Accessibility research
Do relay calls receive the same intake outcome as voice calls?
A field study for testing whether relay calls reach the same services, decisions, and follow-up ownership as comparable voice calls.
The decision this study supports
A business can answer a Telecommunications Relay Service call politely and still give that caller a worse result. The operator may be transferred unnecessarily, asked to call back without relay, denied enough time to type, or routed away from an intake path available to a direct voice caller. A useful review therefore asks whether comparable requests reach comparable decisions, not merely whether staff accepted the connection.
The U.S. Department of Justice explains that covered entities must accept TRS and video relay calls and treat them like other calls. Its ADA effective communication guidance also explains that communication methods must fit the person and the situation. Those statements are authoritative accessibility inputs. They do not supply a call-center score, a universal handle-time target, or evidence that any particular business falls under a specific legal rule.
This protocol is designed for a small business evaluating an approved virtual-assistant intake lane. It is operational research, not legal advice or a claim about VirtualAssistantCallCenter customer results. Management remains responsible for determining applicable duties and approving the services, scripts, disclosures, identity checks, and escalation boundaries used in its own operation.
Define parity before reviewing calls
Parity means access to the same legitimate outcome under the same business rules. It does not mean that every interaction has identical duration or wording. A relay conversation may take longer because each turn passes through a communications assistant. Longer duration by itself is not a defect. The meaningful comparison is whether the caller can convey the request, receive material information, correct errors, choose an available option, and leave with the same type of owned next step.
Write a case-matching rule before collecting records. Match on request class, operating-hours window, location or service area when relevant, language lane, urgency category, and the policy version in force. Do not compare an after-hours emergency relay request with a routine daytime voice inquiry. If no credible comparison exists, report the relay case descriptively rather than forcing it into a pair.
Declare outcomes such as completed intake, scheduled appointment, accurate information, accepted transfer, owned callback, or documented ineligibility. A vague disposition such as "handled" cannot establish parity. A caller who receives an owned callback can have a valid outcome even if the original call did not complete every step, provided the same fallback would be available to a comparable voice caller.
Build an evidence record without collecting excess data
Use one pseudonymous case identifier from first answer through closure. Record the incoming channel as relay, video relay, speech-to-speech, direct voice, or unknown only when the source supports that classification. Capture request class, queue, hours status, script version, transfers, identity-check branch, corrections, offered choices, disposition, owner, promised follow-up window, acknowledgment, and final state. Keep raw phone numbers, recordings, transcripts, disability information, and account details in their approved operational systems.
The review table should not infer a disability or demand a diagnosis. Relay use is a communication channel, not evidence about a person's capacity, identity, or authorization. The relay operator's presence also must not be treated as a third-party account representative without examining the business's applicable rule. Reviewers should use event evidence and approved policy, not assumptions about the caller.
Separate event time from the time a platform stored the event. Retain the timezone and clock source for queue arrival, answer, transfer, callback creation, owner acknowledgment, and closure. If a transcript is necessary for a declared review question, restrict access and extract only the coded finding needed for analysis. Delete working extracts under an approved retention schedule.
Observe the interaction at decision points
Code whether the initial answer recognized the relay format without suspicion or forced restart. Then test whether the assistant allowed complete turns, addressed the caller rather than the communications assistant, and avoided imposing an ordinary voice pace. Examine whether menus, verification prompts, one-time codes, or transfer bridges functioned through the chosen channel.
At each material choice, determine whether the caller received the same information a matched voice caller would receive. This includes available appointment windows, service boundaries, costs only when approved for disclosure, fallback choices, and the reason for any refusal. Record whether correction opportunities were meaningful. A spelling or number readback that moves too quickly for a response is not equivalent to a confirmed field.
Follow the case beyond disconnect. Test whether promised messages used an accessible and approved channel, whether the downstream owner could continue through relay, and whether a new staff member forced the caller to repeat the entire request. A technically successful transfer can still fail parity if the destination rejects relay calls or has no context.
Sampling and comparison design
Include every eligible relay call in the observation period when volume is low. For each, select one or more voice comparisons through a reproducible rule, such as the next eligible case in the same request class and hours band. Freeze the selection query before examining outcomes. Exclude tests, duplicates, and abandoned calls only under written rules, and publish exclusion counts.
Small samples should be reported as case counts and observed ranges. Do not turn three successful calls into a broad percentage claim. For larger samples, report the proportion reaching each declared outcome, the proportion with an owned unresolved exception, and the distribution of time from arrival to an accountable next step. Show relay and matched voice results separately.
Duration is a diagnostic measure, not the primary success measure. Break it into queue wait, conversation, transfer wait, and follow-up delay. This prevents the expected pacing of relay communication from being mislabeled as agent inefficiency. Review abandonment cautiously because a carrier or bridge failure can look like caller choice.
Classification and review consistency
For every checkpoint assign supported, contradicted, unknown, or not applicable. Supported requires evidence of the defined event. Contradicted means evidence shows a different path. Unknown covers missing, conflicting, or inaccessible evidence. Not applicable is allowed only when the protocol specifies why that checkpoint does not belong to the case.
Have a second reviewer independently code a declared subset containing both relay and voice records. Compare decisions about outcome equivalence, interruption, correction opportunity, transfer acceptance, and follow-up ownership. Preserve disagreements and their resolution. High disagreement may reveal an unclear rubric rather than inconsistent service.
Reviewers should not listen for accent, fluency, or disability as quality shortcuts. They should locate evidence tied to the defined decisions. If recording quality prevents a conclusion, the result is unknown. A clean CRM disposition must not override contrary queue events or a missing owner acknowledgment.
Interpretation, limitations, and action
This design measures documented process outcomes. It cannot determine the caller's satisfaction, prove that communication was subjectively effective, or certify ADA compliance. Relay types differ, matched cases may still differ in complexity, and platform logs can omit work completed outside the system. Staff may behave differently when they know a review is underway.
Treat a disparity as a prompt for diagnosis. Repeated transfer rejection may require destination training or routing repair. Verification failures may require an approved alternative path. Turn-taking problems may require revised examples and quality criteria. Missing follow-up evidence may require one accountable queue rather than another script sentence.
For each change, name the owner, affected request classes, effective time, expected mechanism, validation sample, and rollback condition. Begin a new cohort after material routing, script, or platform changes. Do not erase the earlier result by relabeling cases under the new rule.
The decision rule is straightforward: a relay case demonstrates supported parity only when it reaches the same class of permissible outcome as its valid comparison, with material choices and corrections available and any follow-up accepted by an identifiable owner. Unknown evidence stays unknown. The business should improve the stage where equivalence breaks instead of pressuring callers or assistants to make relay conversations resemble ordinary voice calls.
Sources checked
- U.S. Department of Justice, ADA Requirements: Effective Communication, checked September 28, 2026.
- World Wide Web Consortium, Web Content Accessibility Guidelines 2.2, checked September 28, 2026.
- National Institute of Standards and Technology, NIST Privacy Framework, checked September 28, 2026.