Coverage is a shape problem, not a headcount problem
Most schedules fail for a reason that has nothing to do with how many people you have. They fail because the shape of the supply does not match the shape of the demand. Fourteen agents all starting at 09:00 will drown at 09:30 and idle at 16:00, while the same fourteen agents staggered across the morning may hold service level all day.
So the sequence matters. Requirement first, then shift shapes, then break placement, then publication. Skip a step and you end up hand patching a roster forever.
Step one, build the interval requirement
Run Erlang against each interval separately, using forecast volume and AHT for that interval. Do not staff to a daily average, ever. A daily average will overstaff your quiet evening and underrun your 11:00 peak, and the two errors do not cancel out because service level is not recoverable backwards.
Take the requirement before shrinkage for coverage matching. Shrinkage belongs in headcount planning, not in the placement of the people who are actually rostered on that day.
Step two, choose shift lengths honestly
The natural instinct is one uniform shift length because it is fair and simple. Sometimes it is also unworkable.
A single nine hour shift cannot follow a sharp morning peak, because every agent you add to the peak also adds coverage to an afternoon you did not need it in. If your demand curve has a strong single hump, you need staggered starts. If it has two peaks, you probably need two shift lengths, or split shifts if your labour rules and your people permit them.
Practical guidance:
- Stagger starts in 30 minute increments where you can, 15 minutes if your requirement is very peaked. Wider increments are easier to publish and remember, but they cost precision.
- Shorter shifts, six or seven hours, are your surgical tool for peaks. A handful of part time shifts often solves a coverage problem that twenty full time shifts cannot.
- Keep the number of distinct shift patterns small. Every extra pattern is a real cost in supervision, communication and mistakes.
Step three, place breaks like they matter
Breaks are not administrative detail, they are 60 to 90 minutes of missing capacity per agent, concentrated in the middle of the day if you let it happen.
Rules that hold up in real operations:
- No break in the first hour or the last hour of a shift.
- Lunch in the middle portion of the shift, with a short break on each side of it.
- Never let lunch coverage drop below the interval requirement. This is the single most common cause of a 13:00 service level dip.
- Place breaks on a 15 minute grid so a real time analyst can actually action them.
- Stagger lunches across at least ninety minutes. If everyone eats at 13:00, you have designed a daily outage.
Two fifteen minute breaks and a one hour lunch on a nine hour shift is 90 minutes off the phone, roughly 17% break shrinkage before a single absence, meeting or coaching session. That is why planned headcount always looks so much bigger than the Erlang requirement, and why break placement moves service level as much as headcount does.
Step four, check the schedule, not the plan
Once the roster exists, rerun Erlang against the agents you have actually scheduled, interval by interval. Then look at three things.
The worst interval. One interval two agents short costs more service level than ten intervals with a rounding gap. Fix the worst interval first.
Occupancy. Anything sustained above 90% will come back as attrition rather than efficiency. A schedule that hits service level by grinding people is not a schedule that survives the quarter.
Surplus and where it sits. Late day surplus is normal and useful, it is your coaching and offline window. Surplus in the peak is a genuine error.
When to stop moving people around
Some coverage gaps cannot be closed by rearrangement, and recognising that quickly saves days of futile effort. If the optimiser, or you, cannot close coverage while respecting break rules, the constraint is structural: shift length, start increments, headcount, or opening hours. Escalate it as a structural request with numbers attached, rather than presenting a broken roster as the best available.
Publish like an operator
A schedule is only as good as its adoption. Publish early, keep patterns stable week to week where you can, and be explicit about what changed and why. Agents forgive an inconvenient shift far more readily than a shift that appears without warning. Fairness in rotation, especially of unpopular starts and weekend days, buys you more compliance than any amount of optimisation.
The tools on this site will get you the requirement and a solid first pass roster in minutes. The judgment about which trade offs your operation can live with, that part is still yours.