WFM guideChannel planning

Multichannel contact centre WFM

Q: What is the difference between multichannel and omnichannel contact centres?

Multichannel means a customer can contact you by multiple channels (phone, email, chat, social). Each channel is typically handled separately, with different agents, different queues, and different SLA targets. Data from one channel interaction may not be visible to agents handling a follow-up on another channel. Omnichannel means the channels are integrated: a customer who started a chat conversation can escalate to voice, and the agent can see the full chat history in context. The CRM, case management, and routing systems share state across channels. Omnichannel is the aspiration; multichannel is the operational reality for most contact centres. The distinction matters for WFM because omnichannel routing complexity changes how you model agent supply: an agent handling a voice call is simultaneously unavailable for chat, and the routing engine must account for this in real time.

Q: Why can't you use Erlang C for all contact centre channels?

Erlang C models an M/M/N queue: random Poisson arrivals, exponential service times, N parallel servers, no abandonment. This accurately models a voice queue where each agent handles one call at a time and contacts queue until served. It breaks down for: (1) Live chat, where agents handle multiple concurrent sessions; the correct model is Little's Law for concurrency. (2) Email and tickets, where contacts have an SLA measured in hours or days, not seconds; the correct model is a backlog flow model (drain rate vs. arrival rate). (3) Outbound, where agents initiate contacts; the correct model is a target RPC (right-party contact) model based on dial rate and connect probability. Applying Erlang C to chat systematically understaffs (it assumes 1:1 agent:contact); applying it to email produces meaningless results because the time-to-serve assumption is wrong.

Q: How does channel mix affect contact centre headcount?

Channel mix affects headcount through two mechanisms. First, model selection: different channels require different staffing models, each with different efficiency characteristics. Chat is more efficient per agent than voice because concurrency allows one agent to handle 2–4 sessions. Email is the most efficient for high-AHT work because agents batch and there is no queue impatience. Voice is least efficient per-agent for high-AHT interactions because Erlang C's square root staffing law means adding a few agents produces a large SL improvement, but occupancy constraints are tighter. Second, channel shift cost: when customers cannot resolve their query in their first-choice channel and must switch (e.g., chat escalates to voice, email receives a confusing reply requiring a callback), the second contact has a higher AHT and higher emotional intensity. Channel-shift contacts are a CES (Customer Effort Score) driver and inflate overall volume without being captured in single-channel forecasts.

When a contact centre handles more than one channel, its WFM complexity increases non-linearly. Each channel has a different staffing model, a different service level metric, and a different efficiency profile. Planning them with a single shared headcount pool, or worse, with Erlang C for everything, produces systematic understaffing and invisible overload.

Multichannel vs. omnichannel: the practical distinction

Multichannel

Customers can reach you by multiple channels. Each channel operates independently: separate queues, separate agent teams, separate SLA metrics. Agent handling a chat cannot see the customer's last phone call.

·Easier to staff (channels staffed independently)
·WFM models run per-channel
·Risk: customer repeats same information on each channel
·Reality for most contact centres today

Omnichannel

Channels share context. CRM, routing engine, and case management show the full cross-channel journey. A customer who chatted yesterday has that context visible to the agent who answers the phone today.

·Complex to staff (agent availability spans channels)
·Routing must account for cross-channel agent state
·Advantage: reduces channel-switch repetition
·Aspiration for most; expensive to implement fully

WFM implication: In a true omnichannel environment, when an agent accepts a voice call they are simultaneously removed from the chat queue, and the real-time routing engine must account for this across all channels. This requires routing system integration with the WFM real-time adherence feed, a technical dependency that most WFM tools handle imperfectly. In practice, most omnichannel operations still plan WFM by channel and rely on the routing engine for real-time blending.

One model per channel: why it matters

The most common planning error in multichannel contact centres is applying Erlang C to every channel. Erlang C is correct for voice (one agent, one call, immediate queue). It systematically over-staffs email and under-staffs chat if applied directly.

📞

Inbound voice

Erlang C (M/M/N queue)

Key assumption

One agent per call; contacts queue until served; random Poisson arrivals

AHT range

4–12 min (complex: up to 25 min)

Typical SL target

80% in 20s typical; 90% in 15s for regulated

Staffing note

Square root staffing law: each incremental agent above minimum produces large SL improvement. Occupancy constrained to 80–85%.

💬

Live chat

Concurrency model (Little's Law)

Key assumption

Agents handle 2–4 concurrent sessions; interstitial time between customer responses available

AHT range

8–20 min elapsed; 3–7 min handling time

Typical SL target

90% within 30s first response; 90% within 60s typical

Staffing note

More efficient than voice per agent at low–medium occupancy. Efficiency collapses above 88% voice occupancy if blended.

✉️

Email / tickets

Backlog flow model

Key assumption

SLA measured in hours/days; contacts batch into a backlog; drain rate vs. arrival rate determines queue evolution

AHT range

8–20 min per reply (complex: 30+ min)

Typical SL target

80–95% within 24h or 5 business days (sector-dependent)

Staffing note

No Erlang C minimum, since there is no real-time queue. Under-staffing compounds as backlog; clearing a compounded backlog requires surge capacity.

📱

Social media (DM / comments)

Hybrid: real-time DMs → concurrency; public comments → backlog

Key assumption

Public comments are asynchronous; DMs may require near-real-time response; reputational urgency overrides SLA formulas

AHT range

5–15 min per post/DM

Typical SL target

DM: 60–120 min; public complaint comment: 30–60 min (reputational)

Staffing note

Volume is spiky and event-driven (product launch, service failure). Do not blend with voice without clear routing rules.

📤

Outbound (dialler)

Target RPC model

Key assumption

Agent productivity = dial rate × connect rate × right-party rate × conversion rate

AHT range

3–8 min per connected call

Typical SL target

Defined as contacts per agent per hour, not SLA. Ofcom: ≤3% abandoned calls

Staffing note

Blending inbound and outbound is complex: inbound must take priority, but outbound occupancy disappears when inbound spikes.

Channel-switching cost: the invisible volume driver

What happens when a customer switches channel

1. Initial channel contact

Customer contacts via preferred channel (chat). Partial resolution or confusing response. Issue not resolved.

2. Channel switch

Customer escalates to voice. Customer repeats the entire issue, because chat history is not visible to the agent. AHT of the voice contact is 40–60% higher than a native voice contact because of the recap time.

3. WFM impact

The voice queue receives a higher-AHT, higher-emotion contact not captured in the standard voice forecast. Forecast accuracy drops. The chat contact is already counted in the chat volume report, creating double-counting of demand.

Channel-switch contacts have 35–55% higher AHT

A voice contact from a customer who already tried chat for the same issue requires recap time, has higher emotional intensity, and is more likely to require ACW to document the complaint. This uplift is not captured in standard voice AHT figures because ACD does not tag the reason for the longer call.

Channel-switch rate is a leading indicator of deflection failure

If 15% of voice contacts begin with "I already tried chat/email about this", your deflection strategy is not working: customers are using the lower-cost channel first and escalating when it fails, not substituting. A successful deflection reduces overall contact volume; a failed one adds volume on both channels.

CES is the metric that captures channel-switch pain

Customer Effort Score deteriorates when customers must contact you more than once or switch channels. A high CES alongside stable CSAT indicates customers are reaching resolution but only after significant effort, a pattern that drives eventual churn without flagging in satisfaction metrics.

Running WFM across channels

The technically correct approach to multichannel WFM is to run the right model per channel and sum the results into a total headcount requirement:

Forecast channel volumes independently

Voice, chat, email, and social have different drivers and patterns. Chat volume does not follow voice seasonality exactly; it is shaped by digital product launches, app problems, and digital-first customer demographics. Forecast each channel separately using its own historical data.

Apply the correct model per channel

Voice → Erlang C. Chat → concurrency (Little's Law). Email/tickets → backlog flow model. Do not apply Erlang C to all channels and adjust by a fudge factor; the mathematical foundation is wrong for non-voice channels.

Calculate channel-specific headcount

Each channel output is a minimum seated requirement for the interval. Add per-channel shrinkage to convert to scheduled headcount. Note: blended agents who handle multiple channels reduce total headcount vs. siloed staffing, but only if the occupancy across channels is compatible.

Check blended occupancy

If agents handle both voice and chat, the combined occupancy (voice AHT × voice rate + chat handling time × chat rate) / available time must not exceed 85–88%. ACD only reports voice occupancy; the chat load is invisible to real-time monitoring without WFM integration.

Report by channel and combined

Service level reporting must distinguish performance by channel. A combined "SL" figure that averages voice and email is meaningless: voice is real-time, email is asynchronous. Report each against its own target, and separately report against any combined SLA if contractually required.

Multichannel WFM questions

What is the difference between multichannel and omnichannel contact centres?

Multichannel: customers can contact via multiple channels but each channel operates independently with separate queues, agent pools, and data. Omnichannel: channels share context, with the full customer journey visible to any agent on any channel. Most contact centres are multichannel in practice; omnichannel is the aspiration. WFM is more complex in omnichannel because agent state must be tracked across channels simultaneously.

Why can't you use Erlang C for all contact centre channels?

Erlang C models a single-server queue (one agent, one contact at a time). Chat agents handle 2–4 concurrent sessions, so the concurrency model (Little's Law) applies. Email contacts have SLAs measured in hours/days and enter a backlog, not a real-time queue, so the backlog flow model applies. Applying Erlang C to chat under-staffs (assumes 1:1); applying it to email produces meaningless results.

How does channel mix affect contact centre headcount?

Two mechanisms. First, model efficiency: chat concurrency means each agent handles more contacts, and email batching allows higher productivity per FTE. Voice has the lowest contacts-per-agent-per-hour due to Erlang C occupancy constraints. Second, channel-switch inflation: failed digital deflection adds higher-AHT repeat contacts to the voice queue that weren't forecast in channel-specific models.