Service Level During Work Migration

Service level through a simulated work migration (synthetic data, shared pool). Source-site service holds through most of the consultation period, then falls from around the announcement to a trough during the transfer waves; the destination opens below the service floor and recovers as it ramps. Lines are weekly medians and bands the 10th–90th percentiles across 2,000 simulated futures; neither is a single trajectory.

Service level during work migration concerns the service a customer operation delivers while its work moves from one site, owner or workforce to another, as in a site exit, a consolidation, an outsourcing or a transfer to a partner. The planning artefact usually produced for such a program is an end-state reconciliation: starting headcount, less attrition, transfers and releases, equals the population at the finish. That reconciliation can be correct while saying nothing about the weeks between the decision and the last cutover, which is where service losses tend to arise. During those weeks supply and demand leave the source site on different schedules. Supply leaves through attrition that often cannot be replaced; demand leaves in steps set by client contracts and transfer waves. Because queueing systems respond non-linearly to staffing, a modest gap between the two schedules can produce a disproportionate fall in service level. The situation corresponds mainly to the ownership-or-location change in Migration Archetypes, where customer demand is unchanged but the workforce serving it moves.

The end-state view and the transition view

A migration program is commonly summarised as a headcount waterfall: the population at the start, the number expected to leave by attrition, the number transferring with their work, the number released, and the number remaining. The waterfall answers where the program ends up. It is a statement about stocks at two points in time.

Service level is a statement about flows in every interval between those points. It depends on how much work remains at the source site in a given week and how many productive hours are available to serve it that week. Two programs with identical waterfalls can deliver very different service during the transition, depending on when the attrition occurs relative to the work leaving. For that reason the transition view needs its own model, reconciled to the waterfall but not derived from it.

Why service falls during the transition

Supply leaves before demand

In many jurisdictions an employer planning collective dismissals must inform and consult employee representatives before the dismissals take effect. In the European Union, Council Directive 98/59/EC requires consultation "in good time with a view to reaching an agreement", and projected collective redundancies take effect no earlier than 30 days after the competent public authority is notified. The Directive sets a minimum: member states may apply provisions more favourable to workers, and national implementing law, collective agreements and works-council arrangements can extend the effective period considerably. Where a transfer of an undertaking is involved, separate provisions on employee rights apply.

For planning purposes the practical effect is a period, referred to in this article as a consultation freeze (a descriptive label rather than an established term), during which, depending on national law, collective agreements and the employer's own practice, a program commonly defers announcing its plans to staff, moving work and replacing leavers. The freeze is a constraint on the source site's supply, not on its demand. Customers continue to contact the operation at the usual rate while the workforce serving them shrinks through ordinary attrition.

Attrition rarely stays at its ordinary rate. Downsizing is associated with higher voluntary turnover among remaining employees, and with higher sickness absence among those who stay. A migration program can therefore expect attrition and absence above their ordinary rates once the downsizing is known, which may coincide with the period in which replacement is constrained.

Demand leaves on contract dates

Work leaves the source site in steps rather than continuously. Some clients transfer with their work at a planned wave. Some leave the organisation altogether at the end of a fixed-term contract, or after a notice period if the contract is rolling or contains a termination-for-convenience clause. Some are moved to a different internal platform. The timing of each step is uncertain, and the steps that most relieve the source site (transfers) are often the ones the consultation freeze delays.

Transition work competes for the same hours

Staff who will transfer, or who will hand over work to a receiving team, usually need training in the receiving site's systems and processes, knowledge-transfer sessions and shadowing. These activities draw on the same staff, in the weeks immediately before each cutover, when the source site is at its thinnest. After cutover, the receiving team handles work more slowly for a period while it builds proficiency, a pattern described by the speed-to-proficiency curve. Research on organisational learning finds that knowledge acquired through experience transfers incompletely even between shifts of the same plant, which suggests larger losses when work moves between organisations or sites.

Queues respond non-linearly

Service level in a staffed queue is not proportional to staffing. Under the Erlang C model a pool operating at high occupancy loses service level much faster than it loses agents: a small reduction in productive hours without a matching reduction in demand can take a pool from comfortably above its target to far below it. The effect is described in Erlang Sensitivity and the Staffing Cliff.

The same arithmetic applies to the receiving site. Larger pools achieve a given service level at higher occupancy, because the safety staffing a queue requires grows roughly with the square root of its load rather than in proportion to it. A migration plan that assumes a productivity "synergy" by transferring the work with proportionally fewer staff at proportionally faster handling times holds occupancy constant but reduces the size of the pool. At the same occupancy a smaller pool delivers a lower service level, so a synergy that is fully realised in productivity terms can still cost service. The shorter handling time makes a fixed answer-time threshold longer relative to each call, but under Erlang C this exactly cancels the smaller spare capacity (staff minus offered load) in the waiting-time tail; the probability of waiting still rises, so the service level falls whatever the pool size. Pooling Architecture in Service Workforces discusses the same scale economies, which also produce the dedication penalty.

The combined pattern (migration valley)

Taken together, these mechanisms produce a characteristic pattern, called in this article the migration valley (a descriptive label rather than an established term). Service holds through most of the consultation period because attrition takes time to accumulate. It falls around the announcement and the first transfer waves, when elevated attrition, rising absence and pre-cutover training coincide with work that has not yet left. It then eases at the source as work departs and recovers at the destination as the receiving team ramps. The depth and timing of the valley depend on the length of the consultation period, the attrition response, the training plan and how much contingency capacity is available, all of which are uncertain in advance.

Depth depends on the service model

The same synthetic migration (shared pool, 2,000 simulated futures) read two ways; blue is the source site, orange the destination. Solid lines and bands: service level under Erlang A (median and 10th–90th percentiles), in which waiting callers abandon after an average patience of two minutes; dotted: the median share who abandon. Thin dashed: the median service level under Erlang C, in which nobody abandons. Red lines mark a 0.70 service floor and a 10% abandonment cap; the grey band is the median consultation freeze. The shortage is identical; Erlang C reports it as queues that never clear, Erlang A as service held up by callers giving up.

How deep the valley looks depends on the queueing model as much as on the staffing. Under Erlang C nobody abandons, so a pool that falls below its offered load shows service falling towards zero. Under Erlang A, callers who wait longer than their patience hang up, which relieves the queue: service measured as calls answered within the threshold divided by all calls offered stays above zero, and much of the shortage appears as abandonment instead. In the synthetic example behind the figure, with a two-minute average patience, an eight-minute handle time and a 60-second answer threshold, the shared pool's worst week has a median across simulated futures of about 0.85 service level with about 10% of callers abandoning under Erlang A, against about 0.27 under Erlang C (lower than the lowest weekly median plotted, because each future's worst week falls in a different week); the share of simulated futures breaching a 0.70 floor falls from about 88% to about 4%. Neither reading is the more optimistic one in substance, since both describe the same shortfall of capacity; reporting the abandonment rate alongside the trough makes the two readings comparable.

Caller patience therefore becomes one of the most influential inputs. It can be estimated from routine queue data: under the exponential-patience assumption of Erlang A, the mean patience is approximately the total time all callers spent waiting, answered and abandoned alike, divided by the number who abandoned. Measured patience distributions are often not exponential, so the estimate is a first approximation. Customers who abandon may also try again, so the extra abandonment during a transition partly returns as demand in later weeks.

Modelling the transition

Because the inputs are uncertain and the response is non-linear, a single deterministic projection understates the risk; see The Flaw of Averages. A transition model is more informative when it is stochastic and weekly:

• Demand is built per client. Each client is given a probability of transferring, leaving or re-platforming, informed by relationship health and contract terms, and a departure week derived from its contract type, expiry date or notice period, or its transfer wave.
• Supply is built per pool. Attrition runs at a base rate multiplied during the freeze and again after the announcement; staff are split into transferring and releasing groups at the announcement; transferees move with their work; releases follow the work out and are subject to a statutory notice period; training hours reduce productive time before each wave.
• Service is computed each week from workload and productive hours: an Erlang-type service level for interactive work, and a backlog measure for deferrable work.
• Both sides are simulated together within each draw, so that a long consultation period is compared with the depleted supply that the same long period produces. Repeating the draw several thousand times gives a distribution of service level for every week, from which the trough, the probability of breaching a service floor and the duration of any breach can be read.

The same model reconciles to the program's end-state waterfall: its mean headcount flows can be compared line by line with the plan's figures. Wiki:Packs/Migration Service-Level Simulation implements this design with an intake layer for client and staffing workbooks. The Migration Scenario Modeler is a browser-based version of the weekly model for a single blended team, with a choice of channel-balancing policy and of Erlang A or Erlang C for delivered service. It accepts a book of business described by contract type, relationship health and the likelihood of each outcome, and an optional split of the team into transferring and released groups; its uncertainty bands narrow as assumptions are marked as confirmed. The Migration Health Pack describes a weekly governance surface through which the output of such a model can be reported.

Levers

The levers available to a program differ in when they must be arranged and in which part of the valley they affect.

Lever	Mechanism	When it must be arranged
Contingency capacity	Adds productive hours from another site or partner, usually at a handling-time penalty for unfamiliar work	Before the announcement; it concerns capacity outside the affected site and so can be arranged earlier than levers that act on affected staff, subject to any information obligations that apply
Training timing	Moves off-floor hours away from the weeks before cutover, or spreads them	Before the first wave
Retention payments	Reduce attrition among staff who will be released or who transfer late	At the announcement
Wave sequencing	Changes when each block of work, and its staff, leaves the source	After the freeze ends
Release timing	Holds released staff for a period after their work has left, as a buffer against slipped waves	After the freeze ends

In a synthetic illustration built with the pack, the timing of pre-cutover training and the availability of contingency capacity moved the depth of the valley more than retention payments or the length of the consultation period did: the shared pool's median trough rose from 0.28 in the base case to 0.69 with most training moved after cutover and to 0.87 with ten contingency FTE, against 0.32 with retention payments and 0.36 with a short consultation period. That result depends on the illustrative parameters and should be re-established with an organisation's own inputs.

Limitations

Erlang C assumes that callers never abandon. When offered load reaches or exceeds staffed capacity Erlang C has no steady state, and implementations typically return a service level at or near zero, whereas in practice callers abandon and the queue is partly relieved. The Erlang A model incorporates caller patience and gives more realistic estimates under overload. Under Erlang A an overloaded queue reaches a steady state in which part of the offered demand abandons; service measured as calls answered within the threshold divided by all calls offered stays above zero, and the shortfall appears as abandonment instead. A transition model based on Erlang C should report utilisation (offered work divided by available capacity) alongside service level, so that a near-zero trough is read as a statement about overload rather than a literal forecast. See Erlang-A and the comparison above.

Several effects are commonly left out of transition models and bias them in known directions. Client departures that respond to poor service during the transition are usually omitted, which makes the valley look shallower than it may be. Cross-pool overflow is usually omitted, which makes it look deeper. Fungible headcount within a pool overstates the resilience of teams dedicated to single clients.

Maturity Model Position

At Level 1–2, migrations are planned as an end-state headcount reconciliation, and service during the transition is managed reactively as it deteriorates. At Level 3, the transition is modelled week by week with deterministic assumptions for attrition and transfer dates, and service is projected against a floor. At Level 4, the transition is simulated with explicit uncertainty in the consultation period, the attrition response and client fates. Levers are compared as scenarios, and contingency capacity is contracted ahead of the announcement on the basis of the simulated breach probability. At Level 5, transition risk is an input to the decision about whether and when to migrate, and blended human and automated capacity at the destination is modelled as part of the receiving pool.

See Also

• Migration Archetypes
• The Migration Health Pack
• Migrating a Book of Business
• Wiki:Packs/Migration Service-Level Simulation
• Erlang Sensitivity and the Staffing Cliff
• Pooling Architecture in Service Workforces
• Erlang-A
• Employee Attrition and Turnover
• Scenario Planning and Contingency Staffing
• Offshoring and Nearshoring

Interactive tools

• Migration Scenario Modeler — migration.wfmlabs.com. Week-by-week voice, chat and email service through a freeze, runoff, attrition, an absence surge, transfer waves and borrowed capacity, under Erlang A or Erlang C, with a scenario export for further analysis.
• Erlang Suite — erlangcalculator.wfmlabs.com. Single-interval Erlang C and Erlang A staffing.