Surveys that branch like a real conversation

A good conversation on a doorstep never runs down a fixed list of questions. It follows the person. Say you support the campaign and I'll ask if you'll volunteer. Say you don't and I'll ask why, gently. A survey that can't do that – that marches every person through the same rigid script – isn't listening, and people can tell.

Uprise surveys branch. Here's what that gives you.

Skip logic and terminal branches

Every answer can change what comes next. With per-option skip logic, each choice a respondent makes can route them to a different question – so the survey adapts to the actual conversation instead of forcing the conversation to fit the survey.

And sometimes the right move is to stop. Terminal branches let an answer end the survey cleanly. If someone tells you they're moving out of the electorate next week, you don't need to walk them through five more questions – the branch closes the conversation on a natural note and moves on.

The effect is a survey that feels like it's paying attention, because it is.

Branch where the conversation actually forks

Branching is easy to overdo. A survey with a branch on every question becomes impossible to reason about and produces data nobody can compare.

The useful test is whether the next question genuinely changes. Usually there are only two or three real forks in a doorstep conversation:

• A supporter gets asked to do something – volunteer, take a sign, come to an event.
• An opponent gets asked why, once, and then you stop.
• An undecided gets the issue question, because that is the one that tells you what would move them.

Everything else is the same question for everyone, and should stay that way.

Four question types, one builder

Real conversations aren't all multiple choice, so the builder gives you the types you actually need:

• Single choice for a clean either/or.
• Multi choice when more than one answer is true.
• Yes/no for the quick reads that keep momentum up.
• Text for the things you'd never predict – the comment that tells you what the electorate is really thinking.

Mix them freely inside one survey. A yes/no can open the door, a single-choice can sort the response, and a text field can catch the detail that turns into your next talking point.

One text field, and no more

A word of warning on the text type, because it is the one that gets abused. Free text is the only way to capture the thing you could not have predicted, and it is also the slowest thing to record on a doorstep and the hardest thing to analyse afterwards.

One per survey. Put it at the end, make it optional, and treat what comes back as a source of talking points rather than as data you will count. If you find yourself adding a second free-text field, what you actually want is a single-choice question and you have not yet worked out the options.

The same survey on the doors and over text

Here's what ties it together: the survey you build works both on the doors and over SMS. One survey, two channels.

That matters because your program isn't one channel. A canvasser runs the survey face to face on a Saturday; the same instrument goes out over text to the people you couldn't reach in person. You're not maintaining two versions that slowly drift apart, and you're not comparing results from two different questionnaires and hoping they line up. It's the same questions, the same branches, the same data – however the conversation happened.

Wired to canned replies and dispositions

A survey answer shouldn't just sit in a database. It should do something. So each option can be wired to the two things that make outreach move:

• An automatic canned reply. Choose an option and the matching response can fire back on its own – the right follow-up text, sent the instant someone answers, without an organiser typing a word.
• A disposition. Each option can map to a disposition, so the answer feeds straight into how you're recording and scoring that contact.

Set it up once and the survey stops being a form and starts being a machine. Someone taps an answer over SMS, the right reply goes out, and the contact is scored – all from the same choice.

Test it on ten people before two thousand

Field the survey with your own team first – an hour, ten conversations, real ones. You are looking for three specific failures:

• A branch that goes somewhere silly. Easy to miss on screen, obvious the first time a person walks it.
• A question people ask you to explain. If a volunteer has to gloss it, the wording is wrong.
• An option nobody ever picks. Either it is redundant or it is phrased in a way nobody recognises themselves in.

Fixing those three costs an hour before the program starts. Finding them in week three costs you every response gathered until then, because changing the instrument mid-program makes the data either side of the change incomparable. If you must change it, note the date and treat it as a new series.

Build once, learn everywhere

The reason branching surveys are worth the care is that they're where your program learns. Every door and every text is a chance to find out where a person actually stands – but only if the instrument is good enough to follow the conversation and disciplined enough to record it the same way every time.

Build the survey once. Run it on the doors and over text. Let each answer trigger its own reply and its own disposition. What you get back isn't a pile of raw responses – it's clean, consistent intelligence about your electorate, gathered the same way no matter how you reached the person.

Stop making people answer questions in an order that suits your form. Build the survey that follows the conversation instead.

Read more

• Building a branching survey – the full build, from the decision backwards.
• Reading your results – how dispositions roll onto the five-point scale.
• Running a doorknock weekend – fielding the survey with a real team.
• Launching an SMS program – the same instrument, over text.