Reports and KPIs
Pilot and rollout, weeks 1 to 12. Targets are set during discovery; every figure here is illustrative.
Illustrative data
Step 22 of 24 · Presenter note
The KPIs from the concept, with Windchill topic reuse and TM segment reuse measured separately. All numbers here are illustrative.
CHF 318KExternal cost avoidedPilot period, against Acolad rates
61%Share of Acolad scope replacedBy spend, Windchill flow
5.8 dAverage turnaroundAcolad baseline 9.5 days
98.6%Publications succeeded first timeFailures retried safely
81%Windchill topic reuseTopics not sent for translation
55%TM exact-match reuseSegment level, all languages
17%AI units corrected by the loopBefore any human review
29%AI units edited by proofreadersHuman correction rate
Turnaround per request, in days
Findings per 1,000 AI units, by category
Terminology99
Accuracy: mistranslation40
Accuracy: omission32
Negation or condition8
Numbers and units6
Tags and placeholders10
Fluency and style94
Locale conventions33
CriticalMajorMinor
TM matching by language
Exact match (reused)Fuzzy matchNew (AI translated)
IT61 / 18 / 21
FR60 / 19 / 21
ES59 / 19 / 22
NL58 / 19 / 23
ZH57 / 19 / 24
DE57 / 20 / 23
SV57 / 19 / 24
CS56 / 20 / 24
PL56 / 20 / 24
PT-BR55 / 20 / 25
DA55 / 20 / 25
TR55 / 20 / 25
FI54 / 20 / 26
PT54 / 20 / 26
HU53 / 20 / 27
NO53 / 20 / 27
FR-CA52 / 21 / 27
RO52 / 21 / 27
SK52 / 21 / 27
AR51 / 21 / 28
HR51 / 21 / 28
BG50 / 21 / 29
ET50 / 21 / 29
SL50 / 21 / 29
ES-LA49 / 22 / 29
CA49 / 22 / 29
LT49 / 22 / 29
ZF48 / 21 / 31
LV48 / 22 / 30
ID47 / 22 / 31
HE47 / 22 / 31
UK47 / 22 / 31
VI46 / 22 / 32
MS45 / 23 / 32
TH44 / 22 / 34
SR44 / 22 / 34
KM36 / 21 / 43
MY34 / 20 / 46
AI units edited by proofreaders, highest 12 languages
MY Burmese49%
KM Khmer47%
SR Serbian35%
TH Thai33%
ZF Chinese (Traditional, Hong Kong)32%
AR Arabic32%
MS Malay31%
VI Vietnamese31%
HE Hebrew31%
CS Czech30%
HR Croatian28%
LT Lithuanian28%
03060%
Khmer and Burmese need the most human correction, as the Phase 0 benchmark predicted.