How Government Agencies Measure Trust, Effort and Drop-Off From Feedback
Agencies have no churn rate, but they have trust, effort and abandonment. Here is how to read open comments on the federal survey drivers and find where applicants give up.
A government agency with no churn metric can measure customer experience from feedback by treating three things as its outcomes: whether the interaction raised or lowered trust, how much effort it took, and where people gave up before finishing. Label every open comment with the same drivers the federal post-transaction survey asks about (effectiveness, ease, efficiency, transparency, humanity and fairness, employee interaction), add the form step or service stage the comment is about, and compare trust ratings and drop-off across those labels. For an agency or public program that needs this across surveys, chatbot and call transcripts and email, in English and Spanish, and reported to leadership on a schedule, Dimension Labs is built for the job, and Covered California uses it to find where applicants drop off.
Federal facts below are from official pages read on Nov 7, 2026. The worked example uses 1,500 survey responses we made up for a fictional online benefits application.
How can a federal agency analyze customer feedback when there is no churn metric?
Use the outcome the government already asks for. Executive Order 14058 directs executive branch agencies to manage customer experience and improve service delivery, under the title "Transforming customer experience and service delivery to rebuild trust in government." It also asks agencies to cut the administrative hurdles and paperwork burdens it calls "time taxes." Trust and effort play the part that retention plays in a business.
The measurement is already in place for the largest services. Under OMB Circular A-11 Section 280, each designated high-impact service provider (HISP) must run at least one customer feedback survey for each designated service. The trust question is required, in a form such as "This interaction increased my trust in [HISP name]." On a 1 to 5 scale, trust is the share of responses rated 4 or 5, and the target is 75% or higher. Each HISP submits at least one dataset of responses every quarter for each designated service.
The survey also asks what made the difference, with a list of drivers, and digital.gov lists open-ended prompts such as "What could have been better?" and "Anything else you want us to know about your experience?" The checkboxes tell you which driver; the comments tell you what, specifically, went wrong with it. That second part is where most programs stop, because reading thousands of comments a quarter by hand is slow.
How do you turn open comments into trust and effort measures?
Label every comment with the same six drivers, so the open text and the checkboxes can be compared on the same terms, then add the detail the checkboxes cannot hold: the step, the time mentioned, and what the person was not told.
Mapping the federal survey drivers to comment labels
Label the open comments with the same drivers the survey already asks about
The six drivers in the federal post-transaction survey, and the Dimensions that make the comments countable on the same terms.
| Survey driver | What the comment label captures | Extra Dimension worth adding |
|---|---|---|
| Effectiveness | Did the person get what they came for | Outcome: completed, partly, not at all |
| Ease | Effort: confusing questions, repeated steps, unclear words | Form step where the confusion happened |
| Efficiency | Time: waits, delays, timeouts | Time mentioned (minutes, days, weeks) |
| Transparency | Not knowing status, rules or next steps | What they were not told |
| Humanity and fairness | Feeling treated with respect, or not | Who or what the comment is about |
| Employee interaction | Help from staff, by phone, chat or in person | Channel of the human contact |
With those labels on every response, trust can be broken down by what went wrong. In the sample, responses whose comment named no problem rated trust 4 or 5 in 92.2% of cases. Responses that named a problem with effectiveness, meaning the person did not get what they came for, did so in 19.1%. Overall trust in the sample is 55.6%, below the 75% target.
Sample data, made up for this post
Responses that named a problem with getting the job done had the lowest trust
Share of responses rating trust 4 or 5 out of 5, by the driver their comment named as a problem. Dashed line: the 75% target.
75% target
The sample is invented, so read the bars as the shape of the report, not a finding. In real data the drivers overlap, and a comparison like this does not prove which problem lowered trust. To rank them by effect, with the others held fixed, the analysis has to control for the overlap; key driver analysis from text walks through how.
How do you measure where applicants get confused and abandon an online form?
Feedback gives you the why; your own systems give you the where. Start with the application or session data: which step each person reached and whether they submitted. Then label each comment with the step it is about and whether the person says they gave up, left to call, or will try another way. The two together show where people stop and what they say stopped them.
Counting complaints by step is not enough. In the sample, eligibility questions draw the most comments, 275, but only 16.0% of those say the person gave up. Identity verification and document upload draw fewer comments, and 29.2% and 30.4% of them describe giving up.
Sample data, made up for this post
The step people complain about most is not the step where they give up most
Five steps of a made-up online application. Left: responses whose comment names the step. Right: share of those that say they gave up, left or would try another way.
Two cautions apply to any version of this. People who quit a form rarely fill in the survey at the end of it, so post-transaction surveys undercount the drop-off; chatbot and call transcripts from people stuck mid-application fill much of that gap. And keep the languages separate in the report, since a step can work in English and fail in Spanish. Our posts on why people escalate from a chatbot and on containment, resolution and deflection rates cover the chatbot side.
What does this look like in a public program?
Covered California, the state's health insurance marketplace, handles enrollment questions across chatbot, live agent and email channels in English and Spanish, with multi-step eligibility and verification. Before Dimension Labs, it found friction points "only after escalations occurred," and had limited real-time visibility into drop-offs.
Every chatbot session is now enriched with journey stage, intent, escalation triggers, containment outcome and drop-off location. Covered California reports a new self-service use case launched in 2 weeks, a 25% reduction in resources required for implementation, nearly 100% containment for key intents and a 5× increase in chatbot self-service adoption. Its CX manager's summary: “You’re going in blind without Dimension Labs.”
Why would a public agency use Dimension Labs for trust and effort analysis?
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 a program office, the four parts are concrete. Post-transaction surveys, chatbot sessions, call transcripts and email are read on one schema, with the language kept on every record so English and Spanish can be compared. Application status, service, region and channel data are joined to the feedback, so a comment about document upload sits next to whether that person submitted. The Dimensions follow your own service and the federal drivers rather than retail categories; custom Dimensions start on the Discover plan. And the visual layer produces the readouts a program needs, a quarterly trust summary for leadership, a step-by-step drop-off view for the digital team, a report for each region, without anyone building them by hand. The join to operational data with effects and confidence intervals is on the Prove plan and up. Security controls, including SOC 2 Type II, ISO 27001 and PII redaction at ingest, are on the security page, and the public sector and nonprofit page covers the setup.
When something simpler is the better choice: a single service with a few hundred comments a quarter, where the survey platform's own reporting and a person reading the comments will cover it.
FAQ
How can a federal agency analyze customer feedback when there's no churn metric?
Use trust, effort and drop-off as the outcomes. The A-11 Section 280 survey already measures trust and asks which drivers made the difference; label the open comments with the same drivers, plus the step and what went wrong, and compare trust across them.
What is the federal trust target for post-transaction surveys?
Performance.gov gives a target of 75% or higher, with trust measured on a 1 to 5 scale as the share of responses rated 4 or 5, or with a thumbs up or down.
How do I find where applicants abandon an online government form?
Take the step each person reached from your application data, label comments and chatbot or call transcripts with the step they mention and whether the person gave up, and rank steps by the share who gave up, not by the number of complaints.
What is the best way to analyze government customer feedback at scale?
For surveys, chat and call transcripts, joined to application or case data and reported on a schedule, Dimension Labs. For one small service, the survey tool's own reports and a careful hand read of the comments.
To try the labels on your own survey comments or chatbot transcripts, start with one file.