Groundwork
The groundwork.
The essays argue. This page is what they stand on: the numbers a contact center runs on, what each one is for, what right looks like to me, and the misses I keep finding.
The system
One picture, six stages.
Everything on this page hangs off one chain. Demand becomes a forecast, the forecast a requirement, the requirement a schedule, the schedule a day, and the day a read. Each stage can lose the day on its own, and each loses it differently, which is why naming the stage matters more than naming the number.
The chain also draws a fence line. Demand is the weather: it arrives from outside, it does not care what the plan said, and it is the only stage in the picture the operation does not own, though some of the rain is the operation’s own sprinkler: failure demand, John Seddon’s name for the repeat volume a center manufactures for itself when resolution fails. The forecast is a reading of that weather, and readings miss. You do not curse the sky when the forecast is wrong. You get an umbrella, and the umbrella is bought before the rain: the funded tolerance, the intraday levers, the flexibility priced into the plan instead of improvised past it.
Everything to the right of the forecast lives inside the fence, and the instruments swap there: an operation forecasts what it cannot control and diagnoses what it can. Most operating confusion is a filing error across that line: cursing the sky over a misread forecast, or forecasting attrition as though nobody owned its inputs.
The numbers
What each one is for.
Six numbers run most contact centers. None of them is the truth. Each one points at something real, and each one creates a specific trade the moment it carries weight. The pattern below repeats deliberately: the definition, the trade the number creates, the reason it exists, and the paired truth that has to be read beside it. Keeping the trade welded to the word is the discipline, argued at essay length in Read the trade, not the target.
Service level
The share of contacts answered within a stated wait: an 80/20 target means eight in ten answered inside twenty seconds. Whether the denominator counts offered contacts or only handled ones changes the number, and abandonment is the number the threshold hides.
- The trade
- Speed, bought with the conditions underneath it: staffing cost, recovery room, the coaching hour, the patience a hard problem needs. Chased interval by interval, it buys green intervals with canceled coaching and an occupancy debt the report never shows.
- The why
- The promise made about the customer’s wait (waiting is real) and the number the staffing model is solved against.
- Paired with
- What the operation bent to hit it: the trade is the story, not the number.
Occupancy
The share of staffed, ready-or-handling time spent handling contacts rather than waiting for one.
- The trade
- Utilization against recovery. Queues are nonlinear: push occupancy toward its ceiling and the wait explodes, the room to reset disappears, and the operation files the consequences under attrition.
- The why
- Capacity discipline: how hot the floor is running, and how much slack the queue, and the people, have left to absorb a surprise.
- Paired with
- The recovery the number does not show: what running people hot for a quarter does to the people.
Shrinkage
Paid time not available to handle contacts: leave and absence, training and coaching, meetings, and the parts of the day that simply go missing.
- The trade
- Coverage today against capability tomorrow. Collapsed into one percentage, absence and investment get cut with the same knife. And the visible off-phone work, the coaching block and the one-to-one, goes first.
- The why
- The honest distance between paid hours and available hours, priced into the requirement before the schedule is cut.
- Paired with
- Its composition: which kinds, chosen or endured. PTO is not the same as coaching.
Adherence
How closely the day followed the plan: the schedule said a person would be somewhere, and adherence reports whether the day agreed.
- The trade
- Reliability against context. A schedule that is optional cannot plan a floor. A deviation read as a character flaw stops being information, and once the number becomes a verdict, the operation itself starts editing the plan until it behaves.
- The why
- A reading on the pair (plan and day together) and the cleanest evidence available for telling a wrong plan from a sideways day.
- Paired with
- The plan’s own funded assumption. An agent on the assumption is on plan.
Average handle time
The average time a contact takes end to end: talk, hold, and the after-contact work, with the wait in queue counted elsewhere.
- The trade
- Throughput against resolution. Read as a speed target it teaches people to end contacts. Blended across a changing mix it moves on its own: remove the easy work and the average rises with nobody slower.
- The why
- The weight of the work: the denominator every capacity calculation stands on.
- Paired with
- The mix underneath it, and whether the contact was actually resolved.
First-contact resolution
The share of issues resolved in one touch, with no repeat on the same issue inside a stated window, so the issue-matching rule and the window’s length are part of the definition.
- The trade
- Today’s efficiency against tomorrow’s volume: a contact closed without being resolved buys a shorter call now and another contact later. And a baseline cut against the old queue turns automation’s mix shift into someone’s underperformance.
- The why
- Whether resolution actually happened: the outcome read, where handle time is only the effort read.
- Paired with
- Repeat volume: the demand a center manufactures when resolution fails and the forecast then mistakes for growth.
Two of those claims deserve to be handled rather than read. The first is the nonlinearity, and the cliff sits at the offered load, the erlangs of work the interval brings in. At or below that load the queue cannot stabilize. Just above it, one agent moves service level by double digits. Past the requirement, the next agent buys very little. “Just add one person” is exactly right on one side of that line and a waste of a salary on the other, and occupancy has to be read off the same curve.
Simulated values: not benchmarks, not data from any operation. Everything runs in your browser. Open the full instrument.
The second is the pairing. A managed number can improve for a full quarter while the truth paired with it degrades, no dishonesty required, only pressure. Watch it happen:
And because shrinkage and adherence get confused for each other more than any other pair on this page: they live on the same interval grain, and they are still two different readings. Planned shrinkage is scheduled capacity spent off the phones: funded, budgeted, accounted for. Adherence is a reading of how closely the day followed the plan. Nothing on this page charts them against each other, because the join is what everyone assumes and no join is claimed. The operator sheet on the WFM page shows the honest layout instead: one interval row carrying the scheduled headcount split into its activity codes, adherence read beside them as its own column, and the origin of every variance named, so neither reading gets blamed by adjacency.
The standards
What right looks like.
A contact center is closer to a garden than a machine. A machine turns inputs into outputs on demand. A garden turns conditions into outcomes on its own schedule, and almost nothing in it fails where it was caused. The weather happens to both, but the difference between weather and a garden is a gardener.
So right, in a system like this, means a set of conditions kept true rather than a set of targets hit. None of them is a benchmark (every number above gets calibrated to the operation that owns it), and these are the five I keep.
The tolerance is funded and felt
Every competent plan carries an adherence assumption: a funded allowance for the day’s ordinary friction. An agent on the assumption is the plan working as funded, and every surface showing the number should say so. A garden planned to the millimeter dies of its plan. So does a schedule.
Development is budgeted demand
Coaching, training, and one-to-ones belong in the staffing model as demand, not as whatever survives the day. And funding cuts both ways: once the time is protected, the operation is accountable for how it is spent. This is feeding the soil: the investment that looks, on any single day, like nothing grew.
One data grain
Forecast, schedule, and actuals arrive from different systems on different clocks. Nothing compares to anything else until all of them are carried to the same interval grain. Skip that and the analysis is not wrong so much as meaningless. There is no metaphor for this one. It is plumbing, and it comes first.
Attribution before action
Name which gap caused the miss before pulling a lever. Treat an execution problem with a bigger forecast and the operation buys headcount while keeping the miss: water a nitrogen problem and you drown the plant with the problem intact.
The plan loses informatively
The plan is a hypothesis, and it was never supposed to win its argument with the day. No garden ever followed its seed packet, and the difference is what the gardener learns. Fix the actuals whenever they lie. Leave the plan as the day found it. The variance between them is the material foresight is made of.
The common misses
Where operations go wrong, and where each one is taken apart.
The same handful of misreadings shows up in operation after operation, and none of them requires anyone to be bad at their job: most are reasonable reactions to pressure, repeated until they become the furniture. Each one below is developed at essay length. This is the map.
The scheduling department
WFM read as the team that makes the rosters, when the job is strategic inventory management and the inventory is human attention.
Taken apart in WFM as strategic inventory management.
Composition read as performance
Automation takes the easy contacts first, so every blended number moves on its own (handle time up, resolution down) with nobody slower. Read as performance, that arithmetic looks like decline.
Taken apart in The measurement trap.
Adherence pointed at people
A reading on how closely the day followed the plan, used instead as a conformance score on individuals: checked against the clock, graded against an imaginary hundred the plan never asked for.
Taken apart in Adherence is a reading, not a leash.
The schedule settled quiet
Editing the plan after the fact so it matches what occurred. The number walks toward a hundred while nothing about the day improves: the one metric the house games, and the operation’s memory is the price.
Taken apart in The one metric the house games.
Shrinkage collapsed to one number
Absence and investment carried in the same percentage, so the operation cannot tell them apart under pressure, and cuts the coaching block first because it is the only thing a customer will not notice today. The bill arrives two quarters later as attrition.
Taken apart in The hidden cost of attrition.
The wrong gap treated
A missed day answered with whichever lever is nearest instead of the one its origin calls for. Forecasting, planning, and execution gaps need different owners and different fixes, and only attribution can tell them apart.
Taken apart in The diagnostic on the WFM page.
Where next
The argument this page supports.
The essays develop each miss at full length, and the WFM page carries the position all of this groundwork exists to hold up: the forecast is the contract with the floor.