The Value Report · Contoso

Every figure here is an aggregate event count from your own tenant, recorded when the thing happened. Nothing on this page is a survey answer, an estimate or a login, and no figure can be traced to a person.

A win is a job somebody actually completed with Aristo alongside them — counted at the moment it finished, never a login and never a survey. counted

The seven-day figure is the current week, so a board can see whether this is a habit or a launch spike. It is allowed to go down.

A completed move is a strategy move a leader approved that then ran to the end. Approvals are deliberately not counted here — only completions.

None of these numbers is a self-reported opinion, and none of them can be traced to a person.

So what: a board has been shown licence counts and login charts for two years and has learned to distrust both. This page counts finished work — a job somebody completed, a move that ran to the end, a skill somebody else reused — which is the only activity measure that survives the question "so what?".

You can now decide: whether the work is real and habitual rather than a launch spike — the seven-day figure beside the total is what tells you which.

Net minutes people recorded themselves, gained minus lost, aggregated across a floor of five attesters and converted to hours. Below five people this report shows nothing at all rather than a number a small group would make personally identifiable.

Gained, minus lost. The green bar is moments people marked as time they got back, one job at a time. The amber bar is the time Aristo cost somebody — they said so, and it is subtracted in the same ledger. A ledger with a debit column is the only self-reported number a CFO can defend.

The same currency Microsoft measures. Microsoft's Copilot Analytics Labs publishes agent-assisted hours as its value currency, so this figure and a Labs read are quoted in the same unit and benchmark 1:1. The difference is where the number comes from: Labs estimates the value; Aristo pairs a counted, self-attested number with a never-nudged control group, which is Question 3.

So what: this is the number every AI vendor quotes and almost none of them count. Aristo's version is net and self-attested: the people who did the work record the time they got back, and the time it cost them is subtracted in the same ledger. A figure with a debit column is the only self-reported number a CFO can take into a board meeting.

You can now decide: whether to quote this figure externally. It is in the same unit Microsoft's Copilot Analytics Labs publishes, so it benchmarks against a Labs read directly — with a control group behind it that a Labs read does not have.

How it works, in one paragraph you can repeat. On the first day, Aristo splits your cold seats into two groups by a fixed rule — not by hand, and not re-drawn later. One group gets help. The other gets none, on purpose. Some weeks later Aristo counts how many in each group are genuinely using Copilot. The difference between those two percentages is the lift, and it is the only part of this report that answers "would it have happened anyway?" — because the second group is what "anyway" actually looks like inside your own company.

A fixed rule splits your cold seats on day 0 into a helped arm and a control arm that is never nudged. Neither arm is hand-picked and the split is not re-drawn to flatter a result. The lift is the difference between the two arms' activation rates, and every dollar claimed from it uses the lower bound of a 95% confidence interval.

So what: every other number in this document is counted; this one is caused, and the difference is the whole reason the report exists. A rising line proves a line rose. A control group your own company never touched is what tells you whether the line would have risen anyway — and it is the one thing a dashboard cannot retrofit later.

You can now decide: whether the activation you are seeing is worth paying for — this page measures how much of it Aristo actually caused, so the judgement rests on a counted difference rather than on a rising line.

Protected, not returned — $246 a month. Credits steered onto the included seat plus skills reused instead of re-forged. Both are counted events multiplied by a published task price, which makes the dollar an envelope rather than a measurement — so it lives here and never joins the board figure on the cover. estimate

The recaptured licence line is already inside the cover figure and is shown on this page only so the protection story reads whole. It is counted once, on the cover, and never added twice.

Steering and reuse are counted events from your own tenant; the dollar beside each is that count multiplied by Microsoft's published medium-task credit envelope. The licence line is counted differently — a verified reassignment confirmed on re-read — and it is the same money already on the cover, shown here once more so the protection story reads whole.

So what: money not spent is money returned, but it is a different kind of evidence and it is kept separate on purpose. These figures are counted events multiplied by a published task price, which makes each dollar an envelope. Envelopes belong on this page; the cover carries only what was counted outright.

You can now decide: whether the guardrails are set where you want them. Every path on the traffic lights measures against a floor your own leadership named, never someone else's benchmark.

The multiple is the counted return divided by list price from this platform's own pricing constants. Nothing on this page executes anything: the expand move lives on the Frontier Firm Report and the spending ceiling lives on the traffic lights and in your own Microsoft admin center, so there is exactly one place each decision is made.

So what: this is the sentence the whole series was built to earn. The License Report found the money to pay for the work; the Frontier Firm Report turned the seats on; this page says whether it worked, using a control group your own company never touched. Proof is what turns an expansion into a decision — and this report is equally willing to say hold.

You can now decide: expand, or hold. Both are supported answers here, and this page will print the second one just as plainly as the first.