Referral Automation References: Ask What Happened After Go-Live

Why reference calls go wrong

A positive reference call can leave important questions unanswered. "The team was helpful" does not tell you who handled a failed update, what remained with staff or whether the deployment resembled the one you are buying. Set the topics before the call, and distinguish the reference's direct observations from secondhand explanations.

The Agency for Healthcare Research and Quality frames health IT success around its impact on both clinical and administrative workflow, and its Workflow Assessment for Health IT Toolkit is built for people involved in planning, design, implementation and use of health IT in ambulatory care (AHRQ). Use that workflow perspective to organize the reference interview. The call guide below is an original editorial method; AHRQ does not validate this questionnaire or establish the quality of a vendor.

If you want the broader context on vendor diligence, start with how to evaluate healthcare AI vendors and the referral management software buyer's guide. This article covers only the reference conversation.

Step 1: Choose references relevant to the workflow and organization

Ask the vendor for references that resemble your situation on four dimensions:

• Workflow. Inbound referral intake, outbound referral tracking, scheduling follow-up, or prior authorization status tracking are different jobs. Ask for a reference running the workflow you plan to buy first.
• EHR and integration path. Ask which EHR, version and integration path the reference uses. A different environment may still reveal useful support and operating experience, but it does not verify the field-level behavior of your proposed connection. Ask your vendor to demonstrate that configuration separately.
• Size and volume. Record the number of sites, workload and staffing arrangement relevant to the reference's example. Ask which differences from your setting might change the work.
• Time since go-live. Ask what changes the reference has experienced since launch, such as a renewal, vendor release or EHR upgrade. A recent deployment may help explain implementation but cannot yet describe those later events.

Also ask the vendor directly: why these references, and is there a customer who stopped using the product you can speak to? If that conversation is unavailable, record the limitation. Do not infer a reason for the absence or treat it alone as proof about product quality.

Step 2: Ask what the team had to do before and after launch

The most useful reference questions are about work, not satisfaction. Use this list and keep answers in the reference's own words.

Before go-live

1. Who on your side owned the implementation, and roughly how much of their time did it take?
2. What data, access or policy decisions did the vendor need from you before build could start?
3. Did the first live workflow match what you expected to buy, or was it narrowed?
4. Which events mark your implementation timeline: agreement, access ready, testing complete and first workflow processing real work? Which dependencies affected the interval between them?

After go-live

5. What did staff do differently in the first month? What did they stop doing?
6. Which tasks still come back to a person, and how are they queued?
7. How many support tickets or escalations did you raise in the first ninety days, and what were they about?
8. Did the vendor change how the workflow behaves without asking you? How did you find out?
9. What is still on your open requirements list today?

Step 3: Investigate support effort, exceptions and scope changes

Follow up on three sources of work that belong in your operating-cost investigation. A reference can describe them at its site; your own estimate still needs the proposed scope and assumptions.

Support effort. Ask who answers when something breaks, how quickly, and whether the reference pays for a support tier. Ask whether the vendor's accountable owner changed after launch.

Exceptions. Ask what the automation does when a referral arrives with missing demographics, a contradictory insurance record, or an unsupported specialty. The Linear Health Operational AI Evaluation Kit recommends asking a vendor to demonstrate behavior when information is missing, contradictory, late or outside scope, and asks buyers to establish what completion means for the task (Linear Health evaluation kit). If a reference describes an exception as rare or routine, ask what period and records support that description. Keep an impression separate from a measured count.

Scope changes. Ask what was descoped, deferred or re-priced between the proposal and the live workflow. Then ask what was added later and whether the fee changed. Keep recurring fees, implementation effort and internal labor in separate columns; capacity freed on staff is not the same as cash saved.

The same evaluation kit suggests scoring vendor evidence on a 0 to 3 scale from not addressed, to claimed, to demonstrated, to verified in production (Linear Health evaluation kit). Use a separate label for reference testimony so it is not confused with a vendor assertion or your own verified production result. Note whether the speaker directly observed the event and whether supporting records were inspected. The call should identify what you still need to test, not automatically move a proposal to the highest evidence level.

If you are evaluating Linear Health alongside other vendors, you can book a demo and bring the gap list from your reference calls as the agenda.

Step 4: Separate a reference's observations from verified general claims

A reference saying "our referral coordinators spend less time on faxes" describes its experience. Ask which task changed, how the before-and-after periods were measured and what new review work appeared. A vendor's general automation claim answers a different question. Keep testimony, measured site results and advertised product scope separate, and test the assumptions relevant to your own business case.

Use the evidence comparison worksheet below to keep them apart. It is editorial guidance, not a measured instrument.

Evidence comparison worksheet

Labelled synthetic example

The following is an illustrative, invented example. It is not a Linear Health customer, not a measured result, and not a benchmark.

A fictional practice manager speaks with two references for a referral intake vendor. Reference A says staff no longer index incoming documents manually but still resolve uncertain patient matches. Reference B says the first workflow was narrowed to one specialty before another was added. The manager records both accounts with their operating context, then requests a demonstration of an uncertain-match case and the proposed scope and change process. Neither account establishes how much work the manager's own practice will save.

Step 5: Record unanswered questions for the vendor demonstration

Close every call by listing what the reference could not answer. Check for unanswered questions about field-level write behavior in your EHR, security and contracting documentation, pricing at your proposed volume, and the transition process. Assign each gap an owner on your team. The evaluation kit treats privacy, security, safety and authorization questions as gates that a strong overall impression cannot offset (Linear Health evaluation kit). Send those to your accountable reviewers rather than closing them on a reference's say-so.

Use the Operational AI Evaluation Kit to structure one review per vendor, export or print each review, and keep the consolidated gap list in your own decision record. The referral coordination software page describes Linear Health's scope if you want to compare it against what references describe. When you are ready to test the open items in a live walk-through, book a demo and ask the vendor to run your exception case rather than a scripted path.

Frequently asked questions

Sources

Sources accessed October 26, 2026.

• AHRQ, Workflow Assessment for Health IT Toolkit.
• Linear Health, Operational AI Evaluation Kit.