Start here
Seven short pages. The first three explain what the wellbeing check-in, our Intelligent Wellbeing Engine, is and why it is built the way it is. The last three are about running it once your people are switched on.
None of this is homework. If you would rather have twenty minutes on a call than read anything, that is genuinely fine. Email hello@alltoogether.com or call 0161 515 8570 and we will walk you through whichever part you care about.
Why we measure, and why this way
You are about to ask your people three questions every fortnight. Before you do, it is worth ten minutes on why those three, why that often, and what the answers can honestly tell you.
The problem with how wellbeing is usually measured
Most employers do one of two things. They run an annual engagement survey of forty questions, get a report three months later, and act on almost none of it. Or they do nothing, and rely on managers noticing.
Both fail for the same reason: by the time you know something is wrong, it has already been wrong for months. An annual survey tells you about last spring. A manager noticing tells you about the people who show it.
Meanwhile you are already spending money on this. Private medical cover, an employee assistance programme, income protection, a cash plan. That spend is a bet that your people need those things. Nobody usually checks whether the bet is right, or whether the help you have bought is reaching the people who need it.
The check-in exists to make that spend answerable. Not to diagnose anyone, and not to score your culture, but to tell you where support is needed while there is still time to do something about it.
Why three questions instead of thirty
The instinct is that more questions mean better data. The evidence is that they mean fewer answers and worse ones.
This is not a fringe position. The Office for National Statistics measures the personal wellbeing of the entire country using four single questions.
We ask three each fortnight, from a set of nine: the same two every time, so you can see movement, plus one that rotates through the other seven. The first time, everyone answers all nine once, so there is a baseline; after that it is three a fortnight, and the rotation covers the whole set without ever putting a long survey in front of anybody.
Why every fortnight
Frequency is a trade-off. Ask too rarely and you cannot tell a real change from noise. Ask too often and people stop answering.
Three questions takes under a minute. Spaced a fortnight apart, that is minutes of someone's year rather than hours. From roughly six cycles, about twelve weeks, a change at organisation level becomes distinguishable from noise.
Why it is confidential, and why that is enforced rather than promised
Your people will only answer honestly if the confidentiality is real. So it is worth being precise about what it is, because "anonymous" gets used loosely and your employees deserve better than a loose word.
The system knows an employee submitted. It does not let anyone see what they submitted. Responses are stored against an identifier that is not a name, and there is no interface, export, admin role or support function that can retrieve an individual's answers. Not for you, and not for us.
Results are only ever shown as a group figure, and only when at least five people in that group responded. Below five, no figure is calculated or stored at all. That floor is a constraint in the database itself, not a rule in the interface, which means there is no setting to turn it off and no way around it.
Questions touching severe distress carry a higher floor of ten. And the floor applies to each figure shown, not to each query, so slicing the data different ways cannot be used to rebuild a small group.
What you will actually see
This is a real example of a result, including the parts most tools hide.
Three things are deliberate here. Every figure carries a confidence interval, because a score of 61 from nine people is a range rather than a point, and printing "61" alone would claim more precision than the data supports. Thin figures are labelled indicative rather than presented as equal to the rest. And where a figure is withheld, we say so and say why, instead of showing a blank, a zero or a dash that you might read as a result.
What this can and cannot tell you
The fastest way to lose your leadership team's trust is to claim more than the method supports. So, plainly:
| It can | It cannot |
|---|---|
| Show whether a group's experience is moving, and in which direction, from about six cycles | Tell you about any individual, ever |
| Compare areas of working life against each other within your organisation | Diagnose a clinical condition, or substitute for occupational health |
| Show where the support you already pay for is not reaching people | Prove that a change you made caused an improvement |
| Give you a defensible, dated record that you looked | Correct for the fact that people who answer may differ from people who do not |
That last one is why participation is always shown next to every result. A high score from a third of your people is a different fact from a high score from nine in ten, and you should always be able to see which you are looking at.
The one thing worth remembering
This is not a survey tool with a wellbeing theme. It is a measurement instrument built so that the numbers it produces will survive being challenged, by your board, by an employee who is sceptical, or by a regulator. Everything above exists so that when someone asks "how do you know that?", you have an answer.
- Galesic, M. & Bosnjak, M. (2009). Effects of questionnaire length on participation and indicators of response quality. Public Opinion Quarterly, 73(2).
- Wanous, J. P., Reichers, A. E. & Hudy, M. J. (1997). Overall job satisfaction: how good are single-item measures? Journal of Applied Psychology, 82(2).
- Porter, S. R., Whitcomb, M. E. & Weitzer, W. H. (2004). Multiple surveys of students and survey fatigue. New Directions for Institutional Research, 121.
- Rolstad, S., Adler, J. & Rydén, A. (2011). Response burden and questionnaire length. Value in Health, 14(8).
- "Student" (Gosset, W. S.) (1908). The probable error of a mean. Biometrika, 6(1).
- Hundepool, A. et al. Handbook on Statistical Disclosure Control, ch.5.
- Government Statistical Service / GSR disclosure control guidance.
- Information Commissioner's Office, anonymisation guidance.
The full method, including the rotation design and the statistics, is published openly at openworkplacehealth.org and wellbeingengine.io/evidence. We would rather you were able to check us than take our word for it.
What do you ask my people?
Three questions about work and wellbeing every fortnight, under a minute. They come from a set of nine: the same two every time, plus one that rotates through the other seven. The first time, everyone answers all nine once, so there is a baseline; after that it is three a fortnight.
Three per person per fortnight adds up faster than it sounds: a fifty-person company generates well over three thousand answers a year. The design trades depth-per-sitting for honesty, people answer three questions truthfully at their desk, and abandon or speed-click a forty-question survey. The two fixed questions give you a consistent line to watch; the rotating one fills in the rest of the picture over time.
The method behind spreading questions across people
Because the engine decides who gets which rotating question, always the one that person has gone longest without seeing, the gaps in the data are created by the design, not by anything about the person or the moment. That property is what lets answers be pooled across cycles without bias creeping in. The technique is called a planned missing-data design, and it is an established method for covering more ground than any one respondent could reasonably answer.
The two that appear every time:
| Area | What it is really asking |
|---|---|
| Work ability | Capacity to do the job as it is now |
| Burnout | Exhaustion and depletion, not a diagnosis |
And the seven the third question rotates through:
| Area | What it is really asking |
|---|---|
| Workload | Whether the amount of work is manageable |
| Sleep | Often the earliest thing to move |
| Overall health | Self-rated health, comparable with population figures |
| Financial confidence | Whether managing money feels in hand |
| Job satisfaction | How the job itself is sitting with them |
| Intention to leave | Whether they are thinking about going |
| Your benefits | Whether people actually know what they have |
Only where we had to. Wherever a properly validated measure exists and can be used commercially, we use it rather than writing our own, a question that has been tested on thousands of people beats one written in a meeting. And where a well-known scale is licensed for research use only, we do not quietly reproduce it, because that would put both of us on the wrong side of its licence. Some familiar-sounding scales are absent for exactly that reason.
No, and it is a deliberate no. The moment companies edit the questions, answers stop being comparable, with other companies, with population figures, and with your own past. It also breaks our promise to your people that the wording they see has been validated rather than improvised. If something matters to you that the set does not cover, tell us; extending it properly is our job, not a settings page.
We will happily walk through all nine questions and where each one comes from. hello@alltoogether.com · 0161 515 8570
Can anyone see what an individual said?
No. Not you, not their manager, not their chief executive, and not us.
You see one number for your whole company, and only once at least five people have answered. Below five, no number is created at all. There is nothing sitting in a database waiting to be unlocked.
There is no screen that shows it. Not a hidden one, not an admin one. The rule that blocks it is written into the database itself rather than into the software, which matters because software can be changed by whoever is building it that week, and this cannot be changed without a deliberate, recorded alteration to the database that a developer would have to justify.
Show me how that actually works
The table that holds results carries a constraint, a rule the database enforces on every write, saying a row cannot exist if fewer than five people contributed to it. An attempt to write one fails outright rather than being quietly filtered out afterwards.
Individual answers live in a separate place that the employer-facing part of the system has no permission to read. The permission is set at the database level too, so even a mistake in the application code cannot reach across.
Partly, and we would rather say so. If you have eight people, the company figure is a group of eight. That is still enough to hide any one person's answer, but it is a small enough group that someone who knows the team well might make a reasonable guess about a big movement. What the system guarantees is that it will never confirm that guess.
If you have fewer than five people answering, you will simply not see a number, and we will tell you why rather than showing you a blank.
They ask, and everything connected to them goes, their answers, their history, and the code that linked them to it. Afterwards there is no way to tell that they ever took part, which is the standard we hold ourselves to rather than merely marking a record as deleted.
What about figures already published?
Results from cycles that have already closed are not recalculated. That sounds like a loophole and is in fact the opposite: if a company average shifted the moment someone was erased, the shift itself would reveal what they had answered.
This is the cleverest version of the question and it deserves a straight answer. If you could see the whole company and also one department, subtracting one from the other tells you about everyone else. Today it cannot happen, because the only figure that exists is the whole company, so there is nothing to subtract.
When we add department-level results, each will carry the same five-person rule on its own. We will not release that feature until we can also withhold any figure that could be used to rebuild a smaller group by subtraction. That work is underway and the feature does not ship before it.
Two things, said plainly because a page like this is worthless if it only lists strengths.
It cannot stop someone guessing. If a team of six has an obviously difficult quarter and the number drops, people will draw conclusions. The system will not confirm them, but it cannot prevent them being drawn.
And it cannot protect someone who tells you themselves. If an employee talks to you about how they are doing, that is a conversation, not data. It sits outside all of this, and it always will.
We will join a team meeting and answer this question ourselves rather than leaving you to defend it. hello@alltoogether.com · 0161 515 8570
What do the numbers actually mean?
A score from 0 to 100 for your company, where higher is always better, you never have to remember which way a question points. Next to it, a range showing how sure we can be, and a label that tells you plainly when a figure is too thin to lean on.
Workload. 18 of 24 responded · 61, plausibly 54 to 68 · Full
Sleep quality. 9 of 24 responded · 72, plausibly 61 to 83 · Indicative
Financial confidence, Withheld · fewer than 5 responses · Not shown
Three different situations on one screen, and the honest reading of each is different. That is the point of the extra furniture around the number.
"61, plausibly 54 to 68" means the true figure for the group could reasonably be anywhere in that band. If next cycle reads 64, nothing has happened, the ranges overlap almost completely. The range is the part most tools leave out, and it is the part that stops the most mistakes.
Why the range is wider for small groups
With few responses, the spread of the answers is itself uncertain, so we widen the interval to account for that rather than pretending to normal-sized precision. At five responses the multiplier is about 40% larger than the one most software quietly uses by default. This has been the correct way to handle small samples for over a century; most dashboards just do not bother.
Between five and nine responses, the figure is real but thin. Treat it as a prompt to look, not as the basis for a decision you would have to defend.
Not a blank, not a zero, a stated refusal with its reason. A blank invites you to assume nothing is there; a zero invites you to assume something bad. A withheld row means neither: it means fewer than five people answered, so no number was calculated at all. When you see one, the useful response is to help response rates, not to guess at the hidden figure, because there is no hidden figure.
| The mistake | What to do instead |
|---|---|
| Reacting to a single cycle | Wait for a direction across roughly six cycles, about three months. Anything less is mostly noise. |
| Comparing two groups by their scores | Check whether the ranges overlap. If they do, you cannot honestly say one is higher. |
| Reading a high score from few responses as good news | Look at completion first. The people who did not answer are the ones you know least about. |
| Assuming a change was caused by what you did | The measure shows movement, not cause. Record what you changed and when, so you can argue the link honestly. |
Send it over and we will tell you what it can and cannot carry. hello@alltoogether.com · 0161 515 8570
Launching the check-in to your people
About an hour of your time, spread over a fortnight. Most of it is one conversation about how your organisation is arranged, two decisions, and a short message from someone senior.
Two things are being switched on, not one
The first is the part you see: a whole-company picture of what is actually happening, built up cycle by cycle. The second is the part you never see, and it runs on every single answer, the person answering gets pointed back to the support they can already use, whether or not enough of their colleagues answered for anything to show on your side. The checklist below sets up both. The step people skip is the second one.
Before you switch it on
- Tell us how your organisation is actually arrangedWho reports to whom, and what your departments or sites are called. It is the one step that genuinely needs you rather than us. It does not change what you see in wellbeing today; that is explained in the next section, but it is what makes the rest of the platform work, and it decides what you will see when team-level results arrive.
- Check the support list we hold for you is rightEvery answer ends with a route to support. That route is only as good as the list behind it, your cover, your employee assistance line, anything you run yourselves. If we do not know about it, your people get pointed at the national services instead of the thing you are already paying for. Also covered below.
- Decide who sees the resultsMost organisations start with HR and the leadership team, and open manager-level access later once people trust the thing. There is no wrong answer, but decide before you launch rather than after someone asks.
- Get someone senior to ask, in their own wordsThe launch email carries a short message from a named person. It is the single biggest influence on whether people answer, and it is the last section on this page.
What you will see, and what the floor actually means
Right now, wellbeing results are reported for your organisation as a whole. One figure for the company, with its uncertainty shown, cycle by cycle. There is no team-by-team breakdown yet.
The floor applies to that whole-company figure: nothing is computed or stored at all until five people have answered in a cycle. Below five, no value exists anywhere, not hidden behind a permission, not stored and withheld. Between five and nine responses the figure is published but labelled indicative, because at that size the uncertainty around it is genuinely wide. From ten upwards it drops the label.
| Responses in a cycle | What you get |
|---|---|
| Fewer than 5 | Nothing. A row saying the result is withheld, with the reason. No value is calculated. |
| 5 to 9 | A figure, labelled indicative, with a wide interval around it. Directionally useful, not something to act on alone. |
| 10 or more | A figure with a usable interval. This is where it starts earning its place in a decision. |
For most organisations this means the practical question at launch is simply whether you can get five people to answer, which is a much lower bar than it sounds.
Team-level results are coming, and the structure you give us now decides them
We are building the breakdown by department. When it lands, the same rule applies independently to each one: a department needs its own five responses before anything shows for it. That is the reason to think about structure now rather than later.
The shape that works depends on how you actually operate rather than on your org chart. A company of forty split into ten teams of four will see nothing anywhere once the breakdown arrives, while the same company grouped into three or four broader areas will see all of them. Broader is the safer starting point: splitting a group later keeps its history, whereas a group that never produced a result has no history to keep.
What we are asking for, and why
Two things. Who reports to whom, and what your departments or sites are. Neither is so a manager can see individual answers, they cannot, and neither can you.
Reporting lines are what make leave approvals route to the right person, the people directory make sense, and any future manager-level view have a defensible boundary. Departments and sites are what the wellbeing breakdown will use, and if you have more than one location it is worth telling us, because site differences are usually larger than functional ones and get hidden if you only group by team.
Practically this is a conversation on your setup call, not a form. If you have an HR system or a payroll export, most of it comes straight from there and we do the work. If not, a spreadsheet with names, departments and managers is plenty. None of it is permanent, you can change any of it whenever you want, and answers already given stay attached to the people who gave them.
The support list, and why it matters more than the questions
When someone answers, the check-in ends with support they can use. Some of that is universal and always there, the national helplines are shown to everyone, every time, regardless of what anyone answered. The useful part is the rest of it: your medical cover's counselling line, the employee assistance programme, the physiotherapy route, the GP line, the thing in your policy nobody has ever used.
We build that list from the cover we broke for you, so most of it is already there. What we do not automatically know is what you run yourselves or buy elsewhere. So before launch we will ask you to confirm the list and add anything missing.
This is the step that converts a measurement tool into something that helps a person on the day they need it. Skip it and the check-in still works, but it points your people at national charities while your own paid-for support sits unused.
Getting someone senior to ask
People decide whether to answer based on who asked and why. A message from a named senior person, in their own words, outperforms anything written for them, which is exactly why we are not going to hand you a script to paste. A pasted message reads like a pasted message, and your people will know.
What we will do is tell you what it has to contain. Under a hundred words, four jobs:
Specific to you, not to wellbeing in general. "We spend a lot on cover and support and we do not know what lands" works. "We care about your wellbeing" does not, because everybody says that.
Three questions, a fortnight apart, under a minute. Say the number.
Not "it is anonymous", that is a claim. "I only ever see averages of five or more people" is a limit, and limits are believable in a way that claims are not.
The one most organisations leave out, and the one that decides whether cycle six gets answered as well as cycle one.
If it would help, we will draft it with whoever is sending it on a twenty-minute call and it will sound like them. That tends to be faster than editing something we wrote blind.
The first three cycles
The first check-in usually gets the best response you will see for a while, because it is new. Cycle two is often the lowest. Do not read either as a signal about your organisation.
Even "we have had two rounds, here is what we are seeing so far, keep going" is enough. Silence after asking is the fastest way to kill a response rate.
By now you have three points. Look at whether response is holding steady and whether the interval is narrowing, rather than at whether the figure went up or down. A stable response rate is worth more at this stage than a good score.
Do not expect a usable trend before about six cycles, which is roughly three months. Anything you see before that is a starting position, not a direction.
What we do without you asking
The platform sends the invitation, opens each cycle, reminds people who have not answered, and stops reminding the moment they do. You do not need to chase anyone, and you can see everything that has gone out and switch any of it off.
It also does the routing. Every person who answers is shown the support available to them, and where an answer suggests someone might want more than a list, they are quietly offered more, on their own screen, with nothing sent to you, to their manager, or to us. You will not see when that happens, and neither will we. That is the design, not a limitation of it.
Getting enough people to answer
Five answers in a cycle and you get a result. Ten and it stops being labelled indicative. Beyond that, what matters is not the peak but whether the number holds steady. Here is what actually moves it.
Where you are
What works, in order of effect
By a distance the strongest lever. One line in a company update, naming something you did because of what the results showed. It converts answering from a favour into an act with a consequence.
Not HR, and not the platform. Someone whose name people recognise, saying why they personally want to know.
If a team lead says "I have done mine" in a stand-up, their team's response rate moves. If managers ignore it, so does everyone else.
A check-in that lands mid-morning midweek gets answered. One that lands at five on a Friday gets buried. We can shift the send window for your organisation.
Nobody keeps answering a survey that has never produced anything. If you are a small organisation hovering around the floor, the first job is simply getting five people over the line in one cycle, so that something real appears and people can see the point.
What does not work, and one thing that actively backfires
Chasing individuals does not work, and it frightens people who assumed the thing was confidential, because being chased implies someone knows. Let the platform do the reminding.
Making it compulsory does not work either. It converts non-response into false response, which is worse than a gap, because a gap is visible and a lie is not.
When low response is telling you something
A group of people who have stopped answering have usually not become lazy. They have usually concluded that answering does not change anything, or that it is not safe. Both are findings.
Before treating low response as an engagement problem, consider treating it as a result. It is often the earliest signal you will get, and it arrives before any score moves.
It is also the one signal that survives the floor. Response and headcount are visible to you even when no result is, so a cycle where participation halves tells you something on the day it happens, whatever the numbers do afterwards.
Setting an honest target
Aim for steady rather than high. The same two thirds of your people answering every cycle is worth far more than everyone once and a third after, because a stable rate means you are comparing like with like when you look at movement. Volatile participation makes every trend unreadable, and it is the most common reason a figure appears to move when nothing has actually changed.
Turning a result into something that happened
Measurement without action is surveillance with extra steps. This is the loop that stops that, and it is the part that produces evidence you can show.
There are two loops, and only one of them is yours
The fast loop closes in seconds and does not involve you: someone answers, and they are pointed straight back to the support they can use: their cover, the assistance line, the thing in the policy nobody has read. That happens whether or not the company ever reaches five responses in a cycle, and you never see it. The slow loop is the one on this page: what you do, quarter by quarter, about the pattern underneath. Both matter. If only the slow one is running, people are waiting on a committee.
The loop you own
Look at direction across cycles rather than this cycle's number. Note where the intervals are wide, and where rows are withheld.
Not five. One. The organisations that get value from this pick a single area per quarter and leave the rest alone.
An action without a named person and a date is a note. The platform records both, which is what turns a decision into something reviewable later.
Say what you saw and what you are doing. This is also the single strongest thing you can do for your response rate, so it pays twice.
Come back at the review date and record what happened, including if the answer is nothing. A record of an honest failure is worth more than a record of nothing at all.
Choosing what to act on
Three sensible tests, in order:
- Is it moving? A poor score that is stable is a known condition. A decent score that is sliding is a developing problem, and it is usually the better target.
- Can you actually affect it? Workload and manager support are usually within your gift. Financial confidence is largely not, though what you signpost people to is.
- Is the support already bought? This is the one most people miss. If sleep and fatigue are moving and your medical cover includes a sleep programme nobody has used, the action is not a new initiative. It is telling people what they already have.
That third test is why this sits inside a benefits platform rather than a survey tool. Most of the time the answer to a signal is something you are already paying for and nobody knows about.
Which is also why the support list is worth revisiting each quarter rather than only at launch. If a pattern keeps appearing and there is nothing in your list that speaks to it, that is not a signposting problem: it is a gap in what you buy, and it is the most useful thing this will ever tell you at renewal.
What good actions look like
| Weak | Better |
|---|---|
| "Launch a wellbeing campaign" | "Workload has slid for two cycles: send everyone a note about the counselling line in the medical cover, and check usage in six weeks" |
| "Improve manager support" | "Ask managers what would help with the area that moved, and report back at the next cycle" |
| "Monitor the situation" | "Do nothing this quarter, review at cycle 12, recorded here with a reason" |
The third row is not a joke. Deciding not to act, recorded with a reason and a date, is a legitimate action and a defensible one. What is not defensible is looking, seeing something, and leaving no trace of having considered it.
Why the record matters as much as the action
If you are ever asked what you knew about your workforce's wellbeing and what you did about it, the useful answer is not a score. It is a dated sequence: here is what we measured, here is what we saw, here is what we chose to do, here is who owned it, and here is what happened. That sequence is what this section builds, cycle by cycle, without you having to assemble it afterwards.
Ask us anything about this, including the awkward questions. hello@alltoogether.com · 0161 515 8570