A quarter of your meetings never happen, and nobody is counting

Every cost in a business ends up filed under something. That is how you decide what to fix. You look at where the money and the hours are going, you find the worst offenders, and you prioritise.

The trouble with scheduling is that it has nothing to be filed under.

Your HR system is the record of who works for you. Your CRM is the record of your customers. Your ERP is the record of your money. Each of those has a home, a ledger, a place the cost of getting it wrong shows up so someone can act on it. Scheduling has none of that. The effort of coordinating people's time does not sit in its own column anywhere. It has an ephemeral quality. There is no coat peg to hang it on.

So it gets hung on the nearest one instead.

A candidate takes another offer because you could not get them in front of the panel in time. That goes down as a recruiting problem. A deal slips a quarter because the right three people could not find an hour together. That is a sales problem. Someone works another evening to absorb a double booking nobody saw coming. That is not a problem at all, until it turns up months later as burnout, in a different part of the business entirely.

In every one of those cases the thing that actually moved was time coordination. It just never gets called that, because there is nowhere to call it that.

A while ago we ran a hire that took nine days from first interview to offer. Three stages, multiple people on the final panel. Over the same period, the companies we were competing with for those candidates were still trying to get their first or second interview in the calendar.

We saw that for what it was. A scheduling win. Not a lucky quarter, not a heroic recruiter, but the direct result of how we run coordination.

The reason we could see it is that we have separated the scheduling out from the workflow it serves. We know which part of an improvement is workflow specific and which part is scheduling specific. A scheduling improvement makes the workflow better, but we hang the credit on the decision about how to run scheduling, not on the recruiting process it happened to be helping.

Almost everyone else files the identical result under "we run a good hiring process." Which is a lovely thing to believe and impossible to reproduce on purpose, because the cause is spread across the workflow with no home to point at. That is the whole problem in one sentence. A win you cannot attribute is a win you cannot repeat.

One of our recruiting customers, who I will reference rather than name, looked at a couple of months of their scheduling activity once it was sitting in one place. Several thousand requests.

Roughly a quarter of them never happened. Cancelled, expired, or declined.

That drag had been there the whole time. It was never a line on a dashboard, never anyone's target, because there was nothing to measure it against. It was spread across scores of coordinators and thousands of candidates as a vague, shared sense that hiring is just a slog. An unaccountable cost is a wobbly plate you cannot see, so you never learn to prioritise fixing it. You simply keep paying.

Then it had a home, and it became a number. A quarter of the work was producing nothing.

The first thing that jumped out was reschedules coming from the internal side, the interviewing team changing meetings rather than the candidates. That was not necessarily the whole problem. But it was the part that was clearly within reach, the part a company can act on directly, so it was the place to start.

The fix was to encode the rule into the scheduling process itself. Cap how many interviews a person does in a given period. Two a week, say.

That is not a scheduling nicety. It is the social contract between the recruiting team and the interviewing team, written down so that the system honours it. The interviewers get to keep doing their actual jobs. The recruiters still get people hired quickly. Neither side has to keep renegotiating it by hand every week. We are not deciding whether you should interview. We are capturing the terms you already agreed to, and holding everyone to them. Availability, not suitability.

Internal reschedules came down. Which is the point of having the record. You spot the thing you can influence, you change the process, and the number moves.

With the internal side handled, the data pointed somewhere less comfortable, and less within anyone's direct control.

Ask most people running a hiring process who causes the reschedules and they will tell you it is the interviewers. Busy senior people, diaries full, bailing at the last minute. The record said the opposite. Candidates were rescheduling roughly four times as often as the coordinating side did.

You cannot cap a candidate. So this was a different class of problem, needing a different kind of fix. Not a rule to enforce, but a process to make easy and self correcting.

That is where the next set of changes came from, and again they were in the customer's control. Keep self serve rescheduling switched on, so times get settled against live availability instead of negotiated over email. Send a nudge two or three days before a booking link expires, because a meaningful share of links were simply lapsing. Stop rebuilding requests from scratch, since editing an existing one beats starting over and there were hundreds of needless duplicates. And go and find the cancellation pockets, the corners running above 40% against a baseline nearer 11%, and sort them out with the hiring team directly.

None of that is visible if scheduling has no record. You would have gone with the intuition, leaned on your interviewers, and tuned the wrong system entirely, confidently, because the fault was hiding under a workflow that was not where the fault lived.

Here is the part that matters even if you never hire at volume.

A recruitment process is a series of milestone meetings between people who have to find time together. So is a sales process. So is customer onboarding. So is professional services delivery. The ATS is convinced hiring is its problem and the CRM is convinced the deal is its problem, and each is right about its own half. Underneath, the mechanic is identical. People finding time to reach the next milestone. Same interactions, same complexity, same quarter falling out of the bottom.

This is the case for a system of record for time, rather than a scheduling feature bolted separately onto each workflow. A feature per workflow gives every team its own private blind spot. A shared record of the scheduling transaction, tracking its whole life, the reschedules and who initiated them, the cancellations, the expirations, the completions, gives you one consolidated view across all of them. That is the only thing that turns the cost from something ephemeral into a number you can move on purpose. And the faster and more reliably you move it, the faster you get to the only outcome anyone ever wanted. Person in post. Deal closed.

Something has changed since we first started making this argument. There is a new kind of worker turning up in these workflows, and it cannot tolerate the ambiguity at all.

Agents are good at one thing in particular. They orchestrate primary infrastructure. The CRM, the ATS, the ERP. These are the systems where the rules, the compliance, the regulation and the company's own habits are already encoded. An agent does not have to hold any of that in its head. It coordinates across systems that already do, and defers to them for the parts that have to be right every time.

That works in every domain that has such a system. It falls apart in the one that does not. Point an agent at a calendar and there is nothing authoritative to defer to, because the calendar is not a record of anything. It is an access surface people manage for their own convenience, full of holds and blocks and meetings nobody intends to attend. The agent is reasoning over a fiction, confidently.

And it is worse than that, because the agent cannot even do the arithmetic. Hand a model a perfectly accurate clock and it still cannot reliably tell you that a 9am meeting is in the past when it is 11am. Our own team hit this head on building our scheduling agent. Models turn out to be poor calculators behind a very convincing disguise. The fix was not to coax better reasoning out of the model. It was to stop asking it to reason, and to hand it the answer already worked out, past or active or future, as a plain label it could just read. The deterministic substrate does the thinking. The agent orchestrates.

That is the general point. An agent acts probabilistically. Coordinating people's time to a reliable outcome is a deterministic requirement, and you cannot build the second out of the first. So the agent needs a primary infrastructure for time to orchestrate against, exactly as it needs the CRM for customers and the ERP for money. Scheduling has to grow up into one. That is what a system of record for time is, and it is what we built our Temporal Grid to be.

Which quietly turns everything above from good hygiene into a hard requirement. When a person was doing the coordinating, the unaccountable cost was hidden but survivable, absorbed into a workflow and paid without complaint. When an agent is doing it, there is nowhere to hide it, because the agent has no colleague's intuition to fall back on and no memory of the social contract. Either the rules about whose time may be used, for what, and what beats what are captured somewhere it can read them, or it guesses. And it will guess confidently.

Three things.

If you run any process that depends on getting people into a room, ask for this cut of your own data. Not how many meetings you booked. How many you booked that never happened, who moved them, and where the cancellations cluster. If nobody in your organisation can produce that, then that absence is the finding.

When you judge software, stop judging the workflow and start judging the transaction underneath it. Your recruiting tool and your sales tool will each assure you they are doing fine, and within their own four walls they may well be. The quarter that never happens is hiding in the gap between them, in the one thing neither of them keeps a record of.

And if you are starting to hand these workflows to agents, ask what they defer to when the work comes down to time. If the honest answer is a raw calendar, you have not automated the coordination. You have just given the confident guessing to something that never gets embarrassed by getting it wrong.