Matching and routing

When partners' traits are complements (each makes the other more productive), total output is highest with positive assortative matching: best with best. Shimer and Smith (2000, Econometrica 68(2): 343–370) show that search frictions can weaken this.

If a great rep is worth more on a great lead than on a poor one, pair them. Close rate × lead value is multiplicative, which is the definition of complements.

The "deferred-acceptance procedure … always leads to a stable outcome." The side that makes the offers gets its best stable matching. Roth and Shapley won the 2012 Nobel "for the theory of stable allocations and the practice of market design."

Matching is a solved, designable problem, whether for territories, account assignment or routing, and design choices decide who comes out ahead.

A review of call centers as systems where "the behavior of customers and employees is closely intertwined with physical performance measures," and where traditional queueing models are "fundamentally limited."

The base reference before challenging a speed-to-answer target.

In large, busy systems where agents differ in speed, routing each caller to the fastest available agent first minimizes waiting.

Routing that's optimal for speed loads up the fastest agents. It optimizes waiting, not value.

"There is a natural trade-off between minimizing customer wait time and fairly dividing the workload among agents of different skill levels." Their policy "outperforms a common routing policy used in call centers (that routes to the agent that has been idle the longest)."

"longest idle" routing, the fairness default, has a measurable cost.

Agents "exhibit very different performance for the same call type." Simulations on a financial firm's data trace a trade-off curve: the best-resolution rule reached about 90.6% resolution with waits over 7 minutes, while a hybrid rule reached about 88.6% with about 18 seconds' wait.

The closest published evidence on routing by ability: it costs a little speed and buys better outcomes, and smart hybrid rules get most of both.