How We Measure Outcomes
Three rules govern every metric we publish:
- Every metric is tied to a named customer. No anonymous "a leading bank" stats; either the customer goes on record, or we don't publish.
- Every metric is methodology-cited. What was measured, how, against what baseline, over what window.
- Conservative bounds preferred. When we have multiple instrumentation sources, we report the lower bound.
Translation Agents – EDI
EDI's analyst team was spending six figures annually on outside-counsel translation with weeks of cycle time. Translation Agents cut that cost by 85% and compressed per-document cycle time from 12 hours to 2 minutes - with the full audit chain preserved.
- 85% cost reduction – annual translation spend cut by year one
- 360× throughput – 140-page documents from 12 hours to 2 minutes
- 14+ languages live – expanded from 3, no additional vendor cost
Voice Agents – Financial Services
A North American financial-services CX team replaced after-hours voicemail with always-on voice agents. The agent handles inbound calls 24/7 in three languages, with sub-second turn-taking. CSAT went up, not down, after the deployment.
- +18 CSAT points – after-hours quality matched in-hours by month 2
- 62% ops cost reduction – annual call-center spend cut without reducing service hours
- 1.2M calls/month – peak day 50K concurrent, no human queue overflow
QMS Compliance – MedDevice
A German medical-device manufacturer was spending weeks of compliance-team time before every audit cycle. QMS Compliance Agents compressed prep to hours and tripled the client volume the same team could handle - with audit findings dropping 87% from consistent document retrieval.
- Audit prep: weeks → hours – same scope, fraction of the cycle
- 3× client volume – capacity tripled without additional headcount
- ↓ 87% audit findings – consistent retrieval drops auditor findings dramatically
Common Patterns Across Deployments
Looking across these three customers - and the dozens of smaller engagements that haven't graduated to full case studies yet - a few patterns recur:
- The first agent ships in 2-4 weeks. Pilot in days, production in weeks. The bottleneck is usually security review, not LuMay engineering.
- Payback is visible by week 12. Most engagements show ROI-positive metrics in their first quarter; only one outlier took longer (an EU on-prem deployment that required additional compliance work).
- Cost-per-agent drops with the second deployment. Shared platform fabric and institutional knowledge mean the eleventh agent costs less than the second.
- Adoption is the harder problem, not capability.Where we've paired engineering with Academy enablement, adoption sticks. Where we haven't, we've had to come back and add it.