Baselines for an initiative that's already live

Most initiatives are measured after they launch. By then the before-figure can't be measured at the time it described. It is either recovered from records of the months before go-live or reconstructed now. Roiva records which one it was, on the baseline and on every figure calculated from it. A report can then say how much of the value rests on a before-figure that was measured, how much on one that was reconstructed, and how much on an estimate.

How a baseline was measured

Every baseline records how it was measured. That decides what a value entry calculated from it rests on.

Method	What it means	The figure rests on
From history	Averaged from your own months before go-live: a connected system, a warehouse query, or history you imported	A measured baseline
Time study before go-live	Timed by your people on the work as it was done before the AI arrived	A measured baseline
Without-AI sample	Timed after go-live on work done without the AI, by people who don't have it yet or with it switched off	A reconstructed baseline
Benchmark	A published figure from outside your company	An estimate
Estimate	A judgment, with what it rests on	An estimate

An estimate or a benchmark is marked an estimate wherever it appears, and you can't save an estimate without saying what it rests on. A figure whose method nobody recorded counts as an estimate.

Three ways to set it, in order of preference

On an initiative that is already live, each baseline still missing on the Setup tab has a Set the before-figure menu with these choices.

1. The months before go-live. If a connected system or a warehouse metric query holds them, Roiva calculates the baseline itself from the three complete months before the initiative started. If they're in an export or a spreadsheet, import them as readings under Entries → Metric Entries → Import, and the baseline is calculated from them. If you only have the average, choose Enter the months' average and record it as From history, with the months it covers.
2. A sample timed without the AI. Time the same task done without the AI, by people who don't have it yet or with it switched off for a few days, and record the average with the number of tasks timed. Count hands-on minutes, not how long the task was open. Use the same definition of the task as the after-figure uses, and time it across several people: thirty or more tasks across three or more people is a sensible floor. Choose Record a without-AI sample, or run it as a time study, which draws the days, gives an interval and publishes the result for you. See Running a time study.
3. A declared estimate. Enter the figure and what it rests on: a staffing plan, a manager's count, last year's survey. Choose Declare an estimate. It stays labeled an estimate until it's replaced by one of the above. Six months after it was set, the Inbox asks for a measured one. Value calculated from it stays approvable meanwhile, marked as resting on an estimate.

On an initiative that isn't live yet, measure the before-figure now, with history or a time study, before go-live makes the work hard to see without the AI.

What every report shows

Each value entry keeps what it rests on: a measured baseline, a reconstructed baseline, an estimate, or no baseline at all. A deflected ticket priced per unit or a figure someone typed and cited compares nothing with a before-figure, so it rests on no baseline. A deflected ticket priced from a time study of agent minutes rests on that study's baseline. When a figure is compared with more than one baseline, the least measured one decides. What it rests on is frozen when the figure is calculated, so replacing a baseline later doesn't relabel a figure someone has already approved. If the initiative has a signed measurement charter, a figure calculated from a baseline other than the one the charter signed is marked Outside its charter and never approved automatically.

• ROI: the Value card says what approved cash and freed capacity rest on, under each figure.
• The Board Brief: its What the cash rests on row gives the quarter's approved cash compared with an estimated or reconstructed baseline.
• The Value Evidence Register: its Summary sheet totals cash and capacity by what they rest on, and the Register has a Rests on column.
• Reporting tables and the API: reported_value.basis and a value entry's basis carry it as measured, reconstructed, estimated or none.

Cash realized from freed capacity rests on what that capacity rested on, in proportion. See Roiva's figures in Power BI and your data warehouse for the tables.

What a reconstructed baseline doesn't prove

A sample timed without the AI measures the work as it is done today without the AI, not as it was done before go-live. Anything else that changed since then is in it: volume, the mix of work, how experienced the team is, other tools. People who know they're being timed may also work differently. That's why it's labeled reconstructed rather than measured.

An estimate is a judgment. Neither one, and no before-and-after comparison, shows that the AI caused the change. That's a judgment you state in the attribution rule of the initiative's charter, and the methodology page says what the method doesn't prove.

Related

• Metrics, baselines, and readings: what a baseline is, and how it's calculated from history
• What an initiative needs before it has a number: the Setup tab and Missing Inputs
• What makes a figure auditable: frozen inputs and approval before anything counts
• Measurement charters: the baselines, formulas and attribution rule signed before results