Contact center & customer service platforms

The licence is what separates a phone system from a contact centre

Per-seat phone licensing assumes one person, one number, one conversation at a time. Contact centre licensing assumes a queue: calls waiting for the next available agent, routing decisions based on skill or priority, supervisors watching service levels in real time, and reporting that survives an audit. Those capabilities are what the additional licence buys, and they are usually sold per agent rather than per person. A business can therefore run a contact centre on a phone platform with a contact centre licence attached, or on a separate platform entirely, and the difference in cost and complexity between those two routes is what the evaluation should actually be about.

• Queues, skills-based routing and priority rules.
• Supervisor view: live service level, agents, waiting calls.
• Reporting that holds up when someone asks how service performed.
• Channels beyond voice: chat, email, social, messaging.

Concurrency arithmetic decides the licence count

Where concurrent licensing is offered, the number that matters is the peak simultaneous agent count — not the number of people who work on the team. A team of thirty across three non-overlapping shifts may never exceed ten agents at once. A team of ten that all start at nine o'clock will hit ten every morning. That is why the practical question is whether shifts genuinely do not overlap, including handover periods, training, team meetings and overtime, because a single hour of overlap sets the peak for that hour and the licence count for the month. Measure the peak from your own data rather than from an assumption: the busiest hour of the busiest week is a defensible starting point, and the busiest hour of the year is often the honest one.

• Peak simultaneous agents, measured rather than estimated.
• Shift overlap, handovers, training and overtime included in the count.
• The busiest hour of the busiest week, checked against the busiest week of the year.
• Who monitors the peak after the contract starts.

Service level is a definition before it is a number

'Service level' is a fraction with a decision hidden inside it: contacts answered within a stated threshold, divided by contacts offered, expressed as a percentage. The common shorthand — 80/20 or 80/30, meaning eighty per cent answered within twenty or thirty seconds — is a convention rather than a rule. Because the definition carries the number, a quoted service level means little until you know whether the denominator counts contacts offered or only those answered, whether very short abandons are excluded and after how long, and whether callbacks are counted. The companion measures share that property. Average speed of answer is the mean wait of the calls that were answered, so it drifts better whenever impatient callers hang up; abandonment rate supplies the counterweight. Occupancy is productive handling time divided by productive handling time plus available time, and above roughly eighty-five to ninety per cent there is almost no slack left to absorb a burst, which is why waiting time climbs far faster than demand as an operation approaches its ceiling. COPC's CX Standard is the operational reference that defines these terms, and a service metric published without its definition is the most common way a report misleads.

• State the numerator, the denominator, the interval and the abandonment treatment.
• Average speed of answer improves when callers give up — read it with abandonment.
• Above roughly 85–90 per cent occupancy, wait grows faster than demand.
• 80/20 and 80/30 are conventions, not requirements — check the one being quoted.

Erlang C, offered load and what shrinkage does to the schedule

Sizing a queue is arithmetic on four figures, and the platform decides none of them. Forecast the contact volume for each planning interval — fifteen or thirty minutes is conventional — and the average handling time that includes talk, hold and after-call work as the operation defines it. Multiply the two to get the offered load in Erlangs, where one Erlang is the traffic of one continuously busy agent. Feed that, the service-level target and the answer threshold into a queueing model such as Erlang C, which returns the number of agents who must be ready to take a contact. That figure is net, so convert it to scheduled people by dividing out shrinkage: if an interval needs eighty available agents and forecast shrinkage is a quarter, the schedule must carry about a hundred and seven. Erlang C assumes waiting contacts do not abandon; where abandonment is material, the Erlang A extension adds a patience assumption and gives a more honest answer. The discipline is to forecast, staff the intervals that need it rather than the day as a whole, and then compare actual arrivals against the plan, because the model is only as good as the volume and handling-time figures feeding it.

• Offered load in Erlangs is contact volume times handling time.
• The model returns available agents; shrinkage converts that to scheduled people.
• Erlang C assumes nobody abandons; use the abandonment-aware extension where they do.
• Plan at the interval, then re-forecast within it as arrivals diverge.

Payment card data in the call path

If agents take card payments over the phone, the call path becomes part of your payment-card environment and the obligations that come with it. The standard question is whether card data is in scope at all: a platform that lets the agent pause recording, that masks the tones as they are typed, or that hands the caller to a keypad or a payment link so the agent never hears the number, will reduce scope for the business. A platform that simply records everything will expand it, because the recording becomes another place card data exists — with its own storage, access and retention duties. Decide which of those models you are buying before you compare feature lists, because it changes both the configuration and the audit burden.

• Whether card data ever reaches the agent or the recording.
• Pause, masking, keypad capture or payment link — which is offered.
• Where recordings containing payment data would be stored.
• Who has access to them, and for how long.

Recording, consent and retention

Recording a call is two decisions: whether you may, and how long you keep it. Whether you may depends on the law in the states you call into, because some require the consent of all parties rather than one, and the platform's announcement at the start of the call is the usual way that consent is obtained. How long you keep it is a records decision, and it should be written before the platform is configured, since retention settings applied during setup tend to become permanent. Ask who can retrieve a recording, whether retrieval is logged, what happens to recordings when an agent leaves, and how recordings that are no longer needed are destroyed. A recording you cannot retrieve is a liability without a purpose.

• The consent rule in each state you take calls from.
• How consent is obtained and evidenced on the recording.
• A retention period tied to a records policy, not a default.
• Logged access, and a destruction process that actually runs.

Outbound calling carries its own set of duties

Outbound campaigns are governed by consumer-protection rules that apply to the caller, not to the platform hosting the calls. The duties include respecting do-not-call obligations, honouring permitted calling hours, identifying the caller accurately, and keeping abandonment of connected calls within stated limits. Under the FCC's rules at 47 CFR 64.1200 those limits are specific: solicitation to a residence is confined to the hours between eight in the morning and nine at night local time at the called party, and no more than three per cent of live-answered telemarketing calls may be abandoned across a thirty-day campaign, with a call treated as abandoned when it is not connected to a live representative within two seconds of the greeting. A company-specific do-not-call request must be honoured within a reasonable time, and the FCC's revocation rules allow no more than ten business days to act on a revocation of consent. Caller identification must be transmitted, and a prerecorded message must give the business's name and number and a way to opt out. None of those duties transfers to a software vendor because the software makes the call.

• Suppression lists honoured before dialling, not after.
• Calling windows and the abandoned-call ceiling enforced by the system.
• Accurate caller identity on every outbound call, transmitted not just displayed.
• Reports that can evidence compliance if challenged.

Emergency calling when the phone system is an MLTS

An office phone system that serves many users is a multi-line telephone system under federal rules, and those rules carry two duties a contact-centre rollout must configure rather than assume. Kari's Law requires the system to let a user dial 911 directly, without an access prefix such as a leading nine, and to notify a central point at the facility when 911 is dialled wherever it can do so without a hardware or software upgrade. Section 506 of the RAY BAUM'S Act requires the call to carry a dispatchable location — the validated street address plus the floor, suite or room detail a responder needs — and the FCC's rules at 47 CFR 9.16 are keyed to the actual location of the device in use, not the company's headquarters or the carrier's billing address. Remote and mobile softphones are the hard part: a user answering from home or between sites must still be located accurately, so the device and location inventory becomes something you maintain rather than a setting installed once. The operator cannot assume the carrier's general emergency service satisfies the rule; manufacturer, seller, installer and operator each carry distinct responsibilities.

• Direct 911 with no prefix — Kari's Law binds the system, not only the carrier.
• Send a dispatchable location: address plus floor, suite or room.
• Keep a maintained device-and-location inventory for remote and mobile users.
• Location follows the device in use, not headquarters or the billing address.

What to gather before asking for pricing

Contact centre quotes depend on numbers, volumes and channels, so bring all three. The agent count at peak rather than the headcount, the call and message volumes by hour and day, the average handling time, the channels you intend to run, and the systems the platform must integrate with — customer records, ticketing, payments, recording archive. Add the service-level target and the abandonment treatment you will report against, since those set the staffing the platform has to support. Then add the obligations that shape the configuration: whether card payments are taken on the phone, whether calls are recorded and why, whether outbound campaigns run from the same platform, and how emergency locations are kept current for remote staff. Those details change the licence tier, the integration work and the compliance questions more than the choice of vendor does.

• Peak simultaneous agents and total team size, both.
• Volumes by hour, day and channel, with average handling time.
• The service-level target and how abandonment is treated.
• The systems that must integrate, named.
• Recording, payment, outbound and emergency-location obligations.

PUT THE OPTIONS SIDE BY SIDE

Your comparison checklist

Common questions