THE PLAYBOOK FOR CUSTOMER SERVICE LEADERS
Forecasting, triggers, BPO contract, AI, peak SLAs and crisis messages. With ready-to-use templates.
Read the playbook ↓
CONTENTS
Before you start
1. Anticipate
1.1 Rebuild the last peak, hour by hour
1.2 Decide with a cool head: the triggers
1.3 The BPO contract: negotiate flexibility before you need it
1.4 A knowledge base ready for the peak
2. Absorb
2.1 Route each contact to the right skill
2.2 Avoid contacts that should never have existed
2.3 AI: what it really brings during a peak
3. Hold
3.1 Peak SLAs, announced and met
3.2 One dashboard, shared with your BPO
3.3 Organizing teams to last six weeks
3.4 Crisis messages ready to go
4. Learn
In conclusion
INTRODUCTION
This playbook is for customer service and operations leaders who handle large volumes, several channels and often one or more BPOs. It doesn’t replace your WFM tool. It frames the decisions your WFM won’t make for you: who triggers the reinforcement, on which contact reasons, and with what service level promised to the customer.
The examples come from customer services our teams run every day, in e-commerce, transport, insurance and travel. They are anonymized.
Two questions are enough to classify a peak: can you see it coming, and how long will it last?
The four phases below apply to all four cases. When an unplanned peak changes the method, an “If the peak isn’t planned” box spells it out.
Phase
When
The question to settle
Template provided
1. Anticipate
D-60 to D-7
How many contacts, and from when do we switch?
Capacity calculator, trigger matrix, contract checklist
2. Absorb
D-7 to the start of the peak
Who handles what, and which contacts can we avoid?
—
3. Hold
During the peak
What service level do we promise, and how do we steer it?
Peak SLA grid, crisis messages
4. Learn
D+15
What do we keep for the next one?
Post-mortem template
PHASE 1
D-60 to D-7
THE QUESTION TO SETTLE
How many contacts, and from when do we switch?
TEMPLATE PROVIDED
Capacity calculator, trigger matrix, contract checklist
By D-7, three decisions must be made and shared with your BPO: the hourly, per-channel volume forecast, the thresholds that trigger reinforcement, and the contract clauses that make it enforceable.
1.1
Rebuild the last peak, hour by hour
Working at the hour × channel × contact-reason level means having a volume figure for every combination of an hour, a channel (phone, email, chat) and a contact reason (order tracking, dispute, change of address…), rather than a single total per day. It prevents, for example, hourly peaks from disappearing in the estimates.
1
Extract the last comparable peak at this level of detail.
2
Remove what won’t recur (an incident, an isolated campaign) and add what is new this year: a newly opened channel, a broader scope, a different commercial offer.
3
Build three scenarios: low, median, high. Add to the high scenario a cushion calibrated on last year’s gap between forecast and actual.
4
Convert each scenario into FTEs per hourly time slot, counting reinforcements at their first-days productivity.
5
During the ramp-up, compare received vs. forecast every week, and share the gap with your BPO.
6
Set the date by which your BPO receives the final forecast. A forecast sent at the last minute leads to rushed schedules and uncovered slots.
SEEN IN THE FIELD
Forecast error cuts both ways. An insurer received between 62% and 77% of the forecast calls for several weeks in a row. Conversely, a gift-box brand handled 2,000 more tickets than forecast over its season. In both cases, weekly tracking of received vs. forecast made it possible to adjust in time.
Three-scenario capacity calculator
1.2
Decide with a cool head: the triggers
Write down in advance, for each signal, the threshold, the action, the person who decides and the lead time before the action takes effect.
Signal
Threshold (example to adapt)
Action
Who decides
Time to effect
Scheduled hours vs. need, at W+1, W+2 and W+3
Below 90% at W+2
Opening shifts, reaching out to agents, bonuses
Scheduling manager
1 to 2 weeks
Incoming volume
Median scenario exceeded by 20% two days in a row
Switch to the high-scenario schedule
Head of care
24 to 48 h
Forecast at D+3
Above planned capacity
Activating BPO reinforcement
Head of ops
Per contractual notice period
Email backlog age
Oldest ticket beyond 24 h
Overtime targeted at email
Scheduling manager
Immediate
Phone service level
Below target for two consecutive time slots
Shift multichannel agents to voice
Supervisor
Immediate
Escalation rate
Clearly rising over a day
Reinforce the control tower, review the knowledge base
Incident manager
A few hours
Two rules make this matrix useful:
Each signal has two levels: an alert threshold, where you prepare, and an action threshold, where you activate.
Leading signals matter more than the backlog, which only records what has already gone wrong. If your BPO needs 72 h notice, you need a signal that moves 72 h earlier: orders placed, site traffic, bookings, scheduled hours for the coming weeks.
SEEN IN THE FIELD
At one client, the administrative backlog climbed to 900 tasks. The issue had been building for three weeks, but no alert was tied to an action or a name. It took several weeks of daily reporting, agent bonuses and new hires to get back to zero.
The first signal rarely reaches customer service. It comes from technical monitoring, a carrier, a strike notice. Connect these sources to the customer service on-call rota. And when it’s your own company that triggers the peak, choose its timing: after a data leak, an insurer scheduled its communication in two waves, 10 am for corporate clients and 2 pm for policyholders. Customer service knew what time the volume would arrive.
Trigger matrix
1.3
The BPO contract: negotiate flexibility before you need it
A reinforcement that isn’t written into the contract arrives too late or costs too much. Points to check:
A volume floor and ceiling specific to the peak period, and the rule for exceeding the ceiling (written agreement, rate)
A reciprocal forecasting commitment. If you overestimate, you pay a minimum, for example 75% of the announced volume. If you underestimate beyond a threshold, your BPO does its best and the SLAs on the affected channels are relaxed. Everyone has an interest in an accurate forecast.
The ramp-up notice period, and the capacity guaranteed at that notice
The ramp-down notice period: what you pay if the peak fades faster than expected
The billing model during the peak. Billed per productive hour, you carry the productivity risk. Billed per interaction handled, your BPO carries it.
The training lead time for reinforcements, and who funds it
Shared indicators and reporting frequency during the peak
Named escalation contacts on both sides, reachable on key dates
A contract is rarely renegotiated a month before a peak. If you’re reading this too late for this year, at least get an amendment on the ramp-up notice period, and keep the rest for the next renewal.
SEEN IN THE FIELD
At the height of its season, a gift-box brand multiplied its billed volume by twelve in a single month, without exceeding its budget. That’s possible when the ceiling, notice period and billing were discussed beforehand.
1.4
A knowledge base ready for the peak
A procedure that has been out of date for a week produces dozens of errors. During a peak, they hit the reinforcements first, who don’t have the history to correct them on their own.
A dated “what changes during the peak” sheet, with its end date: offers and conditions, delivery or processing times, temporary commercial rules (extended returns, authorized goodwill gestures and their caps).
Macros for the 20 most frequent reasons from the last peak, reviewed by someone who handles tickets every day.
A single owner of updates for the whole period.
A freeze at D-7 on any non-critical change: tools, processes, content.
If an AI answers your customers or assists your agents, this discipline matters twice as much. It answers with what you give it: a contradictory page produces contradictory answers, across thousands of conversations.
PHASE 2
D-7 to the start of the peak
THE QUESTION TO SETTLE
Who handles what, and which contacts can we avoid?
This phase answers two questions: who handles what, and which contacts can be avoided.
2.1
Route each contact to the right skill
The right criterion isn’t internal or external, it’s the skill the contact requires. Give reinforcements, human or AI, the simple, high-volume reasons. For the rest, choose the team that masters the subject and has the authority to decide, whether it’s on your side or your partner’s.
Contact type
Examples
Handled by
Why
Simple and high-volume
Order or case tracking, change of address, question about an offer
BPO reinforcements or AI
Stable rules, easy to document
Simple but sensitive
High-value customer, goodwill gesture above a cap
Internal team, or partner with written delegation
The risk is to the relationship: who decides must be defined
Complex
Dispute, fraud, formal complaint, legal matter, specialized business question
Subject-matter experts, internal or placed by your partner
Expertise and accountability
New
Contact reason that appeared during the peak
Control tower, then documentation
You don’t delegate what you haven’t written down yet
Your most expert teams become the lookouts: they handle the sensitive cases and spot what is going wrong before it spreads. A partner that places experts from your field with you, payroll for example, can perfectly well handle level 2 contacts.
Prefer micro-certification by contact reason over full onboarding. A reinforcement certified on three reasons is productive within a few days. A micro-certification has four elements:
1
The reason’s sheet and its macros
2
A series of real cases already handled, with the right answer
3
A first batch of tickets reviewed by an expert before sending
4
A quality threshold to reach before opening the next reason
Finally, identify the few experts without whom a scope grinds to a halt, and secure their availability on key dates. On specialized scopes, two or three people often carry most of the knowledge.
2.2
Avoid contacts that should never have existed
Multichannel notifications sent at the right milestones cut “where is my order?” contacts by roughly 40%. The same reasoning applies to a case file, a refund or a train: you have the answer, the customer doesn’t.
SEEN IN THE FIELD
At one e-merchant, 8% of tickets came from customers chasing an order confirmation that never arrived or landed in spam. No agent needs to handle those contacts: the email sending has to be fixed.
1
List the steps in your journey that generate contacts. In e-commerce: confirmation, shipping, delay, delivery, return. In transport: delay, cancellation, missed connection. In energy: meter reading, bill, disconnection.
2
For each step, define the message, the channel and the system event that triggers it.
3
Start with anomalies. A delay announced before the customer discovers it avoids a contact, and often an unhappy customer.
4
Each message answers three questions: what is happening, do I need to do anything, when will I hear more.
5
Measure the effect: contact rate per order or per case, between notified and non-notified customers.
Prepare three elements that can be activated in under an hour: a banner on the site and app, a message at the top of the IVR, an automatic reply on email. During an outage, every avoided contact is a contact you won’t handle twice.
2.3
AI: what it really brings during a peak
AI plays two roles, which are not prepared in the same way.
Assist agents. Reply suggestions, history summaries, knowledge base search, ticket classification and prioritization. A Stanford/MIT study of more than 5,000 agents measures an average +15% in tickets resolved per hour with an AI reply-suggestion assistant, with larger gains for less experienced profiles (Brynjolfsson, Li and Raymond, “Generative AI at Work”). This effect mostly benefits reinforcements, because the assistant passes on your best agents’ answers and shortens their ramp-up.
Handle contacts end to end. On asynchronous channels (email, form, messaging) and simple reasons, an AI agent can handle up to 70% of written volume, day and night [source]. Allow one week in a test environment, then two weeks of dry run where the AI drafts and a human validates before sending. Before opening:
A clean, up-to-date knowledge base without contradictions
A closed list of reasons entrusted to the AI
Handover rules to a human: unhappy customer, high-value customer, amount above a threshold, second follow-up
An automation rate measured on your own historical tickets, not the rate advertised by a vendor
Three points to keep in mind:
Don’t cut human capacity in the same proportion. AI absorbs the simple first. What’s left for agents is longer and more sensitive, so average handling time goes up.
If the AI agent isn’t stable in production by D-15, wait until the peak is over. A rollout in the middle of the rush stacks two risks at the moment you have the least margin.
Humans and AI work from the same base and the same escalation rules. Otherwise, the customer receives two different answers depending on who replies.
PHASE 3
During the peak
THE QUESTION TO SETTLE
What service level do we promise, and how do we steer it?
TEMPLATE PROVIDED
Peak SLA grid, crisis messages
During the peak, four topics require a written decision: the promised service level, the indicators tracked with your BPO, how teams are organized, and crisis messages.
3.1
Peak SLAs, announced and met
Set peak-specific service levels, by channel and by contact reason, and have leadership approve them before the peak starts. A 48 h turnaround announced and met damages the relationship less than a 24 h promise that is missed.
Scope
Peak SLA (example to adapt)
How the customer knows
Payment disputes, fraud
Prioritized on all channels, handled the same day
Dedicated queue, immediate acknowledgment
Phone
Usual service level maintained
Wait time announced in the IVR
Chat and messaging
Reply within a few minutes
Queue position displayed
Email and web form
24 to 48 business hours
Automatic reply stating the real turnaround
High-value customers
Usual level
Dedicated queue or number
An honest automatic reply, for example:
AUTOMATIC REPLY
We are currently receiving far more messages than usual. We will reply within 48 business hours. To track your order without waiting: [link]. If your request concerns a payment, please write to us here instead: [link].
3.2
One dashboard, shared with your BPO
Your team and your BPO must read the same figures, calculated with the same definitions and the same source. Five indicators are enough:
Indicator
What it tells you
Frequency during the peak
Backlog by age (under 4 h, 4 to 24 h, over 24 h)
What will go wrong tomorrow
Every hour
Service level by channel, abandons included
Whether the promise is kept, and how many contacts are lost
Every hour
First-contact resolution
Whether we resolve, or generate follow-ups
Every day
Customer satisfaction by contact reason
Where quality drops
Every day
Escalation rate
Whether reinforcements are overwhelmed or the knowledge base is at fault
Every day
Read each indicator alongside the day’s volume. A 99% quality score over a quiet week proves nothing, and an average handling time must be interpreted reason by reason. Add a fifteen-minute daily check-in with your BPO at a fixed time.
3.3
Organizing teams to last six weeks
Secure availability on key dates: announce the peak calendar at least two months ahead, collect presence commitments and plan a counterpart (bonus, priority on following schedules).
Schedule in collective waves rather than case by case. Groups that start and finish together are easier to steer and readjust.
Set up a control tower: a few experts reachable live by all reinforcements, internal and BPO, on a single channel.
Name an incident manager per slot, with written decision-making power: send a crisis message, change an SLA, temporarily close a channel.
Keep a bonus mechanism that can be activated within 24 h to open extra slots when the backlog slips.
Reward quality, not just volume. A challenge based on the number of tickets closed lowers first-contact resolution, which in turn pushes volume back up.
Track fatigue with numbers: absences, satisfaction per agent, drifting handling time. Rotate the hardest positions, such as disputes, and protect breaks.
When the in-house site of a travel company went on strike, its external partner received 100% of calls overnight. Only a team already trained on the scope could absorb that shift. Entrusting part of your flows to a partner all year round is also a way to insure against this scenario. Also keep a named on-call list and a recall procedure: without them, the first quarter of an hour is lost looking for who can decide.
3.4
Crisis messages ready to go
Write the crisis messages and get them approved before the peak.
Scenario
What must be ready
Approved in advance by
Website or app outage
Banner, voice message, automatic reply, agent script
Communications, tech
Payment incident
Customer message, refund or goodwill-gesture guidelines
Finance, legal
Stock-out on a flagship product
Message to affected customers, alternative offered
Sales
Carrier delay or disruption
Proactive message, new estimated date, possible compensation
Logistics or operations
Strike action
Customer message, exchange or refund terms
Management, legal
Goal: a message sent within the hour after the decision.
Also prepare the macro for the question the event will trigger. The day after a strike that kept a news app from delivering a daily paper, escalated tickets were still asking why the paper was missing.
Crisis message library
PHASE 4
D+15
THE QUESTION TO SETTLE
What do we keep for the next one?
TEMPLATE PROVIDED
Post-mortem template
Hold the post-mortem fifteen days after the end of the peak: the data is consolidated and teams still remember what went wrong.
Block an hour and a half with planning, supervisors, your partners and an agent representative. Six questions:
1
Forecast vs. actual, by hour, channel and reason: where is the gap, and why?
2
Triggers: which ones fired in time, which too late, which never served?
3
Avoidable contacts: which reasons exploded when a proactive message would have avoided them?
4
Reinforcements: how many days before they were productive, and what quality by reason?
5
Customers: satisfaction and complaints during the peak, compared with the rest of the year.
6
Teams: fatigue, departures, and what they would do differently.
Leave the meeting with three decisions at most, each with an owner and a date. Update the capacity calculator and the trigger matrix while the numbers are fresh: they are what will save you time next time.
Post-mortem template
IN CONCLUSION
Key takeaways
1 · ANTICIPATE
Rebuild the last peak hour by hour, write down your triggers and negotiate contract flexibility before you need it.
2 · ABSORB
Route each contact to the right skill and avoid those that should never have existed.
3 · HOLD
Promise peak SLAs you can keep, run a dashboard shared with your BPO and keep your crisis messages ready.
4 · LEARN
At D+15, compare forecast and actual, keep three decisions at most and update your templates.