Skip to content
Product
PlatformPricingDimensionsIntegrationsSecurityCompare
Solutions · by use case
Customer, guest and fan experienceProduct feedback and prioritizationRevenue, churn and retentionAgenciesContact center performanceAI agent and chatbot analyticsMulti-location performanceReviews and competitive benchmarking
Solutions · by industry
Hotels and hospitalityRestaurants and food serviceFitness and leisureRetail and e-commerceSports and live entertainmentInsuranceBanking and financial servicesSoftware and SaaSAutomotiveConsumer products and CPGContact centers and BPOsPublic sector and nonprofitReal estateGaming and betting
Learn
Artifacts GalleryCustomer storiesResearchBlogThought leadershipLessons Log in Start now
Blog

How Health Insurance Exchanges Analyze Enrollment Conversations

Six labels on every enrollment conversation, read by intent and language each day, show which questions fill the agent queue. A synthetic worked example and what Covered California changed.

A health insurance exchange analyzes enrollment conversations by giving every chatbot session, call transcript and email the same labels: the stage of enrollment the person reached, what they came to do, whether the bot finished it, what sent it to a person, where an unfinished session stopped, and the language it was in. Read by intent and language every day of open enrollment, those labels show which questions fill the agent queue and where Spanish-speaking applicants get stuck. Dimension Labs is built for this job: it reads every channel an exchange runs, labels each conversation with Dimensions written for enrollment, and turns the result into the daily read and the reports leadership asks for. Covered California uses it this way on its chatbot.

Below: why the 2027 season makes this read matter more, the six labels to use, a worked example on synthetic data, and what Covered California changed once every session carried them. For the federal-agency version, built on trust and effort surveys rather than enrollment, see how government agencies measure trust and effort.

Why do enrollment conversations matter more in the 2027 season?

Because the rules on financial help changed, and people ask about rules they do not understand. Enhanced federal premium tax credits expired, and Covered California puts the federal help Californians lost at nearly $2.5 billion a year. The state raised its own subsidy funding from $190 million to $300 million, and about 200,000 people are newly eligible for help. Every one of those changes is a question an applicant types into a chat window.

The season is also short. In California it runs Nov. 1, 2026 to Jan. 31, 2027; on HealthCare.gov it ends Jan. 15, 2027. A flow that fails in week one and gets fixed in week ten has already failed most of the people who needed it.

Public facts, read Nov 17, 2026

For 2027, the subsidy rules changed, and every change is a new question in the queue

What Covered California and CMS published at the start of open enrollment, and the question each change sends to the chatbot and the service center.

What changedThe question it creates
Open enrollment runs Nov. 1, 2026 to Jan. 31, 2027 in California; HealthCare.gov runs Nov. 1, 2026 to Jan. 15, 2027Am I too late? Which date applies to me?
Enhanced federal premium tax credits expired; Californians lose nearly $2.5 billion a year in federal helpWhy did my premium go up? Do I still get help?
State subsidy funding rises from $190 million to $300 million, for incomes up to 200% of the poverty levelDo I qualify for the state subsidy?
About 200,000 Californians are newly eligible for financial helpI was told no last year. What about now?
The service center is open Monday to Friday, 8 a.m. to 6 p.m., and closed on weekendsWho answers on Saturday?
Sources: Covered California open enrollment 2027 release (Oct. 30, 2026) and CMS plan year 2027 fact sheet, both read Nov 17, 2026. The right-hand column is our reading of what applicants will ask.

What should an exchange label in every enrollment conversation?

Six things, on every session, in every language. Five are the labels Covered California's story names; the sixth, language, is how the same story compares English and Spanish flows. Together they cover what an enrollment team needs to decide what to fix.

The schema

Six labels on every session: the five Covered California's story lists, plus language

What each label records on one enrollment conversation. Example values are illustrative.

LabelWhat it recordsExample values
Journey stageWhere the person is in enrollmentScreening, income, plan choice, documents, payment
IntentWhat they came to doRenew, ask about financial help, report income, compare plans
Escalation triggerWhy the session went to a person, if it didAsked for an agent, no answer, eligibility edge case
Containment outcomeWhether the bot finished itContained, handed off, abandoned
Drop-off locationThe step where an unfinished session stoppedIncome question 3, document upload
LanguageThe language the person wrote inEnglish, Spanish
Labels from the Covered California customer story: "journey stage, intent, escalation triggers, containment outcome, and drop-off location", across English and Spanish. Example values written for this post.

Three choices make the labels useful:

  • Fix the values before the season starts. If "financial help" is split three ways in December and two ways in January, the weeks cannot be compared. Write the intent list and the stage list once, with a definition for each value.

  • Keep drop-off and escalation apart. A person who leaves at the income question and a person handed to an agent at the income question need different fixes: one needs a clearer question, the other needs an answer the bot does not have yet.

  • Label calls and email the same way. Many people who give up in chat call the service center next. If the call transcript carries the same intent and stage, the two contacts can be read as one attempt; following one customer across channels shows how to join them.

What does a first read of two weeks of enrollment chats look like? A synthetic example

We made up a sample to show the shape of the read: 6,000 chatbot sessions at a fictional state exchange in the first two weeks of open enrollment, 4,140 in English and 1,860 in Spanish. Nothing in it comes from Covered California or any real exchange. Overall, the bot finished 66.2% of English sessions and 54.6% of Spanish ones. The total hides where the gap sits.

Synthetic sample

In the sample, Spanish sessions about financial help are contained 25 points less often than English ones

Share of chatbot sessions the bot finished without a person, by intent and language. A fictional exchange, first two weeks of open enrollment.

EnglishSpanish

Renew my current plan907 English, 401 Spanish
86.1%82.5%
Payment and first premium359 English, 153 Spanish
80.5%73.9%
Compare plans and doctors599 English, 313 Spanish
72.1%61.7%
Upload a verification document430 English, 206 Spanish
61.6%50.5%
Will I get financial help?1,044 English, 437 Spanish
57.2%31.8%
Report an income change501 English, 231 Spanish
47.1%31.2%
Medicaid or other program referral300 English, 119 Spanish
46.7%53.8%
Synthetic data written for this post, not data from Covered California or any exchange: 6,000 sessions, 4,140 in English and 1,860 in Spanish.

Split by intent, renewals and payments work in both languages. The gap is in financial help: 57.2% contained in English against 31.8% in Spanish, a difference of 25.4 points. Questions about financial help and income changes are 36.9% of sessions but 52.1% of the sessions the bot did not finish. In this sample, the first fix would be the Spanish financial-help flow, and the escalation-trigger label says what kind of fix: missing content, a question the bot misreads, or a case that needs a person by rule.

One bar goes the other way: program referrals are contained more often in Spanish, on only 119 Spanish sessions. At that size a gap of a few points can be chance, which is why the read shows the count beside every rate. It also tells the team where more data will settle the question within days, since the labels keep running on every new session.

How does an enrollment team turn the labels into changes during open enrollment?

  1. Read containment by intent and language every day. A live integration is enriched every day, so yesterday's sessions carry their labels the next morning.

  2. Rank intents by sessions sent to a person, not by volume. The busiest intent is often renewals, which the bot already handles.

  3. Read the escalation triggers inside the top intent. Sort the fallback questions into gaps you can fill; how to find the questions your chatbot can't answer works through it.

  4. Change one flow and compare before and after. Check the change against the days before it, by language, with the method for checking whether a fix moved the number.

  5. Plan around the deadlines. In California, a plan chosen by Dec. 31 covers all of 2027, so the weeks before each deadline are when a broken flow costs the most.

Remove names, case numbers and contact details from transcripts before analysis; how to remove PII from chat transcripts covers the passes to run. Our security practices are on the security page.

What changed at Covered California once every session carried these labels?

Before, Covered California had tens of thousands of chatbot conversations as raw transcripts, and friction in the enrollment journey was often found only after escalations happened. With every session labeled, the team maps journeys across chatbot flows, finds the drivers of drop-off and escalation, and tunes English and Spanish flows independently. Its CX Manager put it this way: "You're going in blind without Dimension Labs."

Customer story

What Covered California reports after labeling every chatbot session

Results as published in the customer story.

2 weeksto launch a new self-service use case
25%fewer resources needed for implementation
Nearly 100%containment rate for key intents
5×increase in chatbot self-service adoption
Source: Covered California customer story, read Nov 17, 2026.

What does Dimension Labs give an exchange's enrollment team?

Dimension Labs is an omni-channel analytics agent for customer data. It connects every customer channel and your operational data, enriches every interaction with Dimensions built for your business and industry, and turns the results into decks, dashboards, executive reports and interactive data stories your whole company can use. Plans start at $299 a month with no demo.

For an exchange in open enrollment, that comes down to four things:

  • Fully omni-channel: chatbot sessions, call transcripts, email and survey comments are read on one schema, in English and Spanish, so a person who gives up in chat and calls the next day is one story, not two.

  • Easy integration across channels and operational data: contact center and chatbot platforms connect directly, and application status or region can be added as business data. Joining labels to enrollment outcomes and measuring effects with confidence intervals is on the Prove plan ($2,999 a month) and up.

  • Effortless enrichment of every interaction: the six labels above, plus any your program needs, written for health coverage enrollment and usable by the service center, the bot team and policy staff alike. Custom Dimensions start on the Discover plan ($999 a month).

  • A new visual analytics layer: the agent builds the daily read as a dashboard, the weekly summary as a deck or PDF for leadership, and can rerun it on a schedule, with results saved to the project's Final Reports folder and shared by links you can revoke.

When something simpler is the better choice: if the exchange runs only a chatbot and needs containment for a handful of intents in English, the bot platform's own reporting may be enough. The case for Dimension Labs grows with every channel and language you add, because the labels stay the same across all of them.

FAQ

How can a health insurance exchange analyze enrollment chat and call conversations?

Label every chatbot session and call transcript with journey stage, intent, escalation trigger, containment outcome, drop-off location and language, then read containment and escalations by intent and language each day. Fix the intents that send the most people to agents first.

What is the best tool to analyze an exchange's enrollment conversations?

For an exchange with chat, calls and email in more than one language that needs the same labels on all of them and reports for several teams, Dimension Labs. For a single English chatbot with a few intents, the bot platform's built-in reports can be enough.

Should Spanish conversations be translated before analysis?

No. Read them in Spanish, apply the same labels, and keep language as its own field so the two can be compared side by side. How to analyze reviews in several languages explains why translating first loses information.

How fast can an exchange launch and measure a new self-service flow?

Covered California reports launching a new self-service use case in 2 weeks, with a 25% reduction in resources required for implementation.

Which enrollment questions should a chatbot hand to a person?

The ones where a wrong answer costs coverage: unusual eligibility cases, income changes that affect financial help, and program referrals. Label those as escalations a rule asked for, so they are not counted as bot failures; how to measure why customers escalate sets out the split.

Export a week of chatbot sessions or call transcripts with a language column and start with one file: the agent labels every session and shows where each language gets stuck. Every plan from Explore to Scale includes a 14-day free trial.

Dimension Labs/Start with one file From $299 a month

Stop reporting. Start understanding.

Start with one file. Ask why a number moved. Plans from $299 a month, no demo and no contract on monthly plans.