The Pilot That Did Not Scale
After a six-month pilot in a single distribution center, a voice-directed picking and real-time exception system reached 91% adoption and improved outbound accuracy by three points. Nine months into the multi-site rollout, adoption sits at 54%, 38% of pickers keep a paper pick list as backup, overtime is up 11%, and two supervisors have publicly called the system slower. The pilot benefited from a dedicated trainer, a hand-picked team and spare capacity; the rollout landed in peak season with no backfill for training hours. The workforce has 22% annual turnover and speaks nine languages. The union has requested formal consultation on task changes, and two sites are threatening to revert.
1. Leadership has labelled this a resistance problem and wants a communications campaign. How would you diagnose it instead, and what evidence would settle the question?
Read the worked answer
The resistance label is a hypothesis, and usually the wrong one. Test competing explanations with evidence before spending on communication.
Three candidate causes, each with a test:
- Capability or design failure: if trained users still reach for paper because the headset prompts are wrong for their pick patterns, or if error rates rise in the moments paper is used, the system is not fit for the task at that site. Test: compare task time and error rate with and without paper for the same pickers, and review exception logs by site.
- Capacity failure: no backfill during peak means learning is happening under time pressure, which reliably produces abandonment. Test: correlate adoption by shift with overtime and staffing ratio; check whether dropouts cluster in the weeks with highest volume.
- Legitimate change fatigue and industrial-relations risk: two prior initiatives in 18 months plus an unconsulted task change. Test: consultation status against the site agreement.
The decisive distinction is adoption versus proficiency. 54% usage can coexist with poor outcomes if users are unskilled, and 38% paper backup in trained users points more to proficiency and capacity than to attitude. Do not launch a campaign until this is resolved, because a communication push against a real design flaw destroys credibility permanently.
2. Assuming the diagnosis confirms a mix of capacity and proficiency problems with a genuine consultation obligation, what should the next 90 days contain and in what order?
Read the worked answer
Order matters more than content here; sequencing change interventions wrongly is the most common failure.
- Stabilize capacity first. Backfill training hours, schedule training off-peak, and stop opening new sites. Adding sites now multiplies the failure and consumes the credibility needed later.
- Rebuild supervisor capability before worker capability. The two publicly skeptical supervisors are the constraint. Give them task-time data, a role in tuning the system, and an explicit choice to champion or step aside. Peer supervisors convince floors faster than any central message.
- Fix the design defects surfaced in diagnosis, then retrain on the fixed version. Retraining on an unfixed system teaches users that training does not change anything.
- Engage the union early and substantively. Provide task-change data, revised role descriptions and the retraining guarantee. Consultation is a schedule risk, not a communications task, and it should be sequenced before the next site goes live.
- Localize delivery for nine languages and 22% turnover: short role-based modules, floor coaches per shift, and a standing onboarding path so new hires do not arrive into paper workarounds.
- Only then communicate, and communicate results and fixes, not enthusiasm.
3. Which measures should the programme report to the steering committee monthly to track recovery, and what should trigger a pause?
Read the worked answer
Report leading indicators of proficiency, not just adoption. Adoption can be met while the operation is worse off.
Core measures:
- Paper-backup rate by site and shift (target: below 5%). This is the honest adoption figure; 38% today.
- Task time for trained users versus the pre-system baseline, measured per process step, not as an average.
- Exception rate and rework, with root cause categorized as system, data or user.
- Training completion and proficiency certification, split by new hire versus tenured, because turnover means the population renews.
- Overtime and schedule adherence, to expose whether users are absorbing learning cost in unpaid time.
- Site-level trend lines, since aggregate figures hide two sites reverting while others improve.
Pause triggers, agreed in advance: paper-backup rate failing to fall for two consecutive months at a site; task time persistently worse than baseline for trained users; consultation not closed before a scheduled go-live; or overtime above a set threshold.
A pause with a stated condition for restart preserves credibility. Continuing while the numbers deteriorate teaches the organization that the programme's schedule outranks its evidence.