Running Bilingual Outreach Without Running It Twice

Most health centers have solved bilingual outreach on the inbound side. Someone answers in Spanish, the front desk is staffed accordingly, and a language line covers the rest. Then the recall campaign goes out in English to the whole panel, and half the people it was built for never engage with it.

The asymmetry is structural rather than negligent. Inbound is reactive, so the language shows up at the moment of contact and a person adapts. Outbound is built in advance, by whoever has capacity, usually in one language, and the language decision gets made implicitly by the person writing the script.

So the outreach that matters most at a health center, recall lists, open orders, wellness visits, patients who have gone quiet, reaches the English-preferring half of the panel and rolls off the other half. The campaign reports a completion rate that looks acceptable in aggregate and hides the fact that it did not work at all for a large share of the people it targeted.

This is the core panel, not an edge case

The scale of this is easy to underestimate if you have not looked at the national figures recently.

In the Health Resources and Services Administration’s national Uniform Data System figures, 26.97% of health center patients were reported as having limited English proficiency in 2025, which is 8,832,374 people out of roughly 32.7 million patients served. At an individual center the share is often far higher, and it is concentrated in exactly the populations that recall and preventive outreach are designed to reach.

So an English-only campaign at a typical health center is not missing a minority segment. It is missing a quarter of the panel nationally and frequently the majority locally, and it is missing them systematically rather than randomly.

That also means the completion rate is a misleading number. A campaign that reaches 60% overall may be reaching 80% of one language group and 30% of another, and the aggregate is the only figure anyone looks at.

Language access is an obligation before it is a preference

There is a compliance floor here, and it is worth knowing where it sits before treating this as a service upgrade.

Federal regulation at 45 CFR 92.201 requires covered entities to take reasonable steps to provide meaningful access to individuals with limited English proficiency. The rule is written around access to programs and activities, which includes the communication a practice initiates, not only the conversation it answers.

The practical reading is that outbound contact about care a patient is due for is part of the program, so the language it goes out in is not purely a marketing choice. A recall campaign that reaches only English speakers is a program access gap with a paper trail.

This is not a reason to panic and it is not the strongest argument for doing it. The strongest argument is that the outreach works. But it does mean bilingual outreach belongs on the operations list rather than the someday list, and it changes who at the organization has to care about it.

The language field is the campaign, and it is often wrong

Running a campaign in two languages is not two campaigns. It is one campaign with a routing field, and the quality of that field determines everything.

athenaOne carries language preference on the patient record, and the reference sets for languages are part of practice configuration. Reading the panel through GET /patients gives the segmentation directly, so a recall list splits into language cohorts before the first call is placed rather than after someone notices the Spanish-preferring patients are not responding.

The field goes stale in predictable ways. It gets set at registration from a form the patient did not fill out themselves, or it defaults to English because nobody selected anything, or a family member registered the patient and answered for them. A center that has never audited this will find a meaningful share of records defaulted rather than recorded.

So the first campaign should also repair the data. When an outbound call reaches a patient whose recorded preference does not match the language they actually respond in, the correction gets written back with a case through POST /patients/{patientid}/documents/patientcase so the record improves. Run that for a quarter and the routing field becomes trustworthy, which makes every subsequent campaign better without any additional work.

The complication: real bilingual patients do not stay in one language

This is where a two-lane campaign design meets reality and needs a third answer.

A large share of patients at a health center are genuinely bilingual and switch mid-conversation. They answer the phone in English, hear something about insurance or a form, and switch to Spanish for the part that matters. Or a patient whose preference is recorded as Spanish hands the phone to an adult child who takes the rest of the call in English. Neither of these is an error, and a system that locks a call into the language it started in handles both badly.

What works is treating language as a property of the conversation rather than a property of the campaign. The call opens in the recorded preference, follows the patient if they switch, and completes in whichever language the patient is actually using. The recorded preference decides the opening, not the whole interaction.

The handoff has to account for quality honestly. Complex coverage conversations, anything involving eligibility subtleties or a sliding fee determination, and any moment where the patient sounds like they are not following, go to a bilingual staff member or the language line rather than being pushed through. That handoff should carry the language, the reason for the call, and what was already covered, so the patient does not start over. A patient repeating themselves to a third party is the specific experience language access rules exist to prevent.

Build the campaign once and let the language be a parameter

The operational trap is building a second campaign, because a second campaign decays independently.

When the Spanish version is a separate list maintained by a different person, it falls behind. The English recall criteria get updated, the Spanish one does not, and six months later they are targeting different populations for the same program. That is how a center ends up believing it has bilingual outreach while running one current campaign and one stale one.

One definition, one set of criteria, language as a parameter on the contact. Same recall logic, same cadence, same escalation, with the script and the voice resolved at call time from the patient record. When the clinical criteria for a recall change, they change once.

Then report by language cohort as the default view, not as a drill-down someone has to request. Contact rate, completion rate, and appointments booked, split by recorded language. That single reporting change is what makes the gap visible, and a gap nobody can see is a gap nobody fixes. Most centers that build this discover their two cohorts perform differently enough that they would have made different decisions months earlier had the split been on the report.

Key Takeaways

Treat outbound language as an operations problem. Inbound is usually solved because a person adapts in the moment, while outbound bakes the language in at build time.
Split every campaign report by recorded language by default. An aggregate completion rate hides a cohort that is not being reached at all.
Audit the language preference field before trusting it. Defaulted and proxy-entered values are common and they misroute the whole campaign.
Write corrections back when a call reveals the recorded preference is wrong, so the routing field improves with every campaign.
Let the conversation follow the patient’s language rather than locking to the recorded value. Genuinely bilingual patients switch mid-call, and so do handoffs to family members.
Route coverage complexity, sliding fee questions, and any sign the patient is not following to a bilingual staff member or the language line, carrying the context so nobody repeats themselves.
Maintain one campaign definition with language as a parameter. Two parallel campaigns drift apart and the second one goes stale first.

Bilingual outreach fails at health centers for an unglamorous reason: outbound campaigns are built once by someone with limited hours, and building them twice is not realistic. So the second language becomes a project that never starts, while the panel it serves quietly stops responding to recall. An AI team working the panel inside athenaOne can run one campaign definition across every language the center serves, follow the patient when they switch, repair the preference data as it goes, and hand the conversations that need a person to the staff who should be having them.

Sources