Supply-Side Routing
Supply-side routing is the decision of which source of capacity absorbs a given gap between the hours an operation needs and the hours it has: which lever, which pool, which partner, which shift or which worker, and at which horizon. Contact routing decides where a contact goes. Supply-side routing decides where the capacity comes from. The two problems share a capability record and a pool structure, and a change to one changes the other, but they are answered by different components, on different clocks, and with different consequences when wrong. The page treats supply-side routing as a bounded choice over an enumerated option set, which makes it the first application of Workforce Decisioning, and sets out the option set, the eligibility step that precedes the choice, the structure of the decision, the rules that stay deterministic, and the measures that show whether the routing is good.
The option set comes from the Supply Elasticity Lever Inventory. The time axis comes from Variance Routing by Horizon. The eligibility step comes from the capability record and the blocker logic of Placement Engine Architecture. What the page adds is the decision itself.
Why it is a routing problem
A gap of six percent in a pool's afternoon hours can be closed by moving cross-trained agents from a neighbouring pool, by offering overtime, by deferring back-office work, by calling bench hours from a partner, by accepting a lower service level for the afternoon, or by doing nothing and letting the interval loop absorb it through occupancy. Each of these is a route the gap can take. Each has a cost, a lead time, a reversibility and an eligibility condition, and each has consequences for other pools. The real-time analyst who picks one is routing supply, and in most operations does so by habit, by the order the levers appear on a checklist, or by whichever is quickest to execute, which is rarely the cheapest or the most reversible.
Three facts make the problem tractable as a bounded choice, with optimisation applied in code at the checking step.
- The option set is finite and known in advance. The lever inventory enumerates twenty-six levers in five families; for any one gap, at any one horizon, the eligible subset is usually under ten.[1]
- The horizon is determined before the choice. Persistence routing assigns a variance component to a loop by how long it has held its sign; the loop fixes the lead time the lever must meet.[2]
- Eligibility is computable. Whether a worker, a pool or a partner may absorb a gap is a matter of capability records, contract terms and working-time rules, all computed from recorded state; whatever judgement was involved went into the record, not into the day's decision.
What remains after the option set, the horizon and the eligibility are fixed is a choice among a few eligible routes, informed by state a rule cannot fully enumerate: the shape of the gap, what was tried last time, the partner's note about their own staffing, the quality risk of the pool that would lend. That residual is the judgement the decision tier is for.
The option set by clock speed
| Clock | Levers (inventory codes) | Lead time | Reversibility | What it costs |
|---|---|---|---|---|
| Minutes | B5 intraday reallocation; break and offline moves; routing threshold changes | Immediate | Free | Occupancy and service in the lending pool; interruption of offline work |
| Hours | A4 overtime and A3 voluntary time off offered on the day (the inventory lists both at days' lead time for planned use; same-day offers are their fast form); D1 opt-in surge pay; B2 back office as the shock absorber, deferring the day's deferrable work | Same day | Low; an offer can be withdrawn until accepted | Premium hours; deferred work that must be caught up |
| Days | C3 bench and burst hours; A7 shift trades in a marketplace; C6 follow-the-sun lending; E1 shifting simple contacts to automation; E2 the deferrable-work calendar, moving planned work into the trough | One to three days, by notice terms | Moderate; partner hours once called are usually billed | Bench rate; cross-region handling cost |
| Weeks | C1 and C2 contract bands; B1 cross-trained group resizing; A5 variable-hour contracts; A6 contingent and returner pools; class starts | Weeks to a pipeline | Low once committed | Contract notice; training hours; hiring cost |
The clock is the first filter. A gap routed to the interval loop can only take minute- and hour-clock levers; a gap routed to the capacity loop may use them to cover the weeks until the pipeline delivers but should not rely on them beyond that lead time, because carrying a persistent gap on overtime for a quarter is, in the phrase of Variance Routing by Horizon, a capacity decision that has not been taken.[2]
Eligibility before choice
The eligible set is computed in code from three sources, and the decision tier chooses only among eligible options.
May versus should
A capability record states what a worker may be routed to do: product, channel, language, platform, client knowledge, vetting, each with an evidence grade and a published level or an explicit "insufficient data". The worker's standing, their primary assignment, states what they should receive. Supply-side routing moves a worker off their standing assignment onto work they are eligible for: a cross-trained agent whose record shows a published proficiency on the short pool is eligible to lend; one whose record is stale or insufficient is not, however the roster is tagged.[3] Lending a worker whose record is stale produces the staffed-but-short failure: rostered hours meet the requirement, capable hours do not.
Blockers
Between any two pools that might share supply there may be blockers, and they come in three kinds.
| Blocker | Example | Effect on eligibility |
|---|---|---|
| Hard | Language; a contractual dedication; a location or onshore requirement | Not eligible; the route does not exist |
| Soft | A refresher or certification that can be trained | Eligible at a price in training hours and a lead time; not eligible today |
| Priced | Vetting; a contract clause that can be relaxed at cost and notice | Eligible at a stated price and lead time, usually on the days or weeks clock |
The blocker table is what turns "move four people from pool R" into an eligible or ineligible route with a price attached. On the day it is a lookup; the judgement about what counts as a hard blocker or what a relaxation costs was made when the table was written.
Rules
Working-time limits, rest periods, the frozen zone of the published week, contractual notice and the legal ceiling on overtime all remove options from the set in code. A decision tier that is ever asked to choose an option a rule forbids has been given the wrong option list.
The structure of the decision
With the horizon fixed and the eligible set computed, the decision is a fan-out of three primitives on one state version, followed by arithmetic and a gate.
| Step | Component | Content |
|---|---|---|
| 1. State | Signal tier | Gap by cause and persistence; pool, interval and horizon; the eligible options with cost, lead time and reversibility; blockers; the last three dispositions for this pool; any narrative attached to the gap |
| 2. Will the gap persist past the current window? | Proposition | Determines whether a minute-clock route is sufficient or an hour-clock route should be prepared in parallel |
| 3. Which eligible route absorbs the gap? | Choice over the eligible set, any composite routes the gate design allows, plus "none" | The judgement; answered by a typed model or a classifier, with the full distribution kept |
| 4. Quality risk of the lending pool | Score, 0–3 | Whether lending will push the lender's own service or handle time out of band |
| 5. Expected cost and effect | Code | Hours moved × rate; the lender's occupancy and service after the move, by Erlang; the borrower's after |
| 6. Gate | Gate table | Class of the chosen route; released or proposed; confidence threshold for the class |
Step 5 is where the choice is checked rather than trusted. Pooling arithmetic is not monotone: lending agents out of a pool near its limit can cost the lender more service than the borrower gains, and pooling queues whose handle-time distributions differ markedly can increase waiting rather than reduce it.[4][5] The decision tier proposes a route; code computes what the route does to both pools before the proposal reaches the gate.
What stays deterministic
The decision tier does not choose everything, and the boundary is a design decision that should be written down.
- Horizon assignment is persistence routing, a rule.
- Cost ordering within a clock is arithmetic: eligible routes are ordered by cost per gap-hour closed, and the tier's choice is checked against that ordering. Where the tier prefers a dearer route, the proposal row says so and the person at the gate sees both.
- Eligibility is capability records, blockers and rules.
- The frozen zone of the published week and the change-cost schedule of Planning Nervousness and Frozen Zones are rules.
- Service-level differentiation between pools, where some pools are protected and others carry the variance, is a stated policy and can be implemented by queue-ratio rules in the routing engine rather than by the decision tier.[6]
The judgement the tier adds is confined to the residual: which of two or three near-equal eligible routes fits this gap, given its shape, the dispositions of similar gaps and any narrative state. Published intraday-updating models optimise the choice among adjustments given costs that are fully specified;[7] the residual is what those costs cannot express: the shape of the gap, a partner's narrative, the lending pool's quality risk.
A worked example
Tuesday, 11:00 (all figures illustrative). Pool P's afternoon requires about two hundred hours and shows a gap of six percent, twelve hours, from level (contacts above forecast), on the first day of a level excursion this week. Persistence routing, which counts in weekly cycles, leaves it with the interval and short-term loops, so the clock is minutes and hours.
Eligibility, computed:
| Route | Eligible? | Price and lead time |
|---|---|---|
| B5: reallocate four agents from pool R | Yes for two; soft blocker for two (refresher outstanding) | Free for the two; the other two priced at four training hours each, not today |
| A4: twelve overtime hours | Yes, within the weekly ceiling | Premium rate, same day |
| B2: defer the afternoon's back-office batch and lend its three agents certified on P | Yes for the three certified agents | Catch-up tomorrow; no cash cost |
| C3: eight partner bench hours | No for today; forty-eight hours' notice | — |
| None: absorb through occupancy | Yes | Pool P occupancy to ninety-three percent for two hours; service at risk in two intervals |
The fan-out on this state returns: persistence past the window 0.41; route choice B5 0.58, B2 0.27, A4 0.12, none 0.03 (uncalibrated until the class has disposition history); lender quality risk for pool R 1.2 on 0–3. Code computes the effect: two agents lent from R for three hours closes six of the twelve gap-hours; R's service stays inside its band; P's two at-risk intervals fall to one. Ordered by cost per gap-hour closed, B5 and B2 are both free and A4 is the dearest, so the ordering does not contradict the tier's choice. The class is intraday reallocation, released at a threshold of 0.75 on the calibrated probability; at 0.58 the row goes to the real-time analyst as a proposal showing B5 and B2 with their probabilities and computed effects. The gate design allows the signer to compose routes, so the analyst signs B5 for the two eligible agents and B2 for the residual, both logged against the proposal, and the two blocked agents' refreshers are added to the training queue as a priced unblock. At the disposition review the gap is found to have closed by 13:30 and the row enters the calibration set.
The example is deliberately undramatic. The value of routing supply as a decision is not that the tier picks something an analyst would not; it is that every such afternoon leaves a row, the rows accumulate into the shares that Variance Routing by Horizon reports, and the shares reveal the persistent gap that has been carried on overtime for a quarter.
Measurement
- Share of gap-hours by route, by clock, weekly: the diagnostic the control-system series specifies. A rising share on hour-clock routes for a gap the persistence test should have promoted is a routing failure, not a capacity one.
- Routing regret: the cost of the route taken against the cheapest eligible route that would have closed the gap, judged at disposition. Reported as a distribution, not a mean.
- Override rate by class: how often the gate signs a route other than the one proposed, and which. A high override rate on one class means either the option list or the state is wrong for that class.
- Staffed-but-short incidents: gaps that a route appeared to close on rostered hours and did not on capable hours. Each is a stale capability record found the expensive way.
- Calibration of the route choice, by class, from the disposition reviews.
Relationship to contact routing
Contact routing and supply-side routing read the same capability record and act on the same pools, which creates two couplings. A merged pool or a lent agent changes the option set the contact router sees, so a supply-side move should be published to the routing engine as a state change rather than discovered by it. And a contact-routing policy that protects one pool's service at the expense of another's is a standing supply-side decision; it should appear in the eligible set as a constraint with an owner rather than as an invisible default. Skill-Based Routing and Pooling Architecture in Service Workforces describe the contact side; the Value Routing Model decides which pool a contact deserves, which is a different question from which pool can be spared.
Limitations
- The option set is only as good as the inventory and the records. A lever that is not enumerated cannot be chosen; a capability record that is not maintained makes eligible workers invisible and stale ones visible.
- Pooling arithmetic must be run, not assumed. The Erlang effect of a move on both pools is computed per proposal; the example's figures are illustrative and the direction of the effect is not always favourable.
- Partner routes carry contract state the tier cannot see. Bench hours, corridor positions and notice periods live in the contract and must be published as state; a route the contract forbids is a hard blocker, not a low-probability option.
- Calibration of the route choice is slow to accumulate. A pool may route a few gaps a week. Per-class reliability curves take months; until then the class stays proposed.
- The literature prices what it can specify. Safety-staffing results size the buffer;[8] intraday-updating models choose among adjustments with known costs; neither prices narrative state, which is why the judgement step exists and why its calibration must be measured rather than assumed.
Maturity Model Position
- Level 3 — Progressive: the option set and the eligibility rules are written down; the analyst chooses and the choice, the alternatives and the disposition are logged; shares by route are reported weekly.
- Level 4 — Advanced (The Ecosystem Emerges): eligibility is computed from capability records and blockers; the route choice is proposed by a decision tier with a confidence; pooling effects are computed before the gate; minute-clock classes are released on evidence.
- Level 5 — Pioneering: the same decision tier routes supply across in-house, partner and virtual pools; priced unblocks feed the training and contract queues automatically; routing regret is a managed metric.
See Also
- Workforce Decisioning
- Decision Primitives for Workforce Operations
- From Signal to Decision: The OVIX Case
- Supply Elasticity Lever Inventory
- Supply Elasticity in Workforce Planning
- Variance Routing by Horizon
- Placement Engine Architecture
- The Placement Gate
- The Workforce Broker
- Multi-Skill Pooling and the Double-Counting Trap
- Pooling Architecture in Service Workforces
- Cross-Training and Skill Mix Strategy
- Skill-Based Routing
- Intraday Management
- Planning Nervousness and Frozen Zones
- Human Gates and Number Grades
References
- ↑ Supply Elasticity Lever Inventory, this wiki: five families of lever (hours per head, who can do what, where the hours come from, pay and incentive design, demand-side shaping), each scored on reach, speed and cost. The reversibility column on this page is added here and is not an inventory attribute.
- ↑ 2.0 2.1 Variance Routing by Horizon, this wiki: the four-step method that routes a week's gap components to the interval, short-term, capacity or structural loop by persistence.
- ↑ Placement Engine Architecture, this wiki: capability records with evidence grades, published levels and the may-versus-should distinction.
- ↑ Tekin, E., Hopp, W. J., & Van Oyen, M. P. (2009). "Pooling strategies for call center agent cross-training". IIE Transactions 41 (6), 546–561. doi:10.1080/07408170802512586.
- ↑ Multi-Skill Pooling and the Double-Counting Trap, this wiki: why summing pool requirements over-counts and why a lent agent is not a free agent.
- ↑ Gurvich, I., & Whitt, W. (2010). "Service-level differentiation in many-server service systems via queue-ratio routing". Operations Research 58 (2), 316–328. doi:10.1287/opre.1090.0736.
- ↑ Mehrotra, V., Ozlük, O., & Saltzman, R. (2010). "Intelligent procedures for intra-day updating of call center agent schedules". Production and Operations Management 19 (3), 353–367. doi:10.1111/j.1937-5956.2009.01101.x.
- ↑ Bassamboo, A., Randhawa, R. S., & Zeevi, A. (2010). "Capacity sizing under parameter uncertainty: safety staffing principles revisited". Management Science 56 (10), 1668–1686. doi:10.1287/mnsc.1100.1203.
