Could You Draw Your Operations Stack on a Whiteboard Right Now?

Try it as a thought experiment. Somebody hands you a marker and asks you to sketch how work actually moves through your organization: where a patient request enters, which system holds the truth, what talks to what, and where a human has to bridge two things that do not connect.

Most people can draw the EHR. After that it gets harder. There is the phone system, and something for texting, and a separate reminder tool that somebody in marketing set up, and the payer portals, and the fax line that everyone insists is nearly gone. Somewhere in the middle of the diagram you end up drawing a person, because the honest answer to how two systems connect is that Denise copies it across every morning.

That drawing is your healthcare operations stack, and the fact that it is hard to sketch is the finding, not a failure of memory. You cannot optimize an architecture you cannot describe, and you certainly cannot decide where AI belongs in it. A healthcare operations stack that resists description is one nobody is accountable for.

This article is an attempt at a cleaner model. Not a product landscape, since those change quarterly, but a way of thinking about the layers of a healthcare operations stack that stays useful when the vendor names change.

Nobody Designed This

It is worth saying plainly that the current state is not anyone's fault. A healthcare operations stack accumulates rather than gets designed. Each purchase solved a real problem on the day it was made, usually urgently, usually by a different person than the one who made the previous purchase.

The consequences are structural rather than managerial. Capabilities overlap, so three systems can all send a reminder and none is authoritative. Gaps persist, since nothing was ever bought to handle the space between departments. And the integration burden falls on people, because connecting two systems properly costs money that connecting them with a staff member appears not to.

That last point deserves emphasis. Manual integration looks free. It is charged to headcount rather than to the technology budget, which is why it survives review after review while a five-figure interface engine does not.

MGMA's May 2026 poll of practice leaders found that among organizations where AI had not moved productivity, adoption and integration friction were leading explanations, including inconsistent use and poor interoperability that breaks the workflow. The reported obstacle was rarely the intelligence of the tool. It was the stack around it.

The Five Layers of a Healthcare Operations Stack

Here is the model. Five layers, each with a distinct job, each failing in a distinct way.

Layer One: System of Record

Your EHR and PMS. This is where clinical and financial truth lives, and it is the layer you change least often and most carefully.

The job of every other layer is to keep this one accurate. Anything that creates a parallel record, a spreadsheet of pending authorizations, a shared inbox of referrals, a scheduler's private notes, is a symptom of a missing capability elsewhere in the stack.

Layer Two: Integration and Data Exchange

Interfaces, APIs, clearinghouse connections, and increasingly FHIR endpoints. This is the least visible layer and the one that determines what everything above it can actually do.

Organizations consistently underinvest here, because this layer produces no demo. It also matters most: a capability at the workflow layer is worth nothing if it cannot read and write reliably at the record layer.

Layer Three: Workflow Execution

The layer that carries a task from request to completion: applying scheduling rules, verifying eligibility, submitting an authorization, matching a referral to a chart, retrying when a portal times out, and escalating when a person is needed.

In most organizations this layer does not exist as technology. It exists as staff. That is the single most important observation about the typical healthcare operations stack, and it is where the largest opportunity sits.

Layer Four: Patient Engagement

Phone, text, portal, and web. Every channel through which a patient reaches you or you reach them.

Most organizations have plenty here, often too much and poorly coordinated. Adding another channel is the most common response to an access problem and rarely the correct one, since the constraint usually sits at layer three rather than layer four.

Layer Five: Intelligence and Analytics

Reporting, dashboards, forecasting, and decision support. Everything that turns what happened into what to do next.

This layer is only as good as the data the layers beneath it capture. Organizations whose workflow execution is manual have no reliable data about it, which is why access reporting so often stops at call volume. Our list of front desk KPIs covers what becomes measurable once execution is instrumented.

Where AI Actually Fits

Now the central argument, which is architectural rather than technological.

Most healthcare organizations buy AI as a feature of layer four. It answers the phone, so it feels like a communication tool, and it gets evaluated against other communication tools on how well it converses. Purchased that way, it produces a better front door and leaves the building unchanged, which is the most common architectural error in a healthcare operations stack today.

AI Agents belong at layer three. Their value comes from executing workflows end to end against the system of record, using the integration layer, and generating the data that makes layer five meaningful. The conversation is how the work arrives, not what the product is.

This distinction has practical consequences for evaluation. If you are buying an engagement tool, you compare voice quality, channel coverage, and patient satisfaction. If you are buying a workflow execution layer, you compare write-back depth, rule enforcement, exception handling, retry behavior, escalation design, and audit trail. Those are different purchases with different questions and frequently different budget owners.

It also reframes where a Voice AI investment should sit organizationally. A tool bought by marketing to improve patient communication will be measured on satisfaction. A workflow execution layer bought by operations will be measured on completion and capacity. The second framing survives budget scrutiny considerably better.

Worth stating clearly: this is not an argument that AI should occupy every layer. Systems of record should stay systems of record. The intelligence layer benefits from AI but is not replaced by it. The specific claim is narrower, which is that layer three is currently staffed rather than built, and that is where agents create capacity.

Three Patterns That Quietly Consume Your Budget

Three failure modes recur often enough to name.

‍Point-solution sprawl. A tool per problem, each with its own login, its own reporting format, and its own integration. Each was justified individually. Collectively they produce a stack where nobody can answer a cross-cutting question and every renewal is negotiated separately. The tell is that your annual software list has grown faster than your patient volume.

‍The integration tax. Every additional system multiplies connection points rather than adding one. Ten systems that all need to talk have far more possible connections than five. Most organizations pay this tax in staff time rather than in interface budget, which keeps it invisible until someone measures how long people spend rekeying.

‍The shadow stack. The spreadsheets, shared inboxes, personal task lists, and paper logs that staff build to cover gaps. Shadow stacks are excellent diagnostic information, since each one marks a place where the official healthcare operations stack failed to do something necessary. They are also fragile, undocumented, and usually the reason a departure causes a three-month operational dip.

The remedy for all three is the same and unglamorous: know which layer of your healthcare operations stack a new purchase belongs to before you buy it, and refuse purchases that cannot answer the question.

What Actually Changes by 2027

Forecasting is mostly noise, so here is the part that is already written into regulation rather than into predictions.

Under the CMS Interoperability and Prior Authorization final rule, impacted payers must implement and maintain FHIR-based APIs, with compliance dates generally beginning January 1, 2027. Four APIs are covered: Patient Access, expanded to include prior authorization information, plus new Provider Access, Payer-to-Payer, and Prior Authorization APIs. Operational provisions, including decision timeframes of 72 hours for expedited requests and seven calendar days for standard ones, along with specific denial reasons and publicly reported metrics, generally began a year earlier.

For provider organizations the relevant hook is the new Electronic Prior Authorization measure, which MIPS eligible clinicians report beginning with the calendar year 2027 performance period, and eligible hospitals and critical access hospitals with the 2027 EHR reporting period. It is an attestation that at least one prior authorization was requested electronically through a Prior Authorization API using certified EHR technology.

Three implications follow for a healthcare operations stack.

First, layer two stops being optional. An organization whose payer interactions run through portals and fax will be structurally slower than one consuming APIs, and the gap widens as adoption grows. That gap is currently wide: CAQH CORE reports that only around 35 percent of medical prior authorizations are conducted fully electronically using the standard transaction.

Second, the workflow layer gets more valuable, not less. Faster payer responses only help if something acts on them promptly. An API that returns a documentation requirement at 2 a.m. is worth little if the request sits until someone opens a queue.

Third, expect the questions asked of vendors to shift. Interoperability posture, API readiness, and how a product handles both FHIR and legacy transactions become reasonable procurement questions rather than technical trivia. Our guide to insurance verification and prior authorizations covers the operational side of these workflows.

Build, Buy, or Configure, Decided by Layer

The decision is different at every layer, which is why blanket build-versus-buy debates go nowhere.

‍System of record: buy, and change rarely. The switching cost is enormous and the differentiation is low. Configure heavily instead.

‍Integration: buy the platform, own the strategy. Very few provider organizations should be hand-building interfaces in 2027, but every one should know which data flows matter and who is accountable for them.

‍Workflow execution: buy, and buy healthcare-native. This layer requires deep knowledge of scheduling rules, payer behavior, and clinical escalation that is expensive to build and hard to maintain. Generic automation platforms tend to founder on exactly the edge cases healthcare produces constantly.

‍Patient engagement: consolidate rather than add. Most organizations already have more channels than they coordinate.

‍Intelligence: start with what your other layers actually capture. Analytics ambitions usually exceed data quality, and the fix is instrumenting layer three rather than buying a better dashboard.

One cross-cutting rule: prefer components that make the layer beneath them more accurate. Anything that creates a parallel source of truth is adding to your shadow stack regardless of how good it is in isolation.

Here's How Confido Health Can Help

This article argued that the missing piece in most organizations is the workflow execution layer, currently staffed rather than built. That is precisely the layer Confido Health occupies. Our AI Agents take a patient request from whichever channel it arrives on, execute it against your EHR or PMS through real integration, handle the exceptions, escalate what needs a person, and generate the operational data your reporting layer has been missing.

Here is what Confido Health delivers:

• A workflow execution layer, not another channel, completing scheduling and rescheduling, eligibility and benefits verification, prior authorization submission and follow-up, referral and fax intake, refill coordination, payment collection, recalls, and waitlist backfill
• Integration-first approach with 40+ EHR and PMS systems including Epic, Athenahealth, and eClinicalWorks, plus telephony platforms, so layer two is a capability rather than a project
• Operational visibility from execution data, with dashboards showing how calls, scheduling, refills, and payer workflows perform, which is the reporting most stacks cannot produce because the work was never instrumented
• Empathetic, natural conversations with 97 percent patient satisfaction, in more than 20 languages, answering every call on the first ring around the clock
• Proven ROI, with up to 70 percent reduction in staff call burden, 60 percent reduction in cancellations, 80 percent reduction in manual administrative work, 75 percent faster prior authorization processing, and a 15 to 20 percent increase in revenue collections
• Live in under 30 days using expert-approved templates co-built with practicing physicians and operations leaders, so the layer gets added without a multi-year program

Confido Health is more than a tool. It is the execution layer between your systems of record and your patients, doing the work your team currently does by hand between screens.

Want to see which layer of your healthcare operations stack is actually the constraint? Let's get started today.

Still in research mode? Start with our explainer on what an AI voice agent is, then read what matters for COOs.

Frequently Asked Questions

What is a healthcare operations stack?

A healthcare operations stack is the set of connected systems that run administrative work: the system of record, integration and data exchange, workflow execution, patient engagement channels, and intelligence or analytics. Most were accumulated purchase by purchase rather than designed as an architecture.

What are the layers of a healthcare technology stack?

Five layers are useful in practice: system of record such as the EHR and PMS, integration and data exchange, workflow execution, patient engagement, and intelligence. Each has a distinct job and fails differently, so purchasing decisions should be made per layer rather than for the stack overall.

Where does AI belong in a healthcare operations stack?

At the workflow execution layer. AI Agents create value by carrying tasks to completion against the system of record, not by adding another communication channel. Buying AI as an engagement feature produces a better front door while leaving the underlying operational constraint unchanged.

What is the integration tax?

The integration tax is the compounding cost of connecting systems as their number grows, since possible connection points rise faster than the system count. Most organizations pay it in staff time spent rekeying data rather than in interface budget, which keeps the cost invisible during reviews.

What is a shadow stack in healthcare operations?

A shadow stack is the informal set of spreadsheets, shared inboxes, personal task lists, and paper logs staff create to cover gaps in official systems. Each one marks a missing capability, and they are useful diagnostics, though fragile and usually undocumented when someone leaves.

What changes for healthcare interoperability in 2027?

Under the CMS Interoperability and Prior Authorization final rule, impacted payers face compliance dates generally beginning January 1, 2027 for FHIR-based Patient Access, Provider Access, Payer-to-Payer, and Prior Authorization APIs. Related operational provisions and decision timeframes generally took effect a year earlier.

What is the MIPS Electronic Prior Authorization measure?

It is an attestation measure added by CMS under Promoting Interoperability. MIPS eligible clinicians report it beginning with the calendar year 2027 performance period, attesting that at least one prior authorization was requested electronically through a Prior Authorization API using certified EHR technology.

Should healthcare organizations build or buy workflow automation?

Buy, and buy healthcare-native. The workflow execution layer requires detailed knowledge of scheduling rules, payer behavior, and clinical escalation paths that is costly to build and harder to maintain. Generic automation platforms tend to struggle with the edge cases healthcare generates continuously.

Why do AI deployments fail to deliver operational results?

Frequently because of the surrounding healthcare operations stack rather than the model. MGMA polling found integration friction, inconsistent adoption, and interoperability problems that break workflows among the leading explanations where AI had not improved productivity, rather than any shortcoming in the underlying technology itself.

How do you decide where a new tool fits in your stack?

Identify which layer it occupies before evaluating it, then ask whether it makes the layer beneath more accurate. Anything creating a parallel source of truth adds to your shadow stack. Tools that cannot be placed in a layer usually overlap something you already own.