Difference-in-Differences in WFM

Difference-in-differences: the treatment effect is the gap between the treatment group's actual post-intervention outcome and the parallel-trend counterfactual extrapolated from the control group.

Difference-in-Differences in WFM (DiD) is a quasi-experimental method that estimates the effect of an intervention by comparing the change in a treated group to the change in an untreated comparison group over the same period. By differencing twice — before-versus-after and treated-versus-control — it removes both fixed differences between the groups and any trend common to both, isolating the effect of the intervention. It is among the most practical causal tools for workforce management, because interventions are routinely rolled out to some teams, sites, or periods but not others.

How it works

The estimate is the difference of two differences. Take the treated group's outcome after minus before, and subtract the control group's outcome after minus before:

DiD = (Treated_after − Treated_before) − (Control_after − Control_before)

The control group's change estimates what would have happened to the treated group absent the intervention — the counterfactual trend. The treatment effect is how far the treated group departed from that trend, the gap shown in the diagram. The approach was made famous by Card and Krueger's study of a minimum-wage change, which compared fast-food employment in adjacent states.

The parallel-trends assumption

DiD's validity rests on one central assumption: parallel trends — that, absent the intervention, the treated and control groups would have moved in parallel. The pre-intervention periods are used to make this credible: if the two groups tracked each other before the change, parallel trends is more believable afterward. The assumption fails when something other than the intervention differentially affects one group at the same time, or when groups were already diverging. Recent econometric work has shown that staggered rollouts (different units treated at different times) can bias naive DiD estimates and require careful estimators.

WFM examples

• Tool or policy pilots. A new agent-assist tool, schedule policy, or coaching program launched at some sites but not others: compare the change in AHT, service level, or attrition at pilot versus non-pilot sites.
• Staggered rollouts. A change deployed team-by-team over months supplies a natural treated/not-yet-treated comparison at each step (with the staggered-design cautions above).
• Policy boundaries. A regulatory or compensation change affecting one region but not an otherwise-similar region.

Cautions

• Check the pre-trends. Always verify the groups moved in parallel before the intervention; diverging pre-trends invalidate the design.
• Watch for concurrent shocks. Anything that hits one group and not the other at the same time (a reorg, a separate tool change) confounds the estimate.
• Composition changes. If who is in each group changes across the period (attrition, transfers), the comparison is contaminated.
• Staggered timing. With many adoption dates, use modern DiD estimators rather than a naive two-way fixed-effects regression.

Maturity Model Position

In the WFM Labs Maturity Model™, DiD is often the first rigorous causal method an operation adopts because pilots and rollouts are everywhere.

• Level 1–2 (Emerging / Foundational) — pilots are judged on the treated group's before-and-after alone, with no control and no counterfactual.
• Level 3 (Progressive) — evaluations add a comparison group, check parallel pre-trends, and report the difference-in-differences rather than a raw change.
• Level 4–5 (Advanced / Pioneering) — DiD is standard for rollout evaluation, with pre-trend testing and appropriate estimators for staggered designs, feeding go/no-go scaling decisions.

See also

• Causal Inference in Workforce Management
• Potential Outcomes Framework
• Instrumental Variables in WFM
• Regression Discontinuity in WFM
• A/B Testing for WFM Experiments
• Regression to the Mean in WFM