Surveys that run in your own cloud
We hate data.
Specifically, we hate having to hold yours. Your team has already done the hard work of deciding where sensitive data can live and who can access it. Survey Builder works within the cloud environment you already govern. Your team writes and approves the questions in Survey Builder. When someone answers on your site, the response goes straight to your warehouse, and we have no access to it.
Nearly every survey tool is built the same way. You put the vendor’s script on your site. The script comes from the vendor’s servers, and each answer goes back to the vendor’s servers, where it waits in the vendor’s database until you log in to read it.
For most businesses that is a fair bargain. For a pharmaceutical brand it is the start of a long conversation. A patient may type a diagnosis, a side effect, an email address. Now a new company holds those words, and your privacy, security and regulatory people must each be satisfied about that company before a single question is asked. It is not unusual for the paperwork to outlast the campaign.
Survey Builder is arranged the other way round. Everything that touches a respondent runs in a cloud account that belongs to you. We keep only the part that touches nobody: the place where your team writes the questions and your reviewers approve them.
There is no magic in this. It is one decision about where things run, carried through in detail. The details follow.
Thirteen particulars
-
The visitor’s browser speaks only to you.
The survey script and its wording are fetched from storage in your own cloud account. Answers are sent to a collector in the same account. Open the network panel of any browser while the survey is on screen. Our name is not in it.
-
We hold no key to your cloud.
No service account, no password, no standing access. When you publish a survey, we knock. Your cloud checks our signature, then comes and collects the new wording itself. Nothing can be pushed in.
-
We can send you words, never software.
The script that runs on your site is part of what your own administrator installs. What we publish afterwards is a description of a survey: questions, options, rules. Your cloud refuses anything not signed with the key it was given on the day it was installed.
-
Names are kept apart from answers.
Ratings and choices go to one dataset. Names and email addresses go to another, with its own list of readers. An analyst can chart every answer without being able to see who gave it. You write both lists. We are on neither.
-
What is not collected cannot leak.
The collector records no IP address and no browser string. It cuts the query string off every page address, because that is where email addresses and click identifiers hide. It sets no cookie; it tells one visit from another by a random number kept in the browser, which identifies nobody.
-
Every version is frozen and fingerprinted.
Publishing stores the survey exactly as written and seals it with a SHA-256 fingerprint. Each answer is stamped with the version that was on screen when it was given. A year from now you can say precisely which words a respondent read.
-
Reviewers approve changes, not screenshots.
Each version has a plain record of every word a respondent will see, in order, with the skip rules written out as sentences. Between any two versions there is a redline. Your medical, legal and regulatory reviewers read what changed, and nothing else.
-
It waits for consent, and keeps a record of it.
On a site with a consent platform, the survey does nothing until the visitor has agreed: no request, nothing stored. A question that asks for contact details cannot be published without its consent wording, and that wording is saved with each contact, taken from the approved version rather than from what a browser claims it showed.
-
It refuses to publish certain mistakes.
An email address asked for as an ordinary answer, where it would land among the anonymous ratings. A skip rule pointing at a question since deleted. A rating scale the visitor would not see as written. Each refusal comes with its reason, in a sentence.
-
Side effects are expected.
People mention them wherever they are given room to type. A free-text question can carry your adverse-event notice directly above the box, in wording you supply and your reviewers approve like everything else.
-
The form is built for everyone who must use it.
It was written to the WCAG 2.1 AA guidelines. Every control can be reached and worked from the keyboard, and is named for screen readers. It takes your brand’s colors, and corrects any pair of them that would be too faint to read.
-
Installation is one file and one command.
We give your cloud administrator a settings file and a script whose every line they can read. It shows its plan and waits for the word yes. In our own test the whole installation took less than ten minutes. The token that connects the two sides works once.
-
The data is already home, and stays when we go.
Answers arrive in your own data warehouse as ordinary tables, each row carrying the campaign and source that brought the visitor, ready to join your media reporting. You decide how long answers and contact details are kept, and the database removes them on schedule. Should you leave us, there is nothing to ask us to return.
| survey_slug | sandbox-check |
|---|---|
| survey_version_id | 9a8d1752-f50e-4fc7-a135-77b7a99ea7a5 |
| answers | rating: 5 |
| status | completed |
| received_at | 2026-11-17 05:50:16 |
| page_url | https://…/s/sandbox-test-site/sandbox-check.html |
| traffic_source | direct |
| utm_campaign | null |
| device | desktop |
| visitor_id | oa1p4ihh1rsg51thkk5x |
A row from our own test, as it sits in the warehouse. Look for what is missing: no name, no email address, no IP address, no browser string.
This respondent did leave a name and an email address. They are in a different dataset, which the people who can read this row cannot open.
How it compares with the usual choices
There are three other ways to ask a question on a regulated website, and each has its place. We have described them as fairly as we know how. “Usually” means there are exceptions. The left-hand column is a reasonable list of questions to put to any vendor, ourselves included.
| Question | A hosted survey platformYou paste a script and log in to a dashboard | The same, on its enterprise planYour choice of region, perhaps a dedicated instance | Built by you or your agencyYour form, your server, your database | Survey BuilderWritten with us, run by you |
|---|---|---|---|---|
| Where are the answers kept? | In the vendor’s cloud. | In the vendor’s cloud, in a region you choose. | In yours. | In yours. |
| What does the visitor’s browser contact? | The vendor’s servers. | The vendor’s servers. | Yours. | Yours, and nothing else. |
| Who besides you can read the answers? | The vendor’s staff and systems, within what the contract allows. | The same, usually fewer of them. | Nobody, once the agency has handed over. | Nobody. We have no access to give ourselves. |
| Can the script on your site change without your knowing? | Usually yes. It is served from the vendor’s servers. | Usually yes. | Only when you release it. | No. It changes when your administrator reinstalls it. |
| What must be reviewed before the first question? | The vendor: security questionnaire, data processing terms, sub-processors. | The vendor, and a negotiated contract. | Your own design, by your own people. | Less of us: no respondent data reaches us. What your policies still require is for your counsel to say. |
| Is there a record of the exact wording that ran? | Usually a preview link, or screenshots you take yourself. | Usually the same. | If you build one. | Yes. Every version is frozen, fingerprinted and stamped on each answer. |
| Are names kept apart from answers? | Usually one table and one export. | Usually the same. | If you design it so. | Yes, from the start, with separate readers. |
| How long until the first question is live? | An afternoon, once the vendor is approved. | Often months. | Months, and then the upkeep. | An afternoon, after ten minutes of your administrator’s time. |
| What do you get for reporting? | Standard reporting, in the vendor’s dashboard. | Standard reporting, in the vendor’s dashboard. | Whatever you build. | Reporting customized to you, built on industry-standard templates and run in your own account. |
What it does not do
You will find these out in the first meeting, so here they are now.
- It needs a cloud account of your own. We are adding the public clouds one at a time. Ask which are ready before you plan around us.
- It hosts no dashboards. We cannot see your answers, so we cannot keep charts of them. They arrive as tables. The Looker Studio templates below run in your account, not ours.
- It is not a research panel or an email survey service. It asks questions of people who are already on your website, or who follow a link to a page of yours.
- It cannot stop a determined stranger from filling in a public form. Nothing can. It limits where answers may come from and how many arrive; it does not pretend to know who sent them.
- The product is new. It has no wall of customer logos and no audit report to hand you. What we can hand you is everything that will run in your cloud, for your security team to read before they run it. The person who built it is not new; his record is below.
Reports, without handing us the data
A table is not a report. So Survey Builder comes with Looker Studio templates. You copy one into your own account and point it at your own warehouse. It runs under your login, on your data, and we still see none of it.
- What people answered. Each question’s answers, counted and charted.
- Where people dropped out. How many saw the survey, how many started it and how many finished, page by page.
- Who answered. The same results, split by the campaign, source and medium that brought each visitor.
- What changed when the wording changed. Results side by side for each published version of a survey.
The templates read answers only. None of them opens the dataset that holds names and email addresses.
Help, if you want it
The software installs without us. Some teams would rather not do the surrounding work alone. For them, the people who built Survey Builder take on a small number of engagements a year.
- Installation, alongside your cloud team. We sit with your administrator through the install, and answer your security reviewers’ questions about what each piece does and why.
- Tagging and consent. Fitting the survey into your tag manager and your consent platform, so that it appears where it should and stays silent until the visitor has agreed.
- Joining answers to everything else. Modeling in your warehouse, so that survey answers sit beside your analytics and media data. What did the people who came from this campaign actually say?
- Reports made for you. When a template is not enough: dashboards built around the questions your brand team and your reviewers ask.
Where this work means looking at your answers, you give a named person access in your own cloud for as long as the work lasts, and take it away afterwards. The software itself still has none.
Who it is for
We built it for pharmaceutical brand websites, where a page for patients and a page for prescribers answer to the same reviewers, and where a single sentence may take a month to approve and must then appear exactly as approved.
It suits any organization with the same shape of problem: people who ought to be asked, words that must be signed off, and answers that must not wander. Hospitals and health insurers. Banks and their advisers. Public bodies. Any company whose customers are in Europe and whose lawyers have read the regulations.
Who built it
A new product, from someone who has measured websites since 2003.
Survey Builder is the work of David Salinas, who leads the analytics practice dotReturn.
He joined Google in 2003, on its early analytics team, where he wrote the first Google Analytics aptitude test and trained teams across the company’s offices. He went on to direct web analytics at TBWA\Chiat\Day and then, as vice president of business analytics at LotLinx, built the measurement behind more than 500 clients.
His work has included Nike, GlaxoSmithKline and McDonald’s. Survey Builder comes out of it: twenty years of seeing where the data from a website ends up, and who ends up holding it.
- Brands
- Nike, GlaxoSmithKline, McDonald’s
- Early analytics team, 2003 to 2008
- TBWA\Chiat\Day
- Directed web analytics
- LotLinx
- Vice president, business analytics
- Stanford
- BA, English and creative writing
None of the companies named here is a customer of Survey Builder. They are where the habits were formed.
Give us one question and 30 minutes. We will show you where the answer goes.
Here is a test you can make before you believe a word on this page.
Send us two things: the address of one website, and one question you would like to ask the people who visit it.
Thirty minutes later, on a call, you will see three things with your own eyes:
- Your question, running on a page served from a cloud account that is not ours.
- The row your answer writes in the warehouse.
- The network trace: a list of every address the browser spoke to.
Read that list. If our name is on it even once, end the call. You will have lost half an hour. We will have lost the argument.