Tackle problems and improve when data are available and neatly displayed in a Pareto chart is the way everybody expect to go, but this is not so often the case.
How to start in case of greater number of issues but very few data (read recording of sorts) about the problems exist and this few is messy, incomplete and inaccurate?
How to do when collecting new data may take too long and can be deceiving.
How to do when quick action is required ?
>Lisez cet article en français
Measurement is the first improvement step I once wrote in another post, which is still true, but in some circumstances, it’s not possible to set up a dashboard and wait for the first KPIs to come in.
In such a case, one must adapt to the situation and accept it as it is. Available data is scarce, questionable and messy? Just assume the data available is relevant, and then…
1. Use what is handy
Some data may be available and reliable: Production management data is generally well captured and filed as it is connected to P&L. Other data sources may be accurate: invoices or shipment data, for example.
Check what is handy and can be used then go for correlation rather than causation. At this point one needs to find leverage points, not definitive answers.
Ask people who are confronted with the problems. They may give precious indications about their occurrences, conditions, intensity, importance and so on. The best is to conduct an interview without preconceptions and without suggesting answers.
It may be necessary to rephrase and thoroughly check the understanding as the interviewees will use their own words and descriptions.
Ask for the hit parade of problems or “undesirable effects” and use the Interference Diagram.
It may be based on gut feeling and memories but if crosschecked, it may come close to a consensual result. Best is to draw the diagram with a group: people will kind of auto-correct each other, especially people who tend to exaggerate problems in order to get attention to “their” problems.
Again assume what people said is relevant, consider them as subject matter experts.
3. Take samples
Go see and observe, if possible and relevant: measure. The sample don’t have to be statistically meaningful, the objective is still to get a “good enough” picture. Yet too few samples may be biased and too easily criticized by objectors.
When looking at human activities, the work sampling method does fine. It is about tallying the gesture or occupation at the moment the eyes of the observer catch the scene in a prepared form. Tally sheets may do in many cases when looking for samples.
4. Prepare a database
I usually prepare an Excel spreadsheet with columns as data fields and lines as records. When significant amount of lines (e.g. data) are recorded, the analysis can begin with a pivot table.
More sophisticated solutions may be used as well, but time is usually scarce. Furthermore Excel is most often available and many people know enough about it to give a hand.
Study the available data and look for trends, patterns and correlations. When possible, go check the assumptions and reality / likeliness of the findings and discard or confirm scenarios. Keep track of tests and proofs in order to let others scrutinize and assess the analysis.
6. Set up measurements and dashboard
While investigating and coping with few messy data, measurements, data recording and dashboards can be prepared. This is for later confirmation and more accurate analysis in a second step. Yet preparing it soon will yield more data for later use.
At some point the analysis shows leverage points, sources of problems and hopefully helps to find root causes. It is then time to test the assumptions by putting in place solutions and countermeasures.
Each action that alleviates the situation and helps to reduce the complexity is welcome, as it shaves off the bundle.
>Related post: If at least two tell the same