If you have almost nothing to spend, there are still sensible first steps. A free-plan scribe trial costs an hour of set-up. A written fact sheet for your front desk costs nothing and also feeds any later chatbot. A phone greeting that answers the five most common questions in 20 seconds may remove a share of calls before any AI is involved. Spend time first and money second, and let the sheet tell you when money is justified.

Two more sums: billing and recalls

Both were left unpriced above because quotes are needed, but you can still set the bar a tool must clear. Illustration for the same practice.

Billing. The clerk spends 22 hours a month fixing rejected claims, which at $20 an hour is $440 of staff time. Suppose a claim-checking tool cut rejections by a third; that would free about 7 hours, or $147. A tool quoted at $300 a month would not pay for itself on staff time alone. It might still pay if faster payment improves cash flow, but that is a separate input: average days from visit to payment, before and after. So the rule is simple: ask for a quote, then see whether the quote is below the value of the hours it could plausibly free, and treat any cash-flow gain as a bonus you measure rather than assume.

Recalls. Of 210 overdue patients, a practice might send a single message and a booking link. If 10% book and attend, that is 21 appointments. At $70 each that is $1,470, once, not every month, because the overdue list is a stock that shrinks. Spread it over the three months you would take to work through the list, and a tool at $150 a month costs $450. The sum looks good, but it depends on the 10% reply rate, which only your own patients can tell you. So send the first batch of 30 messages, count the bookings, and then decide whether to buy anything at all. A spreadsheet and your existing messaging may do the first round.

Mistakes first budgets run into

Paying per clinician for a clinician who will not use it. A practice buys four scribe seats and two clinicians never log in. Start with one or two seats and add as people ask.

Choosing by headline price. A cheap per-minute phone plan runs out in week two because calls are longer than expected. Check minutes or calls against your own log, and ask about overage before signing; Rosie's page, for example, does not publish its overage rate.

No stop date. A trial with no end becomes a subscription nobody reviews. Put the date in the diary on day one.

Counting every booking as new income. Some callers would have rung back. Discount, as the manager did above.

Where the spend goes after the first job

If the first trial works, you will have a method, a sheet and a clear exit. Repeat it with the next job in line. If it does not, you have learned something cheaply. Either way, do not add a second tool while the first is still being measured: two changes at once leave you unable to say which one moved the number.

Budget questions from practice managers

How much should a small practice budget for its first AI tool?

Start from the job, not a figure. Entry plans for AI phone answering and for scribes list from tens of dollars to a couple of hundred dollars a month per tool or clinician. Budget the plan, an hour or two of set-up, and a short monthly review, and set a stop date before you start.

Should we trial two tools at once?

No. Two trials running together make it impossible to know which one changed your numbers, and they split staff attention. Run one for four to six weeks, record the result, then decide on the next.

What if a vendor will not publish a price?

Treat a quote-only price as a cost band you must get in writing before the trial starts, including set-up, per-user fees and the minimum term. If they will not put it in writing, that is a reason to rank them lower.

Can I promise staff the tool will save them time?

Do not promise it. Say you are running a trial, name the one measure you will watch, and tell them what happens if it does not help. Staff trust a trial with an exit more than a promise.

Further reads

• What Does an AI Receptionist Cost a Dental Practice in 2026? — A phone-side cost breakdown from a close sector.
• Winning Back Lapsed Dental Patients With AI Messages — Message wording for patients who have lapsed.
• Best AI Tool for Medical Billing: Scrubbing, Coding, Denials — Billing tools split by job: scrubbing, coding, denials.
• Is an AI Scribe Worth It for a Small Private Clinic? — Whether a scribe suits a small private clinic.
• Do Small Medical Practices Need an AI Consultant? — Whether a small practice needs outside help at all.
• AI Solutions for Medical Practices: Build, Buy or Hire Help? — Whether to build, buy or bring in help.
• A Worked AI Implementation Plan for a Small Medical Practice — A sequence for adding AI to a practice over a few months.
• AI Chatbot for a Doctor's Website: Scope, Disclosure, Handover — A practice owner's set-up guide: scope sheet, disclosure line, hand-over rules, logging and named chatbot tools with costs.
• AI Claim Scrubber: What It Catches and What It Misses — Modifier, bundling and diagnosis edits are caught; eligibility, documentation and fresh payer rules often are not. Test one on 50 claims.
• AI Receptionist for a GP Surgery: The 8am Rush and Hand-Offs — A practical overflow test and clinical hand-off plan for a surgery considering an AI phone receptionist.
• AI Scribe for Medical Practices: Pick by Country and EHR — A practical decision matrix for choosing a clinical scribe by jurisdiction, EHR connection and specialty fit, with a scored pilot.
• Best AI for Healthcare Providers: What to Buy First by Pain Point — A buying order for a small practice: price your missed calls, late notes and denied claims, then pick the phone agent, scribe or claim check.
• Cliniko for Doctors: Fit, Cost and the AI Tools That Sit Around It — Cliniko is built for allied health. What that means for a doctor's practice, what it costs, and the AI scribes and phone tools that connect to it.
• Healthcare AI Call Centre: When a Group Practice Should Centralise Calls — Five tests for a two-to-six site medical or dental group deciding between one central AI-assisted queue and a receptionist at each site.
• Medical Equipment Lease: Compare Quotes and Check Terms With AI — A checklist for comparing lease quotes on total cost, end-of-lease options and exit terms, with AI to flag clauses before a solicitor reviews them.
• Midwife Software Compared: Notes, Booking, Billing and AI Tools — Five midwife software options checked on vendor pages for notes, consent, portal, billing, data location and AI claims, with a scoring sheet.
• Therapy Practice Management Software: What the AI Add-Ons Cost — SimplePractice, Jane and Carepatron base plans against their AI add-ons, with a monthly sum for a three-therapist practice and the traps in each quote.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Smith.ai AI Receptionist, Dialzara, Rosie and Goodcall pricing pages; Freed and Heidi pricing pages; Tali pricing page (shown in another currency); all checked 4 November 2026.