The right approach to lead-time analysis?
Six Sigma – iSixSigma › Forums › Old Forums › General › The right approach to lead-time analysis?
- This topic has 3 replies, 4 voices, and was last updated 14 years, 2 months ago by
BTDT.
-
AuthorPosts
-
March 16, 2008 at 7:57 am #49608
Hi All,
I am doing a process lead time improvement analysis and would like to know what would be the right approach to analyze and quantify the improvement made.
1. Transform to obtain a normal distribution and perform the analysis with the appropriate tool for normal distribution?
2. Proceed with the underlying distribution, i.e. lognormal distribution?
Any advice would be greatly appreciated. Thanks!
dave
0March 20, 2008 at 2:06 pm #169915
Joseph ProvinoParticipant@Joseph-ProvinoInclude @Joseph-Provino in your post and this person will
be notified via email.hi
A Lead-Time analysis, is quite an extensive job to do, and department often work in silot to reduce there own individual lead-times. Instead of looking at the overall pictures. This surely put some strain on the departments which are at the back of the supply chain, to improve there LT (Lead-Time) whislt the processes at the front are more free and flexible. Also this silot views are really detrimental as they don’t allow to balance priorities within the strategic imperatives of the organisation.Shortening LT at department level could translate in more spending at wider level.
So when looking at lead-time first I would conduct a top level SIPOC of the overall processes of your organisation, then drill down in area of larger returns where there are less conflicts with the other KPI from your organisation..
hope this helps.. Joe0March 20, 2008 at 2:42 pm #169917Dave, ANOVA is powerful enough to analyze your pre-post Lead Times and detect a difference. Or you can go with mood’s median test. Either way, you don’t need to transform your data to perform this analysis.Good luck.
0March 20, 2008 at 3:17 pm #169918Dave:We are in the final Control of a supply chain project. The
demand for the items varies from hundreds of times per week to once every few
years at local and remote locations. We have done all the process mapping,
changed some delivery scheduling and introduced some new items into stock. The
time to receive an order is one of the primary CTQs. The time is a combination
of a log-normal distribution for sourcing and a uniform distribution for
shipping and receiving. The demand curve for order time-of-day is non-normal
and depends on whether the remote locations run double shifts or not.We did not use inferential statistical analysis to determine
the Vital Xs on the project; process mapping, brainstorming were enough to show
what we should be working on. The biggest issue was to show the team and the
stakeholders what the expected improvement would be BEFORE they were going to
commit to changing the process.We managed to pull much of the required data from the work
management system for the last 34 months of operations. This gave us such data
as:·
proportion of inventory vs. special order items broken out by
location·
daily and weekly demand curves by location·
proportion of stock transfers from different warehouse locations·
number of stock-out conditionsThe cycle time for the different process steps (HH:MM:SS) were
modeled using data gathered for about a month of normal operations.When we were calculating the impact on delivery time (the
primary CTQ), we used Monte Carlo simulation to show the impact of optimizing
the delivery schedule and changing stock profiles. Crystal Ball software
generated the VERY non-normal distributions of overall delivery time given the
old and new schedules, and old and new stock inventories. We labeled the
resultant histogram with the P5, P95, median and mean values. We also tabulated
the mean value for the stakeholders.If someone wanted to “prove the Improve,” we could
do either a 2-sample t-test (pretty robust to non-normality) or a Mood’ median
test using the exported simulation data. No one asked for such a test, showing
the overlapping histograms was enough for the stakeholders.The distribution from the histogram could also be used to
construct performance metrics for the Service Level Agreement in the manner of,
“85 percent of orders in a week will be delivered in less than 15 hours
from the time of the order to receipt of the item at the worksite.”Cheers, Alastair0 -
AuthorPosts
The forum ‘General’ is closed to new topics and replies.