A short note before you start
This is not a sales document. It is the document we wished existed when we started talking to independent veterinary practices.
You will not find "transform your practice" in here. You will not find "AI-powered" used as a brag. You will not find a chart that pretends AI fixes the labor shortage, because nothing fixes the labor shortage and pretending otherwise is how vendors lose your trust in 90 seconds.
What you will find:
1. A map of where the day actually breaks (the phones, the refill backlog, the unstaffed hour after the last appointment).
2. The five places AI can carry real load right now, in plain language, with the trade-offs.
3. Prompts and workflows you can copy.
4. A 90-day rollout you can run with the team you already have.
5. The questions to ask any vendor (including us) before you sign anything.
Read it. Steal from it. If at the end you decide your practice does not need outside help, that is a fine outcome. If you decide the leak is bigger than what you can build in-house, you know where to find us.
Yours either way.
Table of contents
Part 1: The reality
1. Why your day feels broken, and why that is not your fault
2. The leak map: where the work actually goes
3. What AI can and cannot do for an independent practice
4. The honest limits
Part 2: The five systems
5. System 1: The prescription-refill agent
6. System 2: The AI visit scribe
7. System 3: The after-hours capture and triage router
8. System 4: The unified client-comms layer over your PIMS
9. System 5: The onboarding and SOP agent that survives turnover
Part 3: The build
10. Working with your PIMS (ezyVet, Cornerstone, Avimark, Shepherd, DaySmart)
11. Prompts and workflow examples you can copy
12. The 90-day rollout
Part 4: The decisions
13. Build, buy, or borrow: how to choose
14. Vendor questions that separate real from theater
15. Governance, safety, and the anxious-pet rule
16. What you are signing up for if you do nothing
The reality
Chapter 1. Why your day feels broken, and why that is not your fault
Independent veterinary practices in the United States are not short on demand. They are short on hours, hands, and quiet. The market is roughly $72.6B with about 57,920 vet businesses, and around 39% of practicing veterinarians are 55 or older. The pipeline of new DVMs is not keeping up with the patient load. Burnout already costs the profession an estimated $2B per year in lost productivity, turnover, and disability claims.
You already know all of that. You feel it in three places every day:
1. The front desk is the bottleneck of the practice, and it never gets a quiet moment.
2. Refills, records requests, and after-hours calls stack on top of the day you planned.
3. New hires take 90 days to be useful, and a lot of them do not last 90 days.
This is the operator's version of the problem, not the headline version. The headline version says "the industry is in crisis." The operator's version says "Tuesday is already underwater by 9:15 and my CSR is crying in the break room." Both are true. Only one of them tells you what to do next.
Here is the part most vendors skip: your team is not slow. Your software is not the villain. Your owners are not greedy. The day is broken because the work has grown around the systems you bought when the practice was smaller, and nobody has paid down that debt for years. Phones ring during exams. Refills sit in a queue that lives partly in the PIMS, partly on sticky notes, partly in the CSR's head. The first 45 minutes of the day are pure triage. The last hour is documentation that should have happened during the day but did not, because the day did not allow it.
AI does not fix the day by replacing people. It fixes the day by absorbing the load that should never have landed on people in the first place.
That is the whole thesis of this guide.
Chapter 2. The leak map: where the work actually goes
Before you talk about tools, you have to know where the hours go. Below is the load map we use in our audits. It is not a survey. It is what we see when we sit in a practice for two days and count.
The front desk
The phone is the epicenter. A small independent practice gets 60 to 200 inbound calls a day depending on size and reputation. Most calls are not new clients. They are existing clients asking the same five questions:
1. Can I get a refill on Apoquel?
2. Is this an emergency or can it wait until Monday?
3. Can I move my appointment?
4. Do you have my records to send to the specialist?
5. What was the dose again?
Each of those calls pulls a CSR out of whatever they were doing. Each one requires a PIMS lookup. Each one ends with a note that may or may not get written down. The cost is not the call. The cost is the context switch and the unwritten note.
The phone is not a marketing problem. It is an operations problem.
The exam room
The DVM enters the room with a problem-list in their head and exits with a problem-list, a SOAP, an estimate, and a client conversation to finish. Charting either happens in the room (slow, breaks rapport) or after hours (burnout). There is no third option in most practices. About 40 to 60 minutes of unbilled charting work piles up by end of day for a full-schedule DVM.
This is where the AI scribe category lives. It is also where the most vendor noise lives, because scribes are easy to demo. We will get to where the noise stops being useful.
The treatment area and the back
Refills, lab follow-ups, callbacks, surgery prep, controlled-drug logs. None of it is glamorous. All of it is required. Most of it is squeezed into the same hour as lunch.
The after-hours window
7:00pm to 7:00am is where 30% to 40% of inbound voicemails land. Studies of after-hours missed calls in healthcare (general, not vet-specific) put the missed-call rate around 34%. We see vet-specific rates higher than that in practices without an after-hours service. Most of those calls are not emergencies. Some are. The cost of treating all of them like voicemail is that the true emergency does not get triaged in time. The cost of treating all of them like emergencies is that the on-call vet never sleeps.
This is the single highest-leverage place AI shows up in an independent practice, and we will spend a full chapter on it.
The new-hire drag
A new CSR gets four days of training and a 90-day probation. By week two they are answering the phone alone. The institutional knowledge of the practice, where the controlled-drug log lives, how to handle the client who only pays cash, what the vet means when she says "back-of-house" on a Wednesday, is in people's heads. When someone quits, that knowledge walks out with them. The next hire reinvents it badly. This is the loop that makes turnover so expensive: it is not the salary, it is the relearning tax.
Where the leak adds up
If you total the unbilled hours across one small practice, the typical number we see is 35 to 60 staff-hours per week that go into work nobody designed for: rework, repeat questions, post-hoc charting, refill chasing, and triaging voicemails the next morning. At $25 to $40 fully loaded per hour, that is between $40,000 and $120,000 a year of labor that never shows up in a price increase. It shows up in turnover and in your owner-vet's evenings.
That is the leak. Now we can talk about what to do with it.
Chapter 3. What AI can and cannot do for an independent practice
We will be specific so nobody has to guess.
What AI can do today, well enough to deploy
1. Listen to a phone call, identify the caller, identify the reason, and write a structured note back into your PIMS.
2. Read an inbound voicemail or text, decide if it is an emergency, route it to the right human, and confirm receipt.
3. Take a refill request, check the patient record, confirm last visit date, check the controlled-drug rules, and either approve, deny, or escalate to a vet, with the reasoning written down.
4. Listen to a 12-minute exam and produce a draft SOAP note that the DVM corrects, not authors.
5. Answer a routine client question (next dose, vaccine schedule, post-op instructions) using the practice's own SOPs as the source of truth, not the open internet.
6. Hold the institutional knowledge of the practice in a place that new hires can search.
That is the floor. Everything in this guide builds on it.
What AI cannot do today, regardless of what the vendor says
1. Diagnose. Not yours, not anyone's. A model can summarize a chart. It cannot decide if the dog has IMHA.
2. Replace a vet tech in the back. The hands-on work is hands-on for a reason.
3. Fix the labor shortage. There are not enough credentialed vet techs in the country. AI does not change that. It changes how much you need them to do work that should not have been theirs in the first place.
4. Run a controlled-drug protocol unsupervised. Anything touching scheduled drugs has a human in the loop, every time, no exceptions.
5. Talk a grieving owner through euthanasia. Do not let any vendor sell you on this. The phones we want AI to handle are the routine ones. The hard ones still belong to your people, with more space to do them well.
The category you should be most skeptical of
"AI receptionist." If a vendor pitches you a fully autonomous voice agent that handles every call, treat that the way you would treat a vendor who said they could do dentals without an anesthesia plan. The work is more complicated than the demo. The right model is an AI layer that absorbs the routine 70%, hands off the rest cleanly, and writes everything back into your PIMS so the human picking up the call has full context.
Chapter 4. The honest limits
Here are the things we tell every prospective client before we take their money. They apply to anyone you talk to, not just us.
1. AI does not make a broken practice good. It makes a working practice less heavy. If the underlying workflows are chaos, AI will automate the chaos.
2. AI will surface problems you have been ignoring. The refill backlog you "kind of" know about will become a number. That number will be larger than you expected.
3. The team needs to be part of the build. Not informed of it. Part of it. If you roll AI in over the top of your CSRs without their input, they will route around it within two weeks.
4. The first 30 days are slower, not faster. You are teaching the system your practice. Anybody who tells you otherwise has not done it before.
5. The wrong vendor will sell you a generic tool and walk away. The right vendor maps your workflow first and builds to it. If you are evaluating us, this is the test we want you to apply.
If those five truths land wrong, the rest of this guide will not save you. If they land right, keep reading.
Part 2: The five systems
These are the five places we see AI carry real load inside an independent practice. They are listed in the order we usually recommend deploying them, which is not the order most vendors push. The order matters because each system makes the next one easier.
## Chapter 5. System 1: The prescription-refill agent
### Why this one comes first
It is the highest-volume, lowest-risk, highest-relief work in the practice. A refill request shows up by phone, by text, by client portal, or by pharmacy fax. Today it gets handled by a human who has to:
1. Pull the patient record.
2. Confirm the patient is current on exams and bloodwork where required.
3. Check the drug for refill rules and controlled status.
4. Approve, deny, or send to the vet.
5. Communicate the result back to the client and the pharmacy.
A small practice runs 30 to 80 of these a week. A medium one runs 150+. None of it is exciting. All of it is interruption.
### What the agent actually does
The agent receives the request through whatever channel the client used. It identifies the patient, the requesting party, and the drug. It pulls the patient record from the PIMS (ezyVet, Cornerstone, Avimark, Shepherd, DaySmart, or another) and runs through your practice's refill policy as a structured checklist:
1. Is the patient current on the required exam? (Pulled from PIMS.)
2. Is bloodwork current for this drug class? (Configurable per practice.)
3. Is the drug refillable, or scheduled, or restricted?
4. Has the dose changed in the last 90 days?
5. Is there an open lab result the vet has not signed off on?
If the checks pass, the agent prepares a draft approval for a vet to review with one click. If any check fails, it sends a short, structured message back to the client explaining what is needed (most often: "you are due for a recheck before we can refill this") and books or offers to book that recheck.
Nothing leaves the practice without a vet's name on it. The agent does not write prescriptions. It prepares them, fully reasoned, for a human to release.
### Where the time goes
The realistic relief is 70% to 85% of the front desk's refill load, depending on how clean your PIMS data is. The vet still does the final approve. What goes away is the chasing, the repeated client calls, the pharmacy faxes that sit in a tray.
### What can go wrong
1. Dirty PIMS data. If the last exam date is wrong in your PIMS, the agent will be wrong. Fixing this is a one-time data cleanup, not an ongoing burden, but you cannot skip it.
2. Edge cases your policy never wrote down. The build process forces you to write your refill policy down for the first time. Most practices have not. Doing it is uncomfortable and then permanent.
3. Compassionate exceptions. Long-term clients, hospice cases, end-of-life pain management. The agent has to know when to step aside. That is configurable, but you have to configure it, not assume it.
## Chapter 6. System 2: The AI visit scribe
### What this is
A scribe listens to the exam (with consent), produces a draft SOAP, and writes it into the PIMS for the DVM to review and sign. The category is well established. HappyDoc and Digitail are two of the bigger names. There are others.
### Where the vendor noise gets loud
Most scribes demo beautifully on a clean exam with one patient, one DVM, and no interruptions. Real exams have a tech walking in, a dog barking, a client interrupting, and a phone ringing in the lobby. Demos are not your reality.
### What to actually evaluate
1. Performance on multi-speaker audio with background noise. Ask for a live demo with a noisy file.
2. Native PIMS write-back. Not a CSV export. A real integration. If they cannot write into your PIMS, the DVM ends up retyping, and you have built a parallel system.
3. Editability of the draft. The DVM should be able to correct in one place, not flip between three screens.
4. Latency. If the draft is not in the chart by the end of the appointment, the DVM will close the exam without it and the moment is gone.
5. Cost per exam at your volume. The category prices per seat, per exam, and per minute. Run your math.
### Build versus buy
For scribes, buy. The category is mature enough that custom is rarely worth it. The thing to custom-build around it is the layer that takes the scribe output, your refill agent's output, and your phone agent's output, and ties them into one clean record. That is where a generalist vendor like Enertia adds value next to a specialist scribe.
### Honest limit
A scribe makes a fast DVM faster. It does not make a slow DVM fast. The gain is real but bounded. The bigger win is that the DVM gets their evenings back.
## Chapter 7. System 3: The after-hours capture and triage router
### What this is
Every inbound call, text, and message between your last appointment and the start of the next day gets received by an AI agent that:
1. Identifies the caller and the patient if it can.
2. Asks two or three structured questions to decide if the situation is an emergency.
3. Routes accordingly: true emergency to your on-call vet or the ER you partner with, urgent-but-not-emergent to the morning triage queue, routine to the standard morning callback list.
4. Confirms receipt to the client with a clear next step.
5. Writes the full conversation and decision into the PIMS so the morning team has context.
### Why this one matters more than the marketing wants you to believe
The unstaffed hour is where churn lives. A client who cannot reach you at 8:15pm with a panicked question about their post-op cat is not going to leave you over that one night. They are going to leave you the third time it happens, when an ER is closer and answers. The after-hours agent is not a sales tool. It is a retention tool. It also keeps your on-call vet from being woken up at 2:14am for a question about kibble.
### The anxious-pet rule (non-negotiable)
Every triage agent we build has a hard rule: when the model is uncertain, when the caller is distressed, when keywords cross into emergency territory (bloat, blocked, seizure, hit by car, ate chocolate, ate xylitol, not breathing, blue gums), the call goes to a human, immediately, without waiting for the model to "decide." Better to wake your on-call vet for a non-emergency than to miss the one that mattered.
If a vendor's demo cannot show you their escalation rules in writing, do not buy.
### What to expect operationally
After-hours voicemail volume drops. The morning triage queue gets cleaner and longer (which is good, because now you can actually see what came in overnight). Your CSRs start the day from a calmer place. The on-call rotation gets less brutal.
This is the system that, more than any other, returns the day to your people.
## Chapter 8. System 4: The unified client-comms layer over your PIMS
### What this is
Texts, voicemails, emails, portal messages, and inbound social DMs all funnel into one inbox that sits on top of your PIMS. The agent in that inbox can:
1. Identify the client and patient.
2. Suggest a draft response from the practice's SOPs.
3. Handle confirmations and rescheduling autonomously, within rules you set.
4. Route anything beyond the rules to the right human.
5. Log every conversation back into the patient record.
### Why this is one system, not five
Every PIMS has a "communications" feature. None of them are good enough to be the only one your team uses, which is why your team is also using text, the phone, the portal, and the front-desk email. The result is that no message has a canonical home. A confirmation lives in text. A refill lives in voicemail. A records request lives in email. Nobody can find anything.
The unified layer makes the PIMS the source of truth without forcing the team to live inside the PIMS UI all day.
### What this is not
A new chat tool for clients to learn. The client never sees a new app. They text the number they always texted, leave a voicemail at the number they always called, or email the address they always used. The change is internal.
### Where this fights with vendors
PIMS vendors have started shipping "their own" comms layer. Some are okay. Most are a checkbox feature, not a product. If you already pay for one and it works, use it. If you are paying for one and your team still keeps a parallel spreadsheet, that is the signal that you need something else.
## Chapter 9. System 5: The onboarding and SOP agent that survives turnover
### What this is
A searchable, conversational layer on top of every SOP, policy, and "the way we do it here" that lives in your practice. New hires ask it questions. Long-tenured staff use it to settle disputes. The owner-vet uses it to enforce consistency.
### Why this comes last in the rollout, not first
It is the most strategic of the five, but it depends on the other four. The reason: most practices do not have their SOPs written down in any usable form. They have them in three Google Docs, one Word file from 2017, a binder behind the front desk, and the head of one CSR who has been there 11 years. You cannot build an SOP agent on top of nothing.
The act of building this agent forces the practice to write its operating manual for the first time. That is a six- to ten-week project on its own, and it is worth doing whether or not you build the agent on top of it.
### The turnover math
A new CSR costs roughly $3,500 to $6,000 to fully onboard (recruiting, training time, ramp loss, mistakes). The average tenure in an independent practice is shorter than the industry would like to admit. If you halve your ramp time and lose one fewer hire per year, the agent pays for itself before the second cycle.
### What this agent does not do
Train. Real training is human. The agent is a reference, not a teacher. It does not replace the 90-day probation, the shadowing, or the slow trust that gets built between a tech and a vet. It just removes the "where do I find the controlled-drug log again?" question from the queue so the human teaching can focus on the hard parts.
---
# Part 3: The build
## Chapter 10. Working with your PIMS
Your PIMS is the source of truth. Nothing AI does is allowed to bypass it. The patient record stays in the PIMS. The financial record stays in the PIMS. Every agent writes back.
A short, opinionated note on the major systems we see:
### ezyVet
Cloud-native, strong API, the easiest of the major PIMS to integrate against. Pricing runs around $245 to $260 per user per month at small-practice scale. If you are on ezyVet, the build is faster and cheaper across all five systems.
### Cornerstone (IDEXX)
Legacy, quote-only pricing, large installed base. Integration is workable but the data model is older and the write paths are more limited. Expect a longer build. Worth doing anyway.
### Avimark
Legacy, common in older practices. Integration is the slowest of the group. The systems still work; the connector is more custom.
### Shepherd
Newer, cloud, flat $299 per month. Good API access. Fast to integrate.
### DaySmart Vet
Around $116 per month, lighter weight, smaller practices. Easy to integrate at the cost of fewer fields available.
The rule that applies to all of them: do not replace your PIMS to make an AI system work. Build the AI system to fit your PIMS. The migration cost of a PIMS switch is almost always larger than the gain from a slightly easier integration. There are exceptions. They are rare.
## Chapter 11. Prompts and workflow examples you can copy
These are not magic. They are the structured instructions we give models inside the systems we build. They will be more useful if you read them as architecture, not as scripts.
### Example 1: Refill agent system prompt (simplified)
```
You are the refill triage agent for [Practice Name]. Your job is to evaluate
incoming prescription refill requests and either:
(a) prepare a draft approval for a DVM to release,
(b) deny with a clear, kind explanation, or
(c) escalate to a named human.
You will receive:
- The refill request (channel, requester, patient name, drug, dose).
- The patient record from the PIMS (last exam date, last bloodwork date,
current medications, weight, species, breed, controlled-drug flags,
open notes).
- The practice's refill policy (defined per drug class).
You must:
1. Confirm the patient is current on exams per the policy.
2. Confirm bloodwork status per the drug.
3. Confirm the drug is refillable and the requesting party is authorized.
4. Flag any open notes from the DVM that affect the refill.
5. Output a structured decision: APPROVE_DRAFT, DENY, or ESCALATE,
with the reason in plain English and the supporting fields cited.
You do not write prescriptions. You prepare them. A DVM releases all approvals.
When in doubt, ESCALATE. Do not guess. Do not invent. Do not refer to the
internet. Use only the patient record and the practice's policy.
```
The actual production prompt is longer and includes the practice's specific edge cases (hospice, long-term clients, recent dose changes, controlled-drug schedule rules). The shape is the same.
### Example 2: After-hours triage decision prompt (simplified)
```
You are the after-hours triage agent for [Practice Name]. You are answering
a call between 7:00pm and 7:00am. Your only job is to decide:
(a) is this an emergency that needs to reach a vet right now,
(b) is this urgent and needs a callback in the morning ahead of routine,
(c) is this a routine question that can wait,
(d) is this something you can answer from the practice's published SOPs.
You will:
1. Ask the client for the patient's name and the reason for the call.
2. If any of the following are present, escalate immediately to the
on-call vet via the agreed channel:
bloat, blocked, seizure, hit by car, ingested toxin, not breathing,
blue or pale gums, collapse, profuse bleeding, prolonged labor,
suspected GDV, severe trauma, eye injury, suspected pyometra,
open fracture, paralysis, severe respiratory distress,
altered mental state, suspected heat stroke, foreign body
(especially linear), or any case in which the client states
the animal is dying.
3. If the situation is urgent but not emergent, schedule the patient
for the first available morning slot and add to the morning triage
queue with a one-paragraph summary.
4. If routine, log to the standard morning queue.
5. Always confirm to the client what will happen next.
6. Always write the full conversation, decision, and reasoning back
into the patient's PIMS record before ending the call.
You do not give medical advice. You do not diagnose. If the client asks
for advice that is outside published SOPs, escalate.
```
### Example 3: Scribe post-processing prompt (simplified)
```
You are the post-processing agent for the AI visit scribe. You receive a
raw transcript and a structured set of fields from the scribe vendor.
Your job is to produce a draft SOAP note in the format the DVM at
[Practice Name] uses, using only the information in the transcript.
You will:
1. Identify the patient, the chief complaint, and the history.
2. Build the Subjective from the client's reported information.
3. Build the Objective from the DVM's stated findings only.
4. Draft the Assessment as a problem list with the DVM's stated
differentials.
5. Draft the Plan with the DVM's stated next steps.
6. Flag anything the DVM said that you could not categorize.
You do not invent findings. If the DVM did not say it, it does not go in
the note. Better to leave a section short than to add detail that was
not in the room.
```
### Example 4: SOP agent retrieval prompt (simplified)
```
You are the operations reference agent for [Practice Name]. You answer
questions from staff about how the practice runs. Your only source of
truth is the practice's SOP library.
You will:
1. Identify the staff member's question.
2. Search the SOP library for the most relevant entries.
3. Answer in plain language, citing the SOP by name and section.
4. If no SOP covers the question, say so. Do not guess.
5. If the question is medical (dose, drug, diagnosis), refuse and
redirect to the vet on duty.
6. If the question reveals an SOP gap, log it for the practice
manager to review and add.
You are a reference. You are not a teacher. You are not a clinician.
```
### A note on these prompts
You can copy these as a starting point. You cannot ship them. The production versions are longer, paired with the data integrations, the escalation paths, the fallback behaviors, and the per-practice configuration. The prompt is 20% of the build. The other 80% is the workflow around it.
## Chapter 12. The 90-day rollout
Here is the rollout plan we use. It assumes a single small-to-medium independent practice with one or two locations.
### Days 1 to 10: The workflow map
Goal: write down how the practice actually runs.
Activities:
1. Two days on-site (or shadowing remotely on video) with the front desk, the techs, and the DVMs.
2. Pull two weeks of phone logs, refill requests, and after-hours voicemails into a spreadsheet.
3. Identify the top 10 repeating patterns. (You will be surprised at how few there are.)
4. Write the practice's first version of its refill policy, its after-hours triage rules, and its top 20 client-facing SOPs.
Deliverable: a workflow map and a policy document. This deliverable has value even if you stop here.
### Days 11 to 25: The refill agent build
Goal: ship System 1 to production.
Activities:
1. Build the PIMS integration for read access on the relevant fields.
2. Build the refill agent against the policy from Phase 1.
3. Run it in shadow mode for one week. The agent prepares decisions; a human still acts. Compare.
4. Tune the edge cases the shadow run surfaces.
5. Cut over: agent prepares, vet approves, system sends.
Deliverable: refills handled in 70% to 85% less staff time.
### Days 26 to 45: The after-hours triage
Goal: ship System 3.
Activities:
1. Configure the call-routing and escalation paths.
2. Write the emergency-keyword list with your DVMs in the room.
3. Pilot for one week with the on-call vet shadowing every call.
4. Tune. Go live.
Deliverable: a calm morning queue and a less-brutal on-call rotation.
### Days 46 to 65: The unified comms layer
Goal: ship System 4.
Activities:
1. Connect inbound text, voicemail, email, and portal messages into one inbox.
2. Build the auto-confirm and auto-reschedule rules.
3. Train the front desk on the new inbox.
4. Cut over by channel, one at a time. Texts first, then voicemail, then email.
Deliverable: every message has a home. The team stops searching for context.
### Days 66 to 80: Choose your next move
Goal: decide whether to add the scribe (System 2) or the SOP agent (System 5) next.
If your DVMs are drowning in evening charting, scribe first.
If your turnover is the bigger pain, SOP agent first.
This is the only fork in the rollout. Both end up in the same place. The order depends on which pain is louder.
### Days 81 to 90: Measure and harden
Goal: prove the work.
Activities:
1. Compare the same two weeks of phone, refill, and after-hours data from before the rollout.
2. Survey the team. Honestly. The numbers do not matter if the team is unhappy.
3. Fix what the survey surfaces.
4. Write the post-mortem.
Deliverable: a clear before-and-after, in numbers and in voices, that the owner-vet can show the board, the spouse, or the bank.
---
# Part 4: The decisions
## Chapter 13. Build, buy, or borrow: how to choose
There are three real options. Pick deliberately.
### Buy
Buy when the category is mature, your needs are common, and your scale does not justify custom.
Scribes are buy. The vendors are good, the price is fair, the build is not worth it.
PIMS comms add-ons are usually buy, with the caveat that they are often not good enough on their own, which is where Build or Borrow comes in.
### Build
Build when the workflow is specific to your practice, no vendor sells what you actually need, or the integration matters more than the surface feature. The refill agent is a build. The after-hours triage with full PIMS write-back is a build. The unified comms layer that sits across all your inboxes is a build.
Build does not mean you do it. It means someone (Enertia or another partner) builds it for you, tuned to your workflow, and you own it.
### Borrow
Borrow when you want to try the workflow without committing. Some categories let you start with a vendor and migrate to custom later. After-hours services are like this. You can start with an answering service, learn what your real call patterns look like, and then build the AI version against that data. Borrowing is a respectable first step. It becomes a problem when you stay borrowed for three years and never actually own the workflow.
### The rule
If the system is going to touch your PIMS write paths, your refill policy, or your after-hours triage, lean Build. If the system is going to listen to an exam and produce a SOAP, lean Buy. If the system is going to be your front line at 9pm on a Saturday, you can Borrow your way in but do not stay there.
## Chapter 14. Vendor questions that separate real from theater
Run these on every vendor, including us. If the answers come back vague, glossy, or "let me get back to you," you have your answer.
1. Show me a deployment in a practice of my size and PIMS. Not a slide, a working system.
2. What does your escalation logic look like, in writing, for after-hours calls?
3. How does your tool write into my PIMS? Not "we integrate," but "here is the API call and the fields."
4. What does month four look like, after the honeymoon? What is the failure mode I should expect?
5. Who owns the workflow you build for me? If we end the relationship, what comes with me?
6. Show me your last three reference calls. Not customers you picked. Calls.
7. What is your stance on AI replacing my staff? (If they have one, walk away. The right answer is that AI absorbs load so people can do work that matters.)
8. What is the smallest, most boring thing you would change in my practice in the first 30 days?
9. What is something your tool cannot do that another vendor's tool can?
10. If I told you my refill policy is not written down anywhere, what would you do first?
The last question is the tell. Bad vendors say "we will build it." Good vendors say "we will help you write it down. Without that, the build will be wrong."
## Chapter 15. Governance, safety, and the anxious-pet rule
Three things that every AI system in a veterinary practice has to respect.
### 1. Humans in the loop on anything medical or controlled
No agent releases a prescription unsupervised. No agent runs a controlled-drug protocol unsupervised. No agent gives medical advice to a client. The line is clear, and it is the same line a human assistant would respect.
### 2. The anxious-pet rule, again
When the model is uncertain or the situation is distressed, the call goes to a human. Always. The cost of one false-positive emergency escalation is one extra phone call to the on-call vet. The cost of one false-negative is something no practice should ever absorb. Set the system to err on the side of waking a human.
### 3. Auditable trails
Every agent action that touches a patient record writes a full log. What it saw, what it concluded, what it did, who approved it. This is for your protection, for the client's protection, and for the patient's protection. If a vendor's tool does not produce an audit trail you can read in plain language, do not deploy it.
### A note on consent and recording
Most states allow one-party recording. Some require all-party. The exam-room scribe should announce itself, and the inbound phone agent should disclose that the call is recorded for record-keeping. Run this past your local counsel; the rules vary and they change.
## Chapter 16. What you are signing up for if you do nothing
This part is short.
The practice keeps running. You keep showing up. The phones keep ringing. The refills keep stacking. The new CSR quits at week 11. The next one starts. The on-call vet keeps not sleeping. The owner-vet keeps charting on Sundays. Revenue is fine. Margins are not great. Turnover eats the year.
You are not behind because the work is impossible. You are behind because the work was never sized for the team you have. AI does not make that go away on its own. It buys you back the hours that should never have been spent the way they were.
If you read this guide and decide you can build all of this in-house, do it. We will cheer you on. If you read it and decide the leak is too big to fix while running the practice, you know where we live.
Either way, the work is the work. The fix is design, not magic.
---
# Appendices
## Appendix A: Glossary
PIMS: Practice Information Management System. The platform that holds your patient records, scheduling, and billing. ezyVet, Cornerstone, Avimark, Shepherd, DaySmart Vet, and a few others.
SOAP: Subjective, Objective, Assessment, Plan. The structure most DVMs use for exam notes.
CSR: Client Service Representative. The front-desk team.
DVM: Doctor of Veterinary Medicine.
LVT/RVT/CVT: Licensed/Registered/Certified Veterinary Technician, depending on the state.
Scribe: A tool (human or AI) that captures the exam and produces draft notes.
Triage: The act of deciding what is urgent, what can wait, and what is an emergency.
Shadow mode: Running a new system in parallel with the old one so you can compare, without exposing clients to the new system yet.
## Appendix B: Reference checklists
### Pre-rollout readiness checklist
1. Your PIMS data is reasonably clean on patient demographics and last-exam dates.
2. You have a written-down refill policy (or are willing to write one in week one).
3. You have a written-down after-hours escalation policy (or are willing to write one).
4. The team knows AI is coming and has been told it is meant to absorb load, not replace them.
5. You have a single point of contact on your side who will own the project for 90 days.
6. You have budget for the integration work, not just the software fee.
7. You have realistic expectations about month one being slower than today, on purpose.
### Vendor evaluation checklist
1. Live deployment in a comparable practice on your PIMS.
2. Documented escalation logic, in writing.
3. Native PIMS write-back, not export.
4. Audit trail that a non-engineer can read.
5. Clear answer on data ownership and exit.
6. A reference call, not a reference logo.
7. A first-30-days plan that is boring and specific.
8. Honest answer about what their tool cannot do.
### Pre-mortem checklist (run this before you start)
1. What is the most likely way this build fails at day 45?
2. Which team member is most likely to route around it, and why?
3. Which client segment is most likely to be confused?
4. What is the rollback plan if a piece of this has to come out?
5. Who is the human on your side accountable for the audit trail being read once a week?
## Appendix C: About Enertia and how to talk to us
Enertia Studios builds custom AI systems for real-world service businesses. We are not SaaS. We are not an ad agency. We are not selling generic AI services. We map how your practice actually runs before we build anything, we integrate into the tools you already use, and what we build, you own.
If you read this guide and want to talk, the next step is an Audit. We sit with your data, your phones, and your team for a few days, and we tell you the truth about where the leak is and what we would do about it. The output is a plan. Yours either way.
You can reach us at enertiastudios.com.
The Audit is the only thing we will ask you to do before we ask you to spend money. If after the Audit you decide to build this in-house, with another partner, or not at all, that is a fine outcome. We would rather be the firm you trusted with the truth than the firm that talked you into a project you did not need.
That is the whole pitch.