Why Bond measures before it advises

Most relationship advice is true and useless. "Communicate better" is correct. It is also impossible to act on, because it tells you nothing about what to change in your specific relationship, on a Tuesday, with this particular person.

The short answer

We built Bond around one inversion: measure before advising. It is a small change in ordering that changes almost everything downstream, including what the product can honestly claim.

Why the category gets stuck

There are two dominant shapes for a relationship app. The first is a journal: write into it, and it reflects your words back. The second is a content library: hundreds of articles, organised by topic, waiting for you to guess which one applies.

Both fail for the same reason. They do not know anything about the couple using them, so they cannot be specific, and specificity is the entire value.

Journaling apps also create a quiet problem. The more honest the writing, the more sensitive it is, and the harder it becomes to do anything useful with it without either reading it or storing it somewhere you would rather it was not.

What measuring first changes

Once you have a score, three things become possible that were not before.

You can compare against a band, not a person. A score of 42 on conflict resolution does not mean you are bad at it. It means you are in a range where couples tend to report a particular set of experiences and where particular things tend to help. That is a statement we can actually make.

You can sequence the work. The result tells you what to do first. Without a measurement, the order is guesswork, and couples usually start with the hardest thing — which is why it goes badly.

You can show change. A relationship improving is slow and invisible. A score that moves from 42 to 58 over three months is not. Making progress visible is most of what keeps a daily ritual going.

The three constraints it forced

One: nothing a user writes reaches a model. If insights are generated from derived numbers — scores, bands, counts, streaks — then the privacy claim is not a promise, it is a consequence of the architecture. We expected this to be a limitation. It turned out to be clarifying: an insight built from numbers has to be specific and useful rather than vaguely reassuring, because it cannot lean on quoting you back to yourself.

Two: individual and couple insights are different objects. An individual insight exists the moment one person finishes. A couple insight only exists when both have completed the same assessment. That ordering is enforced on the server, not in the interface, because the point is that nobody should be answering while being watched. It also means the product has to be genuinely useful to one person alone — which changed the onboarding, the dashboard and the assessment copy.

Three: safety cannot be an entitlement. Once you have a feature matrix, the tempting thing is to put everything in it. Crisis support lines and the safety note are explicitly outside it, and no entitlement check can reach them. That is a structural guarantee rather than a policy we promise to keep.

What we deliberately did not build

We did not build voice tone analysis, and we will not. Inferring emotion from a voice recording is not reliable enough to put in front of a couple in a hard conversation, and being wrong about your partner's tone in that moment is worse than saying nothing.

We did not build a "relationship health score" that tells you whether to stay. That is not a measurement problem, it is a decision that belongs to two people, and dressing up a heuristic as a verdict would be irresponsible.

And we did not build a chatbot. The value here is a specific observation and a specific next step, not an endless conversation with something that does not know you.

What is next

The honest answer is: more measurement, better guidance, and a lot of listening. If you use Bond and something is unclear, the assessments are the part we iterate on most, and we would rather hear about it than guess.