Taking Your AI Baseline

Taking your AI baseline

Most of The GAiGE runs continuously: short pulses on a cadence, month after month, so you can watch a number move. A baseline is the other kind of question. You ask it once, at the start, and you ask everybody: what are people already using, what is that costing you in risk, where does the repetitive work sit, and what could be handed over.

That's what a one-off pulse is for. It opens, each person answers it once, and it closes into a report you can put in front of a leadership team.

This article covers what a one-off is and how it differs from the pulses you already run, which of the four Discovery pulses to use when, what the people answering can and can't do, how they're reminded, what you can see while it's running, how it closes, and how to read the Baseline Report at the end.

A one-off is not a recurring pulse

A recurring pulse has a cadence. It comes round weekly or monthly, the same person answers it again and again, and the value is the trend line.

A one-off has no cadence at all. It has an opening date, a closing date and a completion target, and it is asked of each person exactly once. Once someone answers, they're finished with it, and that holds even if they reinstall their browser or move to a new laptop: the record of who has answered lives with us, not on their machine.

Everything else works the way you already know. A one-off is added from the library, arrives paused, is switched on when you're ready, and its answers stay readable in Reports afterwards. Closing it is a kind of pausing, so archiving and deleting behave exactly as they do for any other pulse.

	Recurring pulse	One-off
How often	On its cadence, indefinitely	Once per person
Controls	Cadence, delivery window, colour	Closing date, completion target
Ends when	You pause or archive it	It reaches its date or its target, or you close it
The report	A rolling window you choose	The whole run, start to finish

Which one to run

The four Discovery pulses come as two pairs. Each one is about three minutes and asks five or six questions.

The baseline pair answers "where are we now".

The Current AI Use Pulse for what people already reach for, how often it helps, and how they learned it. Run this first. It's the one that tells you whether you have an adoption problem or a licence problem.
The AI Risk Pulse for what people understand about the rules and what kinds of information have gone into public tools. Run it alongside the first one, or immediately after.

The opportunity pair answers "where do we go".

The AI Opportunity Pulse for where the repetitive work sits and what people would hand over first.
The Agent & Automation Pulse for which of that work has the shape software could take on, and what stands in the way.

If you're only running one, run The Current AI Use Pulse. If you're building a case for a programme, run the baseline pair first and the opportunity pair a fortnight later, so people aren't asked for twelve minutes of their week in one go.

Full detail on what each one asks and what its report gives you is in the four Discovery pulses.

Setting one up

Pulses page → Library tab → the Discovery chip → Preview the one you want, then Add.
It arrives paused and opens in the builder. Where a recurring pulse has its cadence controls, a one-off has When this runs:
- Stop asking on a date. Leave it blank and we close it 21 days after it opens.
- Or stop at a percentage answered. Leave it blank and we close it at 85%. Whichever comes first.
Switch on Open (we ask everyone once) when you're ready. That's what starts the asking, and starts the clock: the opening date is stamped then, not when you added it.

You can change the closing date and the target at any point while it runs. The questions are locked, like every library pulse, because the report reads specific answers to specific questions.

What people are asked, and what they can do

Someone receiving a one-off sees it in the extension like any other pulse, with three things said up front before the first question: that it's a one-off and roughly how long it takes, when you need it by, and that their answers are anonymous.

They can always stop. Every screen offers a way out: snooze for a few hours, Later, or simply close it. A one-off that someone closes part-way through goes back into the queue rather than being used up, and it tells them so. Nobody is ever held in a questionnaire.

They can't wave a required question through. Every question is either required or optional, and you set that per question in the editor. A required question has no skip control; an optional one does. Every Discovery pulse's open questions are optional, so nobody is forced to write prose about their week, and a half-answered baseline is a hole nobody goes back and fills.

How people are reminded

Reminders are the extension's job, and they tighten as your closing date gets nearer rather than running at a flat rate:

Time left	How often it comes back
More than 10 days	About weekly
10 days to 3 days	About every three days
The last 3 days	Daily

Never more than once a day, whatever the arithmetic says. A one-off with no closing date sits on the middle setting, so nothing can ask someone daily forever. Unsolicited reminders respect your organisation's working hours; someone who opens the extension and asks for a pulse gets it whenever they like.

The words never escalate. The frequency carries the persistence, and the copy stays the same request it was on day one. There is no countdown, no last chance, and nothing red.

One thing worth knowing before you open one: the extension is how pulses reach people, so a one-off can only reach the people who have it installed. If your install rate is 60%, your completion rate has a ceiling of 60%. Pre-adoption organisations are usually where installs are lowest, so it's worth a push on installs before you open your first baseline rather than after.

What you can see while it runs

Counts, and nothing else.

On My pulses and on the library card, the row reads One-off and, once it's open, "62% of your team have answered".
On the Reports page, the Baseline Report leads with the same figure as a dial, plus "26 of 42 people".
By group shows how far along each team is. A group where fewer than five people have answered shows its name and stops there, and a group the pulse reaches fewer than five people in is left out of the list entirely. That means the group rows won't add up to the headline, and the page says so, because a remainder row would hand back exactly the figure the floor exists to withhold.

Completion is a count, not a roll call. The product has no surface anywhere that names who has answered or who hasn't, and it won't get one: that promise is what the answers depend on.

Closing it

Three ways, and they all land in the same place.

Automatically at the target. Once the share of people who've answered reaches your completion percentage, it closes that night.
Automatically on the date. Once the closing date passes, it closes that night.
Manually. Close now on the pulse's row in My pulses. We ask you to confirm, because it stops the asking for everyone.

A closed one-off is a paused pulse: it leaves the delivery queue, its answers stay readable, and you can archive it exactly as you would anything else. If you want more answers, Reopen on the same row starts the asking again from the original opening date, so the report still covers the whole run. Anyone who already answered isn't asked twice.

Reading the Baseline Report

Open Reports, pick the one-off from the pulse list, and you get the Baseline Report instead of the standard sections. It covers the whole run rather than a rolling window, so there's no period to choose.

The completion dial leads: what share of the organisation has answered, and how many people that is. Under it, how far along each team is.

The gauges. Each Discovery pulse carries one 0-100 reading: Usefulness, Verification, Appetite or Readiness. They're the number you can quote in a board pack and the number you can re-run in six months to show movement.

Breakdowns shows the full distribution for every choice question, not an average. On a baseline this is usually where the argument is: "37% have put internal information into a public tool" is a sentence that changes a meeting.

What your team said is the part a consultant actually presents. Every open question gets its own list of themes, most common first, with a count and a share beside each. Open a theme to read the answers behind it, word for word, exactly as people wrote them. A theme heading is always one of the real answers, never a summary we've written on someone's behalf.

Two things to expect:

Five answers per question, minimum. Below that, a question says how many people have answered and shows nothing else. Same floor as everywhere else in the product.
On a very large run, the themes are built from a spread of the answers rather than all of them, and the list says so ("600 of 2,140 answers in 11 themes"). The spread is taken evenly across the whole run rather than the first arrivals, so it isn't just week one's answers. Every answer is still counted in the total.

If the themes come back as a plain list of everything people wrote rather than a grouped ranking, that's the fallback when grouping isn't available. Nothing is lost; it's just unsorted.

The Baseline Report isn't in the downloadable PDF yet. The themes are made to be opened and read, and a printed page can't do that. Present it from the screen for now.

Running it again

A baseline is worth repeating. Six months into a programme, add the same pulse again, run it for a fortnight, and put the two reports side by side: the gauges are comparable, the distributions are comparable, and the themes tell you whether the work you did answered what people actually asked for.