Microsoft offers a few interesting sample datasets. These datasets, provided by obvience.com, are based on real business data that has been anonymized. They are useful to understand how a given analytical approach helps understand what is going on.
THE MICROSOFT DATASET
One is the Supplier Quality Analysis sample. Its comes with a PowerBI dashboard you can download. The sample focuses on one of the typical supply chain challenges: supplier quality analysis. Two primary metrics are at play in this analysis: total number of defects and total downtime in minutes that these defects caused.
To run the dataset in the app, type "supplier quality sample" in the "choose a sample dataset" widget.
If you can't wait, here is a quick preview of the findings.
There has been a significant increase in downtime tied to Rejected defective suppliers, that were almost non-existent the first year. This increase is it tied to more volume (not greater defect severity or change in mix), and is due to "New" volumes.
Some plants and some vendors are doing particularly badly, but the dataset does not have info of whether we are talking simply of "new" plants/vendors or of an actual fall in quality. The problems, in terms of Category, are mostly about the Logistics and Packaging.
VARIANCE ANALYSIS TO UNDERSTAND DEFECTS
Why should one use variance analysis to analyze defects? Variance analysis works on three metrics: revenues (or costs if you want to analyze expenses), units/volumes and prices.
In this context, revenues (rather costs, since they need to be minimized) are the downtime minutes. Volume is the number of defects (again to be minimized). Price is the number minutes of downtime caused by a unit of a given type of defect: defect severity.
Volume variance is the number of downtime minutes given an unchanged defect severity and an unchanged severity mix of defect types.
Price variance is the impact, in downtime minutes, of a change in the severity of a certain type of defect, keeping volumes and severity mix constant.
Mix variance is a measure of the impact on downtime minutes of the change in the relative proportion of the different types of defects, keeping their volumes and severity constant.
The number of minutes of downtime has gone up by 16k from 56 to 72k.
This is equivalent to an increase of over 25%.
The increase seems to be tied to an increased "severity" (impact in terms of downtime) of the defects, rather than to more volumes, that show modest growth.
Logistics and Packaging experience the largest surge of downtime minutes, due respectively to higher severity and increased volumes.
The chart below compares the average weekly values of defect severity (aka unit price), number of defects (aka units) and downtime minutes (aka amount) of the present year (AC) to the corresponding values of the year before (PY).
From the left, the weekly average values are for the year, the most recent six months (compared to the corresponding six months of the previous year), the most recent quarter (compared to the corresponding quarter of the previous year), and the most recent month (compared to the corresponding month of the previous year).
The increase in downtime "amount" appears to have had an impact throughout the year, with a peak in the most recent quarter, driven by a spike in defect "severity" (which, again, is the "price").
The following report shows two dimensions (Category and Defect Type) and confirms what we already suspected: most (17k) of the increase in downtime comes from Rejected supplies. Impact is bad for Logistics but improving in everything else. No Impact is pretty much irrelevant.
We might want to better understand these Rejected 17k minutes of negative variance. One approach is simply plot them. In total Rejected has gone from 5 to 22k, driven mainly by volume.
By Category, Rejected Packaging got much worse driven by more Volume while Rejected Mechanicals got worse driven by severity.
The Skokie Il plant seems to have come out of nowhere with these Rejected defects that it did not have at all the first year.
To capture the interrelations between the different dimensions tied to our 17k Rejected downtime minutes increase of the first result row we can simply summon more detail in terms of Category and Plant:
Info on the Vendors responsible for the downtime increase:
PRICE AND VOLUME VARIANCE
Price variance measures the change in "severity" of same-to-same kind of defects, at constant volume and mix. Note that volume & mix variance is negative, indicating a welcome reduction in downtime minutes tied to volumes, while we have seen that there was in fact a small increase in the volume of defects.
The app will return a different set of combinations if, instead of analyzing total variance, we split variance between price (change of defect severity) and volume & mix.
We get the following "story", that gives us insight also on "Impact" defects:
1️⃣ The vendor Ontotam, either because it is not longer a vendor, or because they have improved, has reduced its number of downtime minutes tied to Impact defects by over 28k. We are talking of Packaging, Unsealed Cartons, for the Rockford plant.
2️⃣ Net of the improvement captured in row (1), downtime tied to Impact defects has increased by 35k minutes due to an increase in the gravity of the issues, as mentioned above.
3️⃣ Rejected product defects, as mentioned, also amounted to more downtime minutes
4️⃣The Plustax supplier also seems to be doing better in Logistics supplies. We might want to check if this is because of a fall in orders or an improvement in quality.
5️⃣ Other than that, other Logistics suppliers have shown an increase in the impact of downtime minutes tied to the volume of defects.
*️⃣ This leaves a positive 12k balance tied to other issues.
The volume has actually gone up, but this has been accompanied by an an improvement in the "gravity mix" of defects. The volume increase in downtime minutes is actually due to more, but on average less serious, defects.
NEW AND LOST VOLUME VARIANCE
Has the volume increase to do with new issues? "New" issues are combinations where at least one element has changed. For example, new suppliers delivering the same product to the same plant, old suppliers delivering new products to a new plant, new suppliers delivering new products to new plants,...
There has been a lot of "bad" new.
Notice that the "mix variance" metric does not capture change in new and lost volume, which remains unchanged when volume & mix is split into "volume" and "mix" metrics.
We might want to give a better look to the impact of "New" combinations on our downtime. The first "Rejected" result seems to be really totally "New" - period zero units are zero.
The second result is more interesting. As the % change value shows, overall this combination (Category=Logistics, Plant=Springfield, Defect Type=Impact, Material Type=Corrugate, Vendor= Reddoit) has seen a healthy 37% decrease. However, if we consider also the "missing" dimensions, namely Defect, there has been an increase in defects: a "New" type of defect not found before in this specific combination.
Indeed we find a "New" defect for the Reddoit Vendor with the already seen combinations: "Poor Pairing" .
Why we have a "New" type of defect might be worth investigating.
ANNEX - VARIANCE BY DIMENSION
In this section the data is analyzed with a standard slicing by dimension approach.
The dataset contains six different dimensions.
Defect Type is important. Defective supplies can: (i) have "No Impact" in terms of downtime - best case -, (ii) be "Rejected", caught before entering the production flow and sent back to the supplier - middle case - (ii) have negative "Impact" in terms of production downtime - worse case.
Note that the small multiple charts are ordered by decreasing amount of change.
The increase of downtime comes from "Rejected" defective deliveries. This is somewhat surprising since these problems get spotted before the defective products go into the production line, and therefore should have a limited impact.
The growth of downtime tied to Rejected is an issue of higher volumes. The impact of Rejected defects has in fact more than quadrupled.
The blue Rejected bubble has come from nowhere, while increase in "price" of the black Impact bubble has been compensated by a loss in volumes.
The increasing severity of the "Impact" defects is compensated by their reduction in volume/mix. No Impact defects have, as their name dictates, no impact, and are decreasing in either number or mix.
The increase in "severity" of Impact defects, at nearly constant volumes, is compensated by a marked improvement in their "mix". In other words, there is now a greater proportion of "less serious" defects than before - even if the seriousness of each single defect type has gone up.
Downtime tied to new volume is an issue for both Impact and Rejected. However, for Impact it is compensated by a reduction of lost volume, which is unfortunately not the case with Rejected.
VARIANCE BY PLANT
Another dimension we might want to look at is Plant. The Skokie plant and the Bangor plant alone represent 15k minutes of increase in downtime.
Note that these two plants are not among the top four plants by downtime. Notice how the improvement of Springfield is tied to a fall in units accompanied by an substantial increase in defect severity.
The negative variance of the Skokie and Bangor plants is tied to increase in volumes. We could further drill down by defect type to understand more.
VARIANCE BY VENDOR
A third dimension is Supplier. We have four suppliers that went from very low downtime impact to over 2k hours of downtime each.
All the other suppliers together increased less. We should investigate if these four "bad" suppliers are in fact really "bad", or they are simply "new" suppliers that last year had much lower sales.
None of these four vendors make the top fours slots in terms of total downtime.
The increase in downtime is tied to more volume. Note that the lack of a variance mix value for Planethouse and J-Lax might be due to these two vendors delivering only one kind of product, and therefore having only one kind of defect, with no possible change of "mix".
VARIANCE BY CATEGORY
Another dimension is Category. Most of the increase of downtime is due to Logistics and Packaging.
Here is the split in terms of price, volume and mix.
The slope chart by Category confirms that significant downtime is in Logistics, due to a hike in the average "price" (severity) of the incidents. Mechanicals is doing even worse, in terms of average severity, but its volume is going down, balancing things out.
The "average weekly values" chart shows how, overall, Mechanicals generated the same downtime impact in the two years. However, more recently, it is doing a lot worse. Logistics and Packaging have performed badly. Packaging seems to be getting better.