Copilot was never the mistake. Where the seats sit is. Chronom reads per-app interaction telemetry, not license assignment, ranks who would actually use one, and watches the Copilot Studio capacity, agents and Azure AI meters that arrived alongside the rollout.
You bought a block of seats to evaluate it. Some people genuinely love it. Most opened it twice in April. The invoice never noticed the difference between the two groups.
Assignment quietly became the metric. Nobody ranked who would benefit most, so seats went where the request came from rather than where the hours were waiting to be saved.
Agents and assistants are consumption-billed. It starts small, compounds every month, and sits outside every seat count and every renewal conversation you have.
Copilot is priced like a license and consumed like a habit. The gap between those two facts is where the money goes - and there's a second bill most teams haven't found yet.
Microsoft 365 Copilot is $30 per user per month, usually committed annually, and it appears on the M365 side of your bill as a seat count.
Everything else AI touches bills on consumption and lands on the Azure side: Copilot Studio message packs, agents published into Teams and SharePoint, Security Copilot compute units, Azure OpenAI and AI Foundry tokens. Two bills, two owners, one rollout - and only the first one gets reviewed at renewal.
$30 per seat / month ≈ $360 per year for every seat nobody opens
The license report tells you a seat exists. Interaction telemetry tells you whether it became a habit, and it sorts almost every deployment into the same three groups.
Seats that never started. Seats that stopped around week three, once the novelty passed. And single-surface seats using one feature - usually Teams meeting recaps - while paying for the whole thing.
Copilot terms don't prorate. Release a seat in month four of a twelve-month commitment and you keep paying for it until the term ends, so the finding is only worth what the calendar lets you bank.
That's why idle seats get reassigned immediately and released at the anniversary. The reassignment returns value from money you've already spent; the release is what actually shrinks the invoice.
Word, Excel, PowerPoint, Outlook and Teams, over months rather than weeks. Assignment counts tell you nothing here.
Never started, lapsed after week three, single-surface. Each group gets a different decision, and only one of them is a release.
Content that isn't search-ready, oversharing permissions and sensitivity labels that silently block responses all read as low usage. Fixing those is cheaper than more seats.
Meeting load, authoring volume, mail throughput and collaboration breadth. This is the list you assign from, instead of whoever asked first.
Reassignment converts committed spend into value today. The release is scheduled for the date where it actually reduces the bill.
Copilot Studio packs, published agents and Azure AI meters get a team, a budget and anomaly detection - so the next AI line item isn't a surprise.
Examples rather than the catalogue - a sample of what the platform reads across assignments, interactions, capacity packs and AI meters.
The question was never whether Copilot was a mistake. It's whether the seats are on the people who convert them into hours - and whether anyone is watching the consumption-billed AI that arrived alongside them.
Copilot use in Word, Excel, PowerPoint, Outlook and Teams, per person, over time. That's what separates a genuine daily habit from the seats that were tried once and quietly abandoned after the novelty passed.
A seat looks identical in the license report whether it's used forty times a day or never again.
Meeting load, authoring volume, mail throughput and collaboration patterns rank the people most likely to turn a Microsoft 365 Copilot seat into hours back. Move seats there instead of buying more, and the adoption story improves at zero net spend.
Adoption dashboards tell you who used Copilot. Nothing tells you who would.
Purchased message capacity and agent packs measured against what's actually consumed. That includes agents published into Teams and SharePoint that outlived the project they were built for and still hold capacity.
Capacity is bought in blocks and consumed invisibly. Nothing reconciles the two for you.
Azure OpenAI and AI Foundry token spend, Security Copilot compute units, and the agent meters that land on the Azure invoice rather than the Microsoft 365 one - each tagged back to the team that created it.
It isn't in your seat count, so it never shows up in the licensing conversation or the renewal.
Copilot value depends on its substrate: content that's search-ready, permissions that aren't oversharing, and sensitivity labels that don't silently block responses. We also flag standalone AI tools duplicating what the Copilot seat already includes.
A content-starved or label-blocked Copilot reads as an adoption problem, and adoption problems get solved by buying more seats.
Copilot terms don't prorate mid-cycle, so the date of a reclaim matters as much as the finding. Every idle seat is mapped to the subscription anniversary where releasing or reassigning it actually banks money.
Release a seat mid-term and you keep paying for it to the end of the term anyway.
Copilot terms don't prorate mid-cycle, so reclaims are timed to your anniversary and we handle the reassignment - the saving is banked rather than theoretical. Seats move to ranked candidates, agents get an owner and a budget, and the meter picks up anomaly detection so the next AI line item isn't a surprise.
Figures on this page describe an illustrative 300-seat Copilot deployment at published rates. Your audit replaces them with your own numbers, tied to named users and interaction history.
Chronom outputs conclusions rather than signals: which seats move, who they move to, when the move actually banks money, and what it's worth over a year.
No recorded interactions in 90 days on any Copilot surface. The people who use it daily keep their seats untouched.
Interactions stopped after week three. Seats move to the highest-ranked candidates instead of new seats being purchased.
Single-surface use only. Enablement first, with the release decision taken at the anniversary if the pattern hasn't changed.
Message consumption measured against purchased capacity across two full quarters, with headroom kept for the agents still in use.
Every consumption-billed agent mapped to the team that created it, with tagging and anomaly detection on the meter.
Every line in a Chronom report is a decision with an owner, a SKU and a dollar figure. Nothing is left for you to go and find out.
“Low Teams usage detected” - a fact you now have to interpret
An alert queue that grows faster than you can triage it
A dashboard that hands the analysis back to you
Reclaiming one round of idle seats is a single win. New assignments follow the same curve, new agents get published, new meters start. Chronom keeps watching both the seats and the consumption behind them.
A Copilot seat assigned this month is evaluated on the same interaction telemetry as the last cohort, so a dormant seat surfaces in weeks rather than at the next renewal.
When a team asks for Copilot seats, you can see whether idle ones already exist in the estate - and who the ranked candidates are - before the purchase order goes out.
A new agent, a spike in token consumption or a Studio pack burning faster than forecast is flagged within a day, not discovered at month-end close.
The assigned-to-active ratio becomes a number you can put in front of an executive sponsor, with the seat movement that produced it attached.
For the right person it pays for itself in a week, and for the wrong person it's the most expensive unused row on your invoice - roughly $360 a year each. The variable isn't Copilot, it's the work: heavy meeting loads, high authoring volume and heavy mail throughput convert into hours back, while people whose day is neither of those things try it twice and stop. Across audited deployments 25–40% of assigned seats are redeployable, which is why the useful question is never 'is Copilot worth it' but 'which of these 300 seats is on the right person'.
It usually improves it. Most of what we find isn't 'cancel Copilot' - it's move seats from people who stopped after week three to the people whose meeting load, authoring volume and mail throughput say they'd use it daily. Your assigned-to-active ratio goes up, the spend stays flat, and the executive who championed the rollout gets a better adoption number than the one they had. Releasing seats outright is only the recommendation where there's nobody left worth moving them to.
No. The connection is read-only through the Microsoft Graph API and reads interaction signals and metadata - which apps a Copilot seat was used in, how often, and when it was last used. Never prompt text, responses, documents or messages. Chronom is SOC 2 Type II compliant and nothing in your tenant changes until you approve a specific action.
From the work patterns Copilot actually accelerates. Meetings attended and recorded per week, documents authored and co-authored, mail volume sent and received, chat and collaboration breadth - the same signals that predict where meeting recaps, drafting and summarisation return the most time. It's a ranking rather than a guarantee, which is exactly what you want when the alternative is guessing or buying seats for whoever asked first.
The opposite - it makes the timing part of the work. Knowing which seats are idle months ahead means the reclaim is scheduled for your subscription anniversary, where it converts into a real reduction rather than a mid-term change you keep paying for. In the meantime those seats get reassigned to ranked candidates, so the money you've already committed starts returning something before the anniversary arrives.
Yes, and they're often the part nobody is watching. Copilot Studio message packs and agent capacity get measured against actual consumption; agents published into Teams and SharePoint get an owner and a budget; Security Copilot compute units get checked against how they're scheduled. Consumption-billed AI is the fastest-growing line in most Microsoft estates precisely because it doesn't look like a seat.
It's covered under the same audit. Azure OpenAI and AI Foundry token consumption, agent meters and anything else billed on usage lands in the cloud domain, gets tagged to the team that created it, and picks up anomaly detection so a new meter surfaces within a day instead of at month-end close. That's the cross-domain effect in a sentence: the AI rollout happens on the licensing side and the bill arrives on the cloud side.
Get a comprehensive audit of your environment and see exactly how much you can save in under 15 minutes.