Referral Bottleneck Analysis: Separate Staff Work From Waiting Time

Referral bottleneck analysis reconstructs how a referral moved from a defined starting event to a defined outcome. Record active staff work, waiting, and dependencies that ran at the same time. Then identify what prevented the next step. A useful analysis explains an observed delay and tests a specific change, while keeping uncertain records and unfinished referrals visible.

Choose the question before exporting timestamps

"Our referrals take too long" can refer to several different problems. Incoming documents may sit before anyone opens them. Completed intake may wait for another office. A ready-to-schedule referral may wait for contact, or a patient may receive an appointment far in the future.

These intervals need different decisions. Start with one question, such as: "Why does accepted intake wait before a confirmed booking?" Write the endpoint precisely. A booking created in the EHR is different from an attended visit, and a fax transmission time is different from the receiving office's first documented receipt.

The referral management process provides the broader lifecycle. This worksheet examines the timing of a selected part of that lifecycle. Include the entry channels, locations, administrative task, observation period, and exclusions in the review record.

Record the clock convention too. Calendar elapsed time includes nights and weekends. Business-hours elapsed time includes only a specified service calendar. Either can answer an operational question, but they cannot be compared under the same label.

Draw what happened

Ask the people doing the work to map a small set of recent episodes. Include a straightforward case, a repeatedly returned case, a case that waited on two things, and an unfinished case. Use this as a diagnostic sample, not a statistically representative estimate of all referrals.

The Institute for Healthcare Improvement's flowchart guidance distinguishes broad process maps from detailed maps that expose rework and complexity. The event worksheet below is an original application to administrative referral timing, not an IHI standard measure.

Start with the sequence staff can substantiate. For each arrow between steps, ask what had to become true before the next action could happen. Two departments appearing next to one another on an organizational chart does not establish a serial dependency.

A coordinator might request a missing document and seek clarification of an administrative scheduling rule simultaneously. If both must be resolved before booking, the later response determines when that gate opens. If either would have been sufficient, the dependency works differently. Record the actual rule instead of guessing from the order of status labels.

Build an event table that preserves the source

Use the following fields for each episode. The investigation should be possible to reproduce without relying on the analyst's memory.

Retain the original event history when a record is corrected. A backdated note can describe when something occurred, but its entry time alone cannot establish that earlier event.

The referral tracking state dictionary helps distinguish episode status from parallel tasks. For this analysis, preserve the task history rather than trying to infer it from the final status.

Measure touch time directly

An item opened at 9:00 and saved at 9:40 does not prove 40 minutes of staff effort. The coordinator may have answered calls or worked on other records. Similarly, an automated event can change a status without using coordinator labor.

For a bounded work sample, observe the handling task or use a suitable activity log whose meaning has been validated. Record interruptions, rework and the person performing the task. AHRQ's time and motion resource describes direct observation of task duration. It is methodological background; its historical clinical examples are not referral productivity benchmarks.

Keep labor minutes and elapsed minutes separate. If two people each work for ten minutes at the same time, that is 20 staff-minutes within ten elapsed minutes. Do not add those 20 labor minutes to the timeline as if they were sequential.

Use the observations to understand the work design. A short sample is not a fair basis for ranking individual employees, setting a universal time allowance, or assuming every referral needs the same effort.

Reconcile a complete example

Consider a hypothetical administrative referral. The observations below use calendar time in one time zone. Monday and Wednesday are labels within a synthetic week, not reported customer dates. Both requested clarifications are required before the task can proceed.

The document and rule-clarification waits overlap. Their combined span is 48 hours, not 76 hours. Adding both durations independently would count the first 28 hours twice.

The total timeline is 60 + 20 + 2,880 + 220 + 10 = 3,190 minutes, or 53 hours and 10 minutes.

Observed handling is 30 staff-minutes. The remaining 3,160 elapsed minutes are waiting under this example's mutually exclusive classification. Request preparation is already included in the 20-minute intake task and is not counted again.

This result does not mean that all waiting was unnecessary, or that 3,160 minutes of payroll could be saved. It identifies where elapsed time accumulated. The team can now investigate why the longer clarification took 48 hours and why another 220 minutes passed after readiness.

Test which change would open the gate sooner

Suppose document clarification had arrived two hours earlier. The administrative-rule clarification would still finish on Wednesday at 10:20. In this example, the earlier document response alone would not change readiness.

Now suppose the rule clarification were resolved after 24 hours instead of 48. The document would still take 28 hours. Readiness would move forward by 20 hours, not 24. That is a conditional timeline calculation, assuming the other dependency remains unchanged.

A 20-hour improvement in readiness is not automatically a 20-hour improvement in booking. Staff coverage, contact availability, appointment supply and later task timing may still determine the final outcome. Observe the complete selected interval after the change.

Use this test to compare plausible interventions. A shared clarification source could address repeated rule questions. An accepted-owner handoff could address work waiting after readiness. Each needs evidence that it solves the observed problem, not just an appealing feature description.

Keep open cases and missing evidence visible

Looking only at completed referrals can make a process appear faster while difficult work remains unresolved. Define a receipt cohort, retain its open cases, and record each open episode's age at the observation cutoff. Do not treat that age as its eventual completed cycle time.

For example, a hypothetical sample contains 50 received episodes. Thirty-two have reliable starting timestamps and confirmed endpoints, ten remain open with reliable starting timestamps, and eight lack reliable starting timestamps. These are mutually exclusive analysis groups; record the eight excluded episodes' operational completion states separately. A completed-case median describes the 32 fully timed completed episodes only. It does not describe all 50.

Report the eight timing exclusions as 8/50, or 16% of the sample. Do not silently remove those episodes from the operational worklist. They need an evidence-repair owner even when they cannot yet support a reliable duration calculation.

Also separate an unresolved timing record from an unresolved patient task. The task may be finished while the historical timestamp is missing. Conversely, a beautifully complete log may describe a referral that has not reached its endpoint.

Compare like work and choose one improvement

Group episodes by the administrative path that changes the work: intake channel, required information, receiving location or assigned workflow. Keep the classification narrow enough to explain a decision without producing tiny groups that encourage false precision.

Compare the same clock convention, start event, endpoint and observation window. Use the referral operations dashboard to publish stable definitions once the investigation establishes them. Use booking conversion separately when the question is how many referrals reach a booking rather than how long a particular interval takes.

Write an improvement card before changing the workflow:

Set the review period and decision thresholds locally before examining the result. Preserve negative findings. Faster handling in one queue is an incomplete success if another queue inherits additional unresolved work.

For referral coordination automation, the most useful opportunity statement names the event to improve, the evidence that will confirm it, and the dependency that automation does not control.

FAQ

Sources

• Institute for Healthcare Improvement, Flowchart, process-mapping context. The referral event table and all scenarios here are original operational examples, ihi.org
• AHRQ, Time and Motion Studies Database, background on observing task duration. No productivity figure or technology-effect estimate is taken from this historical resource, digital.ahrq.gov