Inside a teleradiology company: how the machine works

Buy a read and you receive a report; buy from a teleradiology company and you are actually buying a machine you never see: a worklist engine, a shift schedule, a licensing department, a credentialing office, and a QA program, all running behind every signed report. When service is good, that machine is why. When service degrades, one of those departments is where.

This page walks through the machine part by part, because a buyer who understands the anatomy asks diligence questions that bite. AstraRad is one of these machines, and the tour notes how we run a part where it illustrates: scheduled US shifts, subspecialty routing, 1 in 20 reports double-read blind. Where other companies run the same part differently, the difference is named. If you would rather start from the output side, the SLA page publishes what the machine produces; a written rate card follows a request within one business day.

What happens to a study inside a teleradiology company?

Between last-image arrival and signed report, a study passes through five stations, and the elapsed time at each is managed separately.

Station	What happens	Typical elapsed time when tuned
Intake and validation	DICOM arrives; demographics, order data, and clinical history are checked; priors are matched or fetched	1 to 3 minutes, automated
Assignment	The worklist engine computes eligible readers and places the study	Seconds, continuous
Queue	The study waits, ranked by SLA deadline and priority tier	The variable that separates vendors
Interpretation	The radiologist reads, dictates, and signs	5 to 20 minutes by complexity
Delivery	The signed report returns by HL7 to the RIS and EMR; critical findings trigger a phone call	1 to 2 minutes, automated

Two observations from the table. First, the read itself is a minority of the clock; queue time is where turnaround is won or lost, which is why an SLA clock that starts when the radiologist opens the study, skipping the queue, measures the wrong thing. Second, intake quality is a hidden dependency on you: a study arriving without clinical history or priors reads slower and worse, whatever the machine does downstream. The transport and integration layer itself is standard plumbing, covered in the PACS integration guide.

The worklist engine is the company

Strip away the marketing and a teleradiology company is its assignment logic plus its roster. For every arriving study, the engine intersects three constraints to build the eligible reader set: licensure in the patient's state, credentialing at the sending facility, and subspecialty match to the modality and body part. It then ranks by load and deadline, and re-ranks continuously as SLA timers count down. Commercial platforms describe the same design: Intelerad's assignment engine evaluates subspecialty, credentials, workload, and SLA requirements in real time, per its product literature, and Merative markets equivalent orchestration.

The engine's tuning is legible from outside if you know what to ask:

• Ask what happens when the eligible set is empty. A study from a new facility whose readers are not yet credentialed, or a state where only two panel members hold licenses, has to go somewhere. The honest answers involve escalation to a licensing queue or a named fallback reader; the concerning answer is silence.
• Ask how subspecialty routing is enforced. At AstraRad the eligible set for a pediatric brain MRI is pediatric-trained or neuroradiology-trained readers, full stop; at generalist-pool companies subspecialty is a preference weight the engine can override under load. The difference shows up in report quality on exactly the studies where it matters, mapped in the subspecialty reads guide.
• Ask whether STAT re-sorts the queue or jumps it. Re-sorting means your STAT competes with everyone else's; a true priority lane with dedicated capacity is what makes a STAT tier under one hour hold at all.

How does a teleradiology company staff nights?

Nights are the product most clients are buying, and the staffing model behind them varies more than any other part of the machine. Three designs exist. Domestic staffed shifts: radiologists work scheduled overnight blocks, sleep by day, and read mid-shift at 3 a.m. Offshore day coverage: readers in other time zones work their local daytime, which is the US night; the model that built the original nighthawk industry, with the onshore-offshore consequences covered in what US-based actually means. On-call: a radiologist at home is paged as studies arrive, workable at low volume and a fatigue problem at scale.

The clinical argument is about alertness, and it favors any model where the 3 a.m. reader is mid-shift. The staffing math is about coverage depth: a company needs enough scheduled readers per hour, per subspecialty, per license map, to absorb Tuesday's volume and Saturday's spike without the queue stretching. That depth is expensive, which is why thin-night vendors exist and why the overnight coverage cost analysis finds such wide quality variance at similar prices. AstraRad staffs nights as committed US shifts with rest intervals built into the schedule, and the reader shortage that makes night staffing the industry's binding constraint is quantified in the radiologist shortage geography analysis.

The workstation itself is regulated territory: a 2025 review in the National Library of Medicine's open access archive covers the technical requirements for home-based teleradiology workstations, diagnostic displays, bandwidth, and secure PHI handling among them. Serious companies standardize and audit the home reading environment; it is a fair diligence question whether a vendor does.

Licensing operations: the invisible department

Here is the arithmetic nobody puts on a homepage. A radiologist reading for clients in 30 states holds 30 licenses. A 100-reader panel covering the same map is maintaining on the order of 2,000 to 3,000 license-state pairs, every one with its own renewal date, fee schedule, and continuing education requirements, tracked against a routing engine that must never assign a study to a reader whose license lapsed on Tuesday. This is a standing department with software, deadlines, and headcount, and its quality is invisible until the day it is not.

The Interstate Medical Licensure Compact changed the issuance side: 44 states plus the District of Columbia and Guam now participate, checked September 2026. What the IMLC compresses is the wait for a new license in a member state; what it does not touch is maintenance, and the largest non-member, California, still runs its own licensing on its own timeline, which matters to any vendor promising you 50-state coverage next month. A useful diligence pair: how many California-licensed readers the panel holds today, and how many applications sit in process.

Buyer implications: when a vendor says onboarding your new out-of-state site takes 60 to 90 days, licensing lead time is usually the true constraint, and a vendor already licensed in your states can quote weeks instead. Verification takes minutes: pick two named readers and run them through your state board's public lookup before you sign anything.

Credentialing: every hospital client is a bylaws problem

Separate from licensure, and heavier. Each hospital's medical staff office must credential every reader who will interpret its studies: application, primary source verification, references, committee review, and re-credentialing every two to three years. Multiply readers by hospital clients and a mid-sized teleradiology company is administering tens of thousands of credential files; the difference between privileging and credentialing is a real distinction with real paperwork behind each half.

The efficient path is credentialing by proxy under the CMS telemedicine rules, which lets your hospital rely on the distant site's credentialing decisions through a written agreement. A vendor staffed for it delivers a complete, current credential file per reader on request and names the person who owns your credentialing timeline. A vendor that treats credential files as a bespoke favor is telling you what month two will feel like.

The QA machine, and what leaves the building

Inside a functioning company, QA is a production line of its own: a sampling rule selects reports for blind second reads, discrepancies get classified by clinical significance, a monthly meeting reviews every major case and closes it with the signing reader, and the results route both to clients and back into reader management. AstraRad's protocol double-reads 1 in 20 reports blind; the double-reading methodology explains the design choices, and RADPEER scoring covers the field's framework, including the ACR's own statement that no scoring benchmarks have been established.

The diligence question is about the outputs' destination. A QA program whose findings stay internal disciplines nobody; one that sends clients a monthly report with the discrepancy log has put its reader management where you can audit it.

The commercial back office, and why pricing looks the way it does

One more department shapes your experience without ever touching a study: the billing operation, and its design explains most of the pricing structures you will be quoted. A teleradiology company's costs are dominated by radiologist time, which is bought in shifts, while its revenue arrives per study, which fluctuates daily. Every pricing model in the market is an attempt to reconcile those two curves, and knowing that makes vendor behavior predictable.

Monthly minimums exist because a company that scheduled readers for your forecast volume wants payment for the scheduled capacity whether studies arrived or fell short. Platform and integration fees exist to recover fixed onboarding costs on small accounts. Surge premiums exist where the night schedule is thin and a spike forces overtime or rerouting. None of these is dishonest; each is a cost being passed through with a label on it, and each tells you something true about how the machine behind it is staffed.

A company can also absorb those costs structurally instead of billing them, which is the choice AstraRad made: the billing model stays one line, a fixed price per signed report, with no minimums, no platform fees, and no surge premium. The trade-off runs the other direction and deserves stating: a buyer with high, flat, guaranteed volume can sometimes negotiate a committed-minimum rate below any per-report card, because that buyer is solving the vendor's revenue-curve problem and can charge for doing so.

The back office also runs invoice reconciliation, and its quality is testable in the pilot: every invoice line should tie to a signed report identifiable by accession number, with the priority tier and its multiplier printed on the line. An invoice you cannot reconcile study by study is a process failure that compounds monthly, and it is far easier to reject in week six than in month sixteen. Ask for a sample invoice during procurement; the good ones look boring, and boring is exactly what a controller wants to see.

What the anatomy tells you to ask

Each department above converts to one question a sales deck cannot answer with an adjective, and together they are a 20-minute diligence call:

1. Which staffing model covers 2 a.m. to 6 a.m. on my account, and is it the same on Saturday?
2. What does your worklist engine do when the eligible reader set for my study is empty?
3. How many of your readers hold licenses in my state today, and may I verify two by name?
4. Who owns my credentialing timeline, and do you support credentialing by proxy?
5. What is your QA sampling rate, your major discrepancy rate, and who receives the monthly results?

The pattern in the answers matters more than any single one: a company that answers all five in writing is running the machine deliberately, and a company that cannot is discovering its own operations at the same time you are. AstraRad's answers are standing and published where they can be checked, on the SLA page and through how we work; the rate card against your own case mix arrives within one business day of a request.