NextFluent

How it works

Add the study. Ask it. Settle a word once. Hand your client a page.

Fieldlock is a desk for the surveys you run. Every screen below is the real desk on made-up studies: the Harborline food packing sample with two seasons, a buyer tracker with two waves, and the Kestrel rider survey of 49 questions. Nothing here is a mock-up.

1

Add the study

Drop the SPSS file. A .sav carries the question wording and the answer labels, so nothing has to be typed in. A CSV works too, with a labels file if it has none of its own. The desk reads every question once and notes its grain, its tense and its polarity. Those notes score every question you ask afterwards.

A second file is the next wave. Each wave stays its own file, with its own people and its own columns. A question that was not asked on a wave has no figure on that wave.

The Questions tab: every question in the file with its id, wording, grain, tense, polarity, column and the words it also matches
Every question in the file, as the desk reads it.

2

Ask it in English

Type the question the way your client says it on the phone. Back comes one card per wave. Where the question was asked you get the figure, the wording it was counted from and how many people it rests on. Where it was not asked the card stays blank and names the look-alike it refused to count, with the percentage you would have been handed by mistake.

Add “by region” and the figure is cut, with a letter on each column and the 95% test between them. Add “among those who changed supplier” and the base narrows to those people. Ask for the mean and a scale comes back as a mean. A ranking, a band on an amount or the drivers of a score come back as a table on the card. See them below.

Ask whether something has changed since the wave before and the change comes back as a card of its own: the points moved, and whether that is more than chance would give on these bases. Under it sit the two cards it rests on. Send both cards with the change and your client reads that line under the later card. A study that is new on the desk offers a few questions worked out from its own codebook, so you see what it reads before you type.

The Ask view: a question typed in English, a counted card for last season showing 20%, and a blank card for next season naming the look-alike that was not counted
One question, two waves. 20% where it was asked, no figure where it was not.

3

Settle a word once

People do not ask in the survey’s words. Where your word means the same as the question’s own, a built-in dictionary settles it and you get the figure: “closest dock” for “nearest dock”. Where it does not, the desk stops, names the word and the questions it could mean, and you choose. That choice is kept, so the same wording answers from then on, for you and for your client.

You decide how strict each study is, separately for your desk and for client pages. A figure counted under anything but the strictest setting says so on its receipt.

The desk on the Kestrel rider survey, a made-up sample: a question with one word from another survey question. The card says You choose, names the word and lists both questions with a Choose button each
“Concerned” belongs to another question. The desk asks which you meant.

4

Give your client the page

Each client gets a page of their own, with its own link, its own studies, its own listed addresses and its own cards. They read the cards you sent and, when you open a study to them, they type their own questions and get answers under the same rules. Every question they ask is listed on your desk before the next call.

When a client’s wording has a word nobody has settled, their page gives no figure and says you have it. The question appears on your desk with a Settle button. Choose once, and it answers on their page.

The page carries figures, wording, bases and receipts. It never carries the file, and nothing from one client’s page shows on another’s. You can close the link at any time, and a new link closes the old one.

Your side of one client page, reached from the Client pages list: the agency name and title the client sees, the open switch, the client's link, who can sign in, which studies are on the page, what the client asked, and the cards you sent
Your side of one client page.

What your client sees, and what you see.

On the left, the client’s page on a wording that is not settled: no figure, and a plain line saying their research team has it. On the right, the same question waiting at the top of your Ask screen.

A client page on the Kestrel rider survey. The client asked How jittery are riders about theft. The card is labelled Not settled and says: This wording is not settled yet, so no figure. Your research team has it: once they say which question it means, it answers here
The client’s page: no figure, and who has it.
The Ask screen of the desk with a panel at the top: 2 client questions are waiting for your choice, each with the question as the client typed it, how often it was asked, and a Settle button
Your desk: the same question, with a Settle button.

What every card carries.

A figure on its own is a claim. A figure with its wording, its base and its receipt is a finding. Every card on the desk, and every card your client sees, is built the same way.

A counted card: Last season, Counted, 20%, Unweighted, the wording Delisted because of hygiene score, field Delist3_4, n = 100, a note that the look-alike was not counted, every answer with its share, the actions Send to client page, Copy card, Copy as text, Deck line, CSV and Run again, the receipt hash and a Verify link
  1. The wave. Which file this card was counted on. Nothing is ever blended across waves.
  2. The figure. A percentage, a named box or net of boxes on a scale, a mean, the share in a band on an amount, or a place in a ranking. Weighted when the file carries a weight, with the unweighted figure beside it.
  3. The wording. The exact question the figure was counted from, with its id in the file and the number of people it rests on.
  4. What was left out. When a look-alike question sits in the file, the card says so, so nobody mistakes one for the other.
  5. The actions. Send it to a client’s page, picking the page when you have several, copy it as an image for the deck, copy it as one plain sentence, or take it as a CSV with the receipt hash in the last row.
  6. The change. When you asked whether something changed, a card above the pair carries the points moved and whether that is more than chance on these bases. Sent with the pair, that line sits under the later card on the client’s page.
  7. The receipt. A SHA-256 hash over the study, the wave, the question, the field, the base and every row. Run it again in a year and the hash matches.
  1. No figure. The question you asked was not put to people on this wave, so there is nothing honest to count.
  2. Not counted. The look-alike that is in the file, named, with the percentage a chat on the file would have given you. Here, 25%.
  3. Send the blank. A blank is a finding too. Your client sees that the question was not asked, instead of a number from the wrong question.
  4. The receipt. A refusal gets a hash as well, so the blank can be verified like any figure.
A refused card: Next season, No figure, That question is not in this file, Not counted 25% Would delist because of hygiene score 25 of 100, the actions and the receipt hash

Deeper than a share.

Most questions end at a percentage. The desk also counts a ranking, a band on an amount and the drivers of a score, on the rows of the file, with a receipt that carries the specification. The screens below are the real desk on the second made-up study, a Harborline buyer tracker with two waves.

Rankings, cut by region

Ask “which brand is in the top 3” on a ranking question and a table comes back for every brand on the list: first choice, in the top three, mean rank, each with the people behind it. Add “by region” and the headline figure is cut, with a letter on each column and the 95% test between them. A weight in the file is applied and named.

Two cards, Wave 1 and Wave 2, for Which brand is in the top 3 preferred by region: 77.7% and 79.8% for Harborline, a table of first choice, in the top 3 and mean rank for five brands, and a cut by region with column letters and the 95% test
Every brand’s place on the list, and Harborline’s figure cut by region, on both waves.

What drives the score

Ask “what drives overall satisfaction” and the desk pairs the overall scale with every rated item in the study. Each row is the correlation on the people who answered both, with its share of the explained variation and a tick where it holds at the 95% level. The headline is the top driver and its share. The receipt carries the specification, so the table can be run again and checked.

Segmentation and conjoint stay in your stats package. Write the result back into the file as a column and the desk cuts by it like any other question.

Two cards for What drives overall satisfaction: 74.8% on Wave 1 and 81.3% on Wave 2, Share of importance Print quality r 0.65, and a drivers table with r, share of importance, top 2 and pairs for six rated items
Print quality carries most of the explained variation on both waves: 74.8%, then 81.3%.

Bands, and the table across waves

Ask “how many spend over €50 a week” on an amount and the figure is the share in that band, with how many gave a number. With two waves the line under the cards says whether the move is a real change at the 95% level. The table across waves lists every figure, base and change, and the CSV beside it carries each wave with its receipt hash. Ask it as a change, whether the share has gone up since the wave before, and the change leads as a card of its own, ready to send with the pair.

Two cards for How many spend over €50 a week: 62.6% on Wave 1 and 74.3% on Wave 2, a line reading +11.7 points, a real change at the 95% level, and a table across the two waves with figure, people, change and the test, with a CSV, every wave button
62.6% to 74.3%, a real change at 95%, with the CSV of every wave one click away.

Waves stay apart.

The Waves tab shows what changed between two files: which questions were asked on both, which were dropped, which are new. When you ask, each wave answers for itself. With two waves the card says whether the move between them is a real change at the 95% level or within what chance would give. With three or more it draws the line.

The Debrief tab does this for the whole study at once. When a wave lands, every question asked on both waves is compared and the moves beyond chance are listed largest first, each a card with its two receipts. Under each move is what the file can honestly say about why: the move cut by every classification in the file, with a test on whether any group is apart from the rest; how much of it is change inside the groups and how much a changed mix of them; whether the base changed through routing; which rated items moved with it. Where the file holds nothing, the card says so, and the question goes on a list for the next questionnaire. The debrief sits at the top of the topline, goes into the export, and can be put on the client’s page, held until you show it.

Two questions that read alike are listed as look-alikes, with the cue words that tell them apart. When an ask lands too close between two of them, the desk does not pick. It shows both and asks you to pin one. From then on that wording binds to the question you chose.

The Waves tab: from Last season to Next season, 3 asked on both, 4 not asked on Next season, 4 new on Next season, and under it the questions a client asked that the study cannot answer
What changed from one season to the next.

What the desk will not do.

Guess the question

When a question names something the study never asked about, the answer is no figure, with the reason. When two questions score too close together, or a word in your question is not one question’s own, you are asked to choose. A nearest column is never quietly chosen.

Blend waves or studies

One figure comes from one file. Ask across every study on your login and each study answers on its own tile. One wave’s percentage is never written onto another.

Let a language model write the number

Every count is arithmetic on the rows of the file. No language model sits on the number path, and every receipt says so: model=none. Your response file is never sent to ChatGPT or to any other model.

Try the sample, then bring a file.

The Harborline sample runs on the homepage without a login. Your own file goes on the desk: book a call and we open one with your file on it. The desk is where your real studies live.