Root cause analysis
Root cause analysis na di habit to dey ask
why sometin go wrong until di answer no dey shift again. Fix di root and di symptoms wey dey on top go comot with am; fix symptom
and e go come back. Quality management, IT incident response, patient safety and aviation all dey demand one afta
failure. Di methods dey differ mainly for who dey do di reasoning.
Di methods
Five whysAsk “why” about di failure, den about di
answer, five times. One pesin or small team. E fast and e cheap, and e dey find one chain of causes — di one wey di pesin wey dey ask
don already suspect. E good for single incident wey get single mechanism; e weak wia causes dey affect each oda.
Fishbone (Ishikawa) diagramCauses wey dem sort into
categories — pipo, process, equipment, materials, environment — wey dem draw as bones from one spine. Facilitator dey fill am
from di group suggestions. E wide pass as e deep: e dey list candidates but e no dey tok anytin about which causes dey affect which.
Fault tree analysisLogic tree from top go down, from di
failure through AND/OR gates reach basic events. E strict for engineered systems wia dem know di components and how dem dey
fail. Na analyst dey build am; e need data wey di social world no dey usually get.
Pareto analysisCount di incidents by category and
fix di categories wey cause most of dem. E dey show wia di pain dey, no be wetin dey cause am — na triage tool, no be tool to find
cause.
Structured Democratic DialogueDi pipo wey
dey live with di problem dey write di causes wey fit be, explain dem, den judge — one pair at a time, at 75% bar — whether
each cause get big hand for di oda one. Algorithm dey arrange di answers into influence map
(interpretive structural modelling), and di causes wey dey di bottom of di map na di roots. Fifty years of practice,
hundreds of dialogues wey dem document. Na wetin dis site dey run.
Wen group dey beat di analyst
One question dey decide. Anybodi for di
room know di cause? If machine fail, engineer wey get di logs go find why fast pass fifteen pipo wey dey vote. If deliveries
dey late, patients dey wait, school dey lose teachers or town dey lose trust, di causes dey spread among di pipo
wey dey involved — di night shift know one, di dispatcher know anoda, di customer know di third — and no analyst hold all of dem. Den
di method must collect di causes from evribodi and let evribodi judge di links, or e go find only wetin di analyst
don already believe.
Two findings from di research dey explain why plain group
discussion no dey do dis one either. Pipo wey dem ask make dem rank causes by importance bifor dem examine di links between dem
dey rank dem wrong — di erroneous priorities effect — and group wey dem ask for im views dey spread dem so
wide dat notin get majority — spreadthink. Dem build structured dialogue around di two: no ranking until dem don judge di
links, and dem dey judge links one at a time by supermajority.
Wetin good root cause analysis dey produce
- Map, no be list: which causes dey affect which, with di roots for di bottom.
- Each link with di agreement wey dey behind am, so reader fit see wetin settle and wetin no settle.
- Di arguments dey di record, dem no average dem away.
- Actions wey dem judge against di roots, no be against di symptom wey dey shout pass.
Di seven-stage template ·
Example wey we work out with real numbers ·
How dis site dey run am for any number of groups