Explore the beta

Research

Can personally relevant guidance help students choose food that fits?

This page keeps six things apart, in order: what a first survey observed, what that suggests but does not prove, the health-centered hypothesis, what the Spring 2027 beta will measure, how we would decide to continue or stop, and the methods and limits behind all of it.

  1. Observed
  2. Suggests, not proves
  3. Hypothesis
  4. Beta measures
  5. Criteria
  6. Methods and limits

1 What the survey actually observed

Evidence

Friction, taste, nutrition and a need for speed

Shares of all 158 respondents, in the survey’s own wording where possible. Respondents could select more than one factor, feature or barrier.

How deciding goes

Reported difficulty deciding what to eat at least sometimes58.9%93 of 158
Repeatedly eat the same foods because they’re easiest52.5%83 of 158
Have trouble building a balanced or satisfying meal48.7%77 of 158
Spent four minutes or more deciding36.1%57 of 158

Factors that matter when choosing

Taste / what I’m in the mood for91.8%145 of 158
Nutrition or a balanced meal78.5%124 of 158
Convenience / wait time44.9%71 of 158
Allergies or dietary restrictions15.2%24 of 158
Religious, cultural or ethical food practices12.0%19 of 158

Features selected as useful

Seeing all current campus dining options in one place57.6%91 of 158
Recommendations based on foods I like54.4%86 of 158

What would stop them using a tool like this

Takes too much time64.6%102 of 158
Don’t want to log what I eat38.6%61 of 158
Don’t want another app33.5%53 of 158
Don’t want an app telling me what I “should” eat18.4%29 of 158
82.3%expressed high interest in trying the concept once (4 or 5 out of 5)130 of 158. Interest in trying once. Not adoption, repeat use or demand.
43.0%would most likely use a tool like this before leaving for a meal68 of 158.

About this evidence. These figures come from one survey of 158 university students, collected September 19–29, 2026. It was a convenience sample, and the answers are stated behavior and interest. They do not show that students want health guidance from an app, that Daki improves nutrition or health, that anyone would use it repeatedly, or that institutions would act on what it shows.

2 What this suggests, but does not prove

Enough to justify a test. Not enough to claim impact.

The survey found decision friction, demand for clearer options, interest in preference-based relevance, and a need for speed. These findings support testing a personalized decision tool. They do not yet validate Daki’s health impact.

Reasonable reading

What it suggests

  • Decision difficulty is common but not universal (58.9% at least sometimes; 17.1% often or almost every time; 30.4% said campus dining generally works well for them).
  • Nutrition matters to most respondents, but alongside taste, which matters to even more. Guidance that ignores enjoyment is unlikely to fit.
  • Personal relevance is valued: recommendations based on foods people like were selected as useful by more than half.
  • Speed and low effort are requirements. Logging, extra apps and being told what to eat are all deterrents.
Not shown

What it does not show

  • That students want health guidance from an app. Saying nutrition matters, or selecting a feature as useful, is not the same as wanting or using it.
  • That Daki improves nutrition or health, or that personalized suggestions change behavior.
  • That students will use Daki repeatedly. Trial interest is interest in trying once.
  • That institutions will act on Daki’s patterns, or that waste would fall.
  • Any endorsement by, or agreement with, Harvard or any other university.

Subgroups differ. Among the 90 Harvard-related respondents, 48.9% reported difficulty deciding at least sometimes, compared with 72.1% of the 68 from other universities. In a convenience sample this difference cannot be generalized, but it is a reminder that need varies by setting and that behavior has to be tested in context.

3 The health-centered product hypothesis

Hypothesis

From mission to what is still unproven

Health organizes the story. It is not a result. Each layer below needs the one before it, and only the first two rest on anything observed.

LayerStatementWhere it comes fromStatus
MissionHelp people make food choices that support their health.The purposeStated
Current evidenceStudents report decision friction and interest in clearer, personalized, low-friction help.Survey, n = 158Observed (stated)
Product hypothesisPersonally relevant, explained guidance can improve confidence and perceived fit.The betaTo be tested
Impact hypothesisRepeated use may help people build food intuition and healthier patterns over time.Longer studiesNot yet testable
Unproven outcomeMeasurable health improvement.Not measuredNot demonstrated; not claimed

The model behind the hypothesis: understand the person, interpret the environment, support a confident choice, and, with voluntary aggregation, help institutions improve the environment. See the four steps

4 What the Harvard-specific beta will measure

Planned test

Eight questions for Spring 2027

Proposed measures, to be finalized before any data is collected. The beta will not measure disease prevention or clinical health change.

QuestionWhat we want to knowPossible measure
Health fitDo students identify options they believe fit their health goals, tastes, cultures, dietary needs and circumstances?Self-rated fit with the goals and needs each student stated (not a clinical measure)
ConfidenceDoes Daki increase confidence in a meal decision?Short rating at the moment of choice, compared with browsing alone
UnderstandingDo students understand why an option was suggested?Whether the stated reason makes sense, plus coded reasons for passing
DiscoveryDo students find suitable options they would otherwise have missed?Self-report of options that were new to them
CapabilityDoes use over time help students feel more able to navigate the food environment?Brief check-ins at the start and end of the beta
ReturnDo students come back across multiple meal decisions?Anonymous repeat use within 7 and 28 days
Aggregate patternsWhat needs, mismatches or barriers appear in aggregate?Coded, voluntary signals; groups under 10 suppressed
Institutional usefulnessDo institutional stakeholders find those patterns understandable, new and potentially actionable?Structured interviews with dining and wellness staff

Independent. Daki is not affiliated with, endorsed by, or partnered with Harvard University or Harvard University Dining Services. The beta’s use of published Harvard dining information is not a university integration or an institutional partnership. The beta is a research and learning effort, not evidence of institutional or health impact.

5 Success, failure and stop criteria

How we would know to continue, adjust or stop

Draft criteria. Exact thresholds will be written down before any data is collected, so they cannot be moved afterwards.

Success would look like

  • Most participants rate suggested options as a good fit for the goals and needs they stated.
  • Confidence in meal decisions rises compared with browsing alone.
  • A meaningful share return without prompting over several weeks.
  • Dining or wellness staff say at least one aggregate pattern is new and worth discussing.

Failure would look like

  • Suggestions feel generic, or reasons do not make sense to students.
  • Deciding takes as long as browsing a menu.
  • Most people use it once and do not return.
  • Staff find the patterns redundant or misleading.

Stop or adjust immediately if

  • Menu, allergen or access information shown is regularly wrong.
  • Anyone relies on Daki for allergen safety or medical decisions.
  • Students report feeling judged, pressured or worse about food.
  • Aggregation could reveal an individual’s choices, health details or schedule.

6 Methodology and limitations

How the evidence was gathered, and what it cannot carry

Method

  • An online survey, shared through convenience channels, September 19–29, 2026; 158 responses.
  • Percentages use all 158 responses. Multi-select questions can sum to more than 100%.
  • Harvard-related means the respondent named Harvard as their university; there are no other subgroup claims.
  • Raw responses are not published, and no combination of answers that could identify a respondent is shown.
Limits

Limitations

  • Not representative. Convenience channels; nearly all meal-plan users.
  • Self-reported. What people say, not what they do.
  • Interest, not adoption. Willingness to try once does not predict repeat use.
  • No health measures. Nothing here measures nutrition, health, waste or cost.
  • Students only. Dining professionals have not yet been surveyed.

Students

The early prototype

Built on a dated snapshot of published HUDS information. It covers parts of steps 2 and 3 of the model, not health-goal guidance.

Try the diner experience

Dining teams

The dining-team concept

Simulated data only, to test whether aggregate patterns are understandable and what would make them misleading.

View the dining-team concept

Take part in the research

Students can join the waitlist to take part in the Spring 2027 beta. Dining, wellness and public-health professionals and researchers who want to help design or evaluate it are warmly invited to get in touch.

The waitlist is a short Google Form. Joining does not give immediate access; it helps Daki recruit students for the independent Spring 2027 beta and follow-up research.

Lead dining, student life, wellness or public health?

We would like to learn what would be useful to your institution, and to talk about shaping a potential Spring 2027 pilot, before anything is built for you.

Or write to maverickyasuda@college.harvard.edu