Survey length, before fieldwork

The number
is learned,
not guessed.

Upload the questionnaire. Get the length of interview in about a second — with the range, the time every question takes, and what to cut to bring it down. Trained on historical projects for genuine LOI.

No login. Word .docx. Nothing leaves your organisation.

Brand tracker questionnaire.docx Example
0.0minutes

34 questions timed  ·  15/6 largest grid  ·  2 open ends

Illustrative example. No client questionnaire, name or result is shown anywhere on this page.

Q188.5s
Q1271.3s
Q669.2s
S545.0s
Q338.5s
S130.5s
Q422.4s
Model trained on historical projects for genuine LOI Every question timed A range you can quote Length before fieldwork Model trained on historical projects for genuine LOI Every question timed A range you can quote Length before fieldwork
How it works

Three steps from the questionnaire on your desk to a length you can defend — and none of them is a survey pilot.

The wheel is one upload broken into its sections — 16.5 minutes across five — in the same colours the estimator uses on its own result screen.
Where the time goes Example · 16.5 min
16.5minutes
  • Attribute ratings6.9
  • Screening4.3
  • Open feedback2.6
  • Firmographics1.7
  • Wrap-up1.0
01 — READ

Parse the document

Every question found, classified and sized: single and multi-code, scales, rankings, open ends, and every grid down to the cell. The same parser that built the training data reads your file.

02 — TIME

Time each question

Each question is timed against what respondents actually did on completed projects — not a rule of thumb, and not a per-question guess typed into a spreadsheet.

03 — CALIBRATE

Return a range, not a number

Survey overhead, transitions and mode are added, then the total is calibrated into P25–P90 against completed fieldwork, so you get an interval you can quote.

The pipeline

What happens between the upload and the number.

Five stages, and only the last one is a model. Most of the work is reading the document accurately — get that wrong and no amount of modelling recovers it.

  1. Questionnaire
    upload
    A Word .docx, as written
  2. Parse &
    count structure
    Questions, grids, options
  3. Feature
    extraction
    The inputs the model was trained on
  4. Trained
    model
    Historical fieldwork data
  5. Prediction
    time
    A range, per section and per question
Understand the questionnaire

We don't count questions. We measure the effort behind them.

Each question is transformed into measurable signals that describe how much reading, thinking and interaction it requires.

Questionnaire
.docx
01Question detection Single Choice Multi Select Matrix Ranking Open Ended
02Feature extraction Reading Effort Options Response Complexity Interaction Question Structure
Converges into Question Time
Profile
How the model learns

The number is learned, not guessed.

Statistical modelling identifies the factors that influence respondent effort. Machine learning captures more complex relationships, while validation and calibration keep predictions grounded in observed survey behaviour.

Statistical foundation
  • Regression Analysis
  • Robust Statistics
  • Mixed-Effects Modelling
Machine learning
  • Ridge / LASSO
  • Gradient Boosting
  • Random Forest
Converges into LOI Prediction
Engine
Estimated LOI Expected Range Key Drivers
Predicted LOI
Actual Survey Behaviour
Model Calibration
Improved Prediction
Calibration feeds back into every prediction that follows.
What you get back

We will tell you our error bar. Ask anyone else for theirs.

Enough detail to defend the number in the room.

The length of interview

Example: 16.5 min

The headline figure, calibrated — not the raw model output.

The full distribution

Example: 11.8 – 41.8

P25, P50, P75 and P90, so you can set incentive and field time honestly.

Every question, timed

Example: 34

A duration for each question and each section, summing exactly to the total.

What to cut

Example: −5.2 min

Which grid, which open end, and what removing it saves — before you field it.

Health scores

3

Complexity, fatigue and mobile-friendliness, each traced to the questions driving it.

Straight from Word

.docx

The questionnaire as it already exists. No reformatting, no template.

Export the estimate

PDF · PNG · CSV

The number, the range and every question — ready to drop into a proposal or a client deck.

Cut and re-check

− +

Remove a question and watch the total fall, then put it back. Try a shorter questionnaire before you commit to one.

Your next questionnaire is already on your desk.

Upload it and see the number in about a second. Nothing to install, no account, and the file never leaves your organisation.

Model trained on historical projects for genuine LOI · estimate returned in ~1s

Let's begin →