Value First, Then Route
Value first, then route is the operating sequence of the Value-Based Planning Model. Every interaction type is valued before anyone decides who handles it; routing is then designed on value and capability together; each destination is staffed with the model that fits its work. The Value-Based Planning Model page states the model itself, and states the Erlang inversion as a change of planning input; this page names the stages the presentation puts in the old chain's place. This page is the mechanic: the stage-by-stage chain that replaces the deflection chain, the three-stage routing rule that runs it, and the reason one staffing formula cannot serve the three destinations. It borrows the pools from the Three-Pool Architecture and the scoring from the Value Routing Model without re-deriving either.
Two chains
The planning chain most contact centers run has four steps, and the third has been in continuous use for a century.[1] The chain that value-based planning substitutes has five, three of them new.[2]
| Old chain | New chain | |
|---|---|---|
| 1 | Inbound volume — all inquiries, by channel | Total workload — all inquiry types, all channels |
| 2 | Deduct deflection — IVR, bots, self-service | Value classification (new) — high, medium, low |
| 3 | Plan with Erlang — offered volume to agents; a single formula | Routing decision (new) — human, collaborative, AI |
| 4 | Headcount | Staff by channel (new) — where channel means handler class: human, collaborative, AI; not chat, email, phone |
| 5 | Value output — spend against value derived per dollar |
Four differences carry the argument. Deflection is removed as a step. Automation is no longer a subtraction taken before planning begins; it is a destination work is routed to, with its own cost, failure modes and place in the plan. Total workload is never reduced up front. The first stage counts everything, including the work that will end up with systems. The unit of staffing changes. The old chain staffs by media channel, because that is how queues were built; the new chain staffs by handler class, because that is what determines the staffing model. The output changes. The chain no longer ends in a headcount. It ends in a ratio of spend to value derived, which treats productivity the way the marketing-science literature says it should be treated: as a decision variable set for profit, not a quantity to be maximized.[3]
Why the old chain fails under agentic automation is the subject of Planning for Agentic AI Is Not a Carve-Out; this page takes that as given and describes what runs in its place.
Stage one: classify by value
Every interaction type receives a value score on a 1-to-10 scale. The presentation illustrates the stage with ten types scored from 10 down to 2, spanning judgment-heavy expert work at the top and simple status transactions at the bottom.[2] Two features of the stage matter.
The unit is the interaction type, not the individual contact: classification is a planning act done in advance on a taxonomy the routing platform can distinguish, not a judgment made on each arrival. The score is composite. Its four sub-dimensions, each scored for human and for AI handling, are the Value Routing Model's subject; the score here is that model's output. The band structure the scores fall into, and why a formula that ignores it under-resources the work that matters, is developed in Tier 1 Is Not One Thing.
Stage two: route by design
Routing runs on two dimensions, not one. The first is the value score from stage one. The second is AI capability for the type: a planning-time estimate of how completely a system can handle that type, distinct from the containment rate measured after deployment. The presentation groups illustrative types into three bundles, each carrying a routing rule:[2]
| Bundle (illustrative scores) | Rule | Destination |
|---|---|---|
| 10, 4, 2 | High value, or low AI capability | Specialist |
| 8, 6, 6 | Medium complexity; AI-assistable | Collab |
| 6, 3, 2 | Low value and high AI capability | AA |
The bundles are illustrative, and not every placement in them satisfies the published thresholds; a score of 6 in the AA bundle, for instance, would resolve to Collab under the AA threshold the Three-Pool Architecture states. What the bundles show is that value alone does not determine the route. A type scored 2 sits in the Specialist bundle because no system can yet complete it, and the same score sits in the AA bundle because for that type one can. Nor does capability alone determine it: a type scored 10 goes to people whatever a system could do with it, because the value at stake justifies the human hour. The presentation states the Specialist admission rule explicitly as AI capability under 30 percent, or value score 8 or above; the full heuristic, including the threshold for the AA pool and the sensitivity of the thresholds, is carried by the Three-Pool Architecture.[2]
"By design" distinguishes this stage from what a routing platform does in real time. Skill-Based Routing executes an allocation contact by contact, on competence and availability. Stage two decides the allocation at planning time, on value and capability, and hands the platform a design to execute. The design is where the human-machine boundary is set, and the classical warning about automation is that the boundary is usually set by default, with people receiving whatever the designer could not automate.[4] Routing by design makes the boundary a deliberate decision, re-taken each cycle as capability moves.
Stage three: staff each pool with its own model
Each destination is staffed by a different model. The presentation labels the three as Erlang plus a complexity premium, the N* cognitive portfolio model, and a transaction cost model.[2] Staffing by handler class at stage four follows from this: a pool is the set of work that shares a staffing model.
- Specialist — queue-based, with a complexity premium. Its work is heterogeneous enough that the Three-Pool Architecture staffs it by simulation rather than closed-form Erlang; Erlang C with a complexity-premium multiplier is the approximation it offers to operations without simulation capability. Either way the input changes: handle time in this pool is higher and more variable than the pre-automation average, because the low band has left. The Service Demand Rebound Model quantifies that shift as the Complexity Premium, and The Hardening Residual carries what it does to every other sized quantity.
- Collab — the N* cognitive portfolio model. The pool is not queue-based. People oversee several concurrent system sessions and intervene on judgment calls, so the binding constraint is cognitive load rather than time on task, and the model solves for the number of sessions one person can carry. The derivation is on Cognitive Portfolio Model (N*).
- AA — a transaction cost model. The pool has no occupancy constraint. Its cost includes a rate per interaction, the expected cost of the interactions it fails to complete, and the demand its capacity induces. The last two are The Escalation Tax and the Service Demand Rebound Model; the full five-layer cost model is on the Three-Pool Architecture page.
The presentation's rule is that three pools need three staffing models, and one formula cannot serve all three.[2] The reason is structural. Erlang-family models assume that a server handles one customer at a time, that arrivals and service times come from homogeneous populations, and that every customer admitted to service completes and departs.[1] They are silent on work that fails after service. The Collab pool breaks the first assumption, because one person carries several sessions. The Specialist pool strains the second, because its residual work is too heterogeneous for a single service distribution, which is why the sibling pages staff it by simulation. The AA pool falls outside the third, because its failures leave the pool and land in a different one, so the cost that matters most is the one the queue model does not see. The operations literature already surveys a body of staffing models developed for structurally different call-center settings rather than a single formula.[5] Three pools make the choice of model explicit.
Architecture versus deflection planning
The presentation closes the mechanic by contrasting deflection planning with workforce architecture.[2] Deflection planning answers one question: how much volume leaves the human queue. Architecture answers four: which handler class carries which work, at what cost, where its failures go, and how the answer changes as capability moves. Three properties follow. The boundary is bidirectional: work moves to systems as capability rises and back to people as it fails, and both directions are routing outcomes. Failure paths are explicit: an interaction the AA pool cannot complete has a designed destination and cost, which the cascade formula on The Escalation Tax prices. And the design is re-run each cycle, because stage two depends on capability and capability moves; the routing decision is a planning deliverable with a cadence, not a configuration set once.
Maturity Model Position
Under the WFM Labs Maturity Model™, the old chain is the appropriate mechanic through Level 3, where automation is scripted and containment stable enough to treat as a deduction. The five-stage chain is the operating mechanic of Level 4. Its prerequisites are the model's: an interaction taxonomy the platform can route on, a value score per type, and a capability measure per type that is refreshed each cycle. At Level 5 stage two runs continuously, and the three pools are re-balanced as capability signals arrive rather than at planning intervals.
See Also
- Value-Based Planning Model — the model this mechanic serves; owns the Erlang inversion and the four building blocks
- Planning for Agentic AI Is Not a Carve-Out — why the old chain fails under agentic automation
- The Value of an Hour — the time taxonomy that gives stage one its meaning
- Tier 1 Is Not One Thing — the three value bands the scores fall into
- Value Routing Model — the four sub-dimensions behind the value score
- Three-Pool Architecture — the three destinations, the full routing heuristic and its thresholds
- Cognitive Portfolio Model (N*) — the Collab pool's staffing model
- The Escalation Tax · Service Demand Rebound Model — the two cost terms in the AA pool's model
- The Hardening Residual — the sizing consequences of the Specialist pool's harder residual
- AI Containment Rate and Its Workforce Implications — the measured outcome that AI capability is not
- Skill-Based Routing — the real-time execution of a routing design
- Multi-Objective Optimization in Contact Center — the surface the routing thresholds are tuned on
References
- ↑ 1.0 1.1 Gans, N., Koole, G., & Mandelbaum, A. (2003). "Telephone Call Centers: Tutorial, Review, and Research Prospects". Manufacturing & Service Operations Management 5 (2), 79–141. doi:10.1287/msom.5.2.79.16071.
- ↑ 2.0 2.1 2.2 2.3 2.4 2.5 2.6 Lango, T. (2026). Adaptive: Building Workforce Systems for an (Unpredictable) Future. Presentation, SWPP Annual Conference, slides 39–40. The two chains, the three-stage routing mechanic and the illustrative value scores are the presentation's framing.
- ↑ Rust, R. T., & Huang, M.-H. (2012). "Optimizing Service Productivity". Journal of Marketing 76 (2), 47–66. doi:10.1509/jm.10.0441.
- ↑ Bainbridge, L. (1983). "Ironies of Automation". Automatica 19 (6), 775–779. doi:10.1016/0005-1098(83)90046-8.
- ↑ Akşin, Z., Armony, M., & Mehrotra, V. (2007). "The Modern Call Center: A Multi-Disciplinary Perspective on Operations Management Research". Production and Operations Management 16 (6), 665–688. doi:10.1111/j.1937-5956.2007.tb00288.x.
