Pradeep Kandpal

Diagnostic Analytics is one of the data analytics techniques that analyses a dataset to arrive at root causes of events, behaviours, and outcomes.  It is primarily conducted to provide insights on various factors that are responsible for a problem at hand and tends to uncover the “WHY” behind the data.  The data source, quality and reliability is paramount while conducting a Diagnostic Analysis.
 
Diagnostic Analytics primarily represents the Current State in a problem-solving domain which connects the dots between Descriptive Analytics (what is wrong?) and Predictive Analytics (what is likely to happen?).  The findings of these provide further insights on Prescriptive Analytics (Future Course of Action).
 
In a DMAIC framework of six sigma, maximum value can be derived from Diagnostic Analytics in the Analyze phase. 
 
Examples & Use Cases:
 
RCA Techniques: 5 WHYs & Pareto analysis to find out of the root cause - For e.g. A 5-why analysis revealed increased usage of UPI transactions to be the root cause of CASA ratio decline in a leading bank.  A pareto analysis showed that discounted products which correspond to 20% of the overall merchandise are contributing to around 80% of the sales.  Clinical Diagnostic tests use patient's tests results data to generate a complete summary based on the insights derived post comparing it against the standard and also against patient's past data. The physician in turn could do RCA to derive meaningful conclusions as to why this is happening. Hypothesis testing: To test an assumption that better wages outside is contributing the most to the attrition in a leading organization. A sample of exit interview data was subjected to a statistical test (1 proportion test).  The test result was found to be statistically and practically irrelevant and rejected the assumption. Correlation & Regression Analysis: Many stock broking platforms have built-in algorithms based on pattern recognition, correlation & regression analysis to derive meaningful conclusions so that their investors can make informed decisions. Anomaly Detection: Network analysis make use of built-in control charts to detect any anomalies that may shed further light on the assignable causes of frequent downtimes and network jams.  

All "known risks" before manifesting themselves as issues/problems would've been unknown once.  Such known risks after obtaining valuable insights from key stakeholders, domain experts and SMEs become part of a document known as Risk Register. In project management, this Risk Register is an output of a process called as "Identify Risk". 
 
The Risk Register is generated to document all the identified risks both positive (opportunities) and negative (threats) along with other details which more or less fall under the following heads:
 
 
Criteria to Enter a Risk Item:  It is important to identify as many risks, as early as possible at the beginning of a process or a project lifecycle so that you are better prepared to handle any unforeseen circumstances that might announce themselves as critical issues in the later stages.  Any uncertain event that if occurs results in a negative or positive impact on one or more parts of the project or a process can be a part of the risk register.
 
Criteria to Remove a Risk Item:  Irrespective of the number of risks identified at the beginning of the process or a project, the old risks might lose their sheen and are not relevant enough, probability and impact wise to be a part of the risk register.  Such risks that do not show any signs of manifesting themselves as issues or problems after having reached their corresponding risk triggers can be removed from the risk register.
 
Application of Risk Register in Process Management: 
 
In a process management, the Risk Register can be a source of all the identified risks so far.  These are then evaluated and rated with domain experts and SMEs on the following parameters:
 
>> Urgency - The speed with the risk response is to be applied.
>> Proximity - How imminent the risk is?
>> Dormancy - How soon do we feel/discover the impact of the risk once it would occur?
>> Manageability - How easily the risk response can be implemented?
>> Controllability - The degree to which the outcome of risk can be controlled?
>> Detectability - Ease with which it would be known that the risk is about to happen.
>> Connectivity - The degree to which a risk is connected to other risks. More the connectivity, higher the criticality.
>>Strategic Impact - If the risk has an impact that would affect the overall strategy of the organization.
>> Propinquity - Perception of the end user - Criticality of the risks perceived by a key stakeholder/end user.
 
After all the risks are passed through these parameters, a list of prioritize risks is generated.
 
Of these prioritized risks, negative risks can be addressed by using popular Risk Management tools such as FMEAs and Probability and Impact Matrix with a view towards reducing the probability and impact of those risks.  The more detailed and up to date your risk register is, the more value your can derive from these tools. 
 
For positive risks, all efforts should be made to increase the probability and impact of these risks.  Once these risks manifest, then they should be exploited to derive maximum value out of them. 
 
For Example:
1). Risk register containing a list of identified risks before rolling out a completely new process.
2). During process re-engineering all possible areas where things could go wrong can be listed as risks in a risk register.
3). In an established process, a risk register should be constanly refined by constantly re-visiting the risk items. New risks should be included, if any and old risks should be removed.
4)  During the pilot process in the improve phase of a DMAIC project, all risks can be included in the risk register.
5) The assumptions and constraints that are listed before rolling out a new process should be constantly revisited to check if any of them are leaning towards becoming a risk can be thus included in the risk register.  
 
Conclusion:  From a process management perspective, it is also important to note that risk identification in and of itself is an iterative process where new risks might get added to the risk register if the process steps are modified due to updates from the client or as a result of a process improvement.  Similarly, old and irrelevant risks might be removed from the risk register as the process matures.
 
 
 
 
 

A Placket-Burman design exhibits the following differences when compared to a regular 2-level design:
 
Plackett-Burman Designs
Regular 2 -Level Designs
1.  These are highly fractional resolution III designs where the main effects are confounded with the 2-way interactions.
1.  These could be either full factorial designs or fractional factorial resolution IV or V designs where there is no confounding of main effects with 2-way or other higher order interactions.
2.  Only used to screen main factors from a long list of potential factors.
2.  These are used to screen as well as to optimize the main factors.
3.  The number of experimental runs in these designs are in multiples of 4 starting from 4 to 128 and the number of factors has to be one less than the number of runs.  For e.g., A design with 32 runs can screen main effects of up to 31 factors.
3.  The number of experimental runs in these designs is expressed as 2k (full factorial) and 2k-n (fractional factorial) where k denotes number of factors and n could be 1, 2 or 3 for 1/2 factorial, 1/4 factorial or 1/8 factorial designs.
4. 7 factors with 2 levels would warrant just 12 runs in a PB design as show below:
 
 
Design Summary:
Factors:    7    Replicates:    1
Base runs:    12    Total runs:    12
Base blocks:    1    Total blocks:    1 
4. 7 factors with 2 levels would require 128 runs for a full factorial and 64 runs in a fractional factorial as shown below:
Design Summary - Full factorial
Factors:    7    Base Design:    7, 128
Runs:    128    Replicates:    1
Blocks:    1    Center pts (total):    0
Design Summary - Fractional factorial
Factors:    7    Base Design:    7, 64   Resolution:    VII
Runs:    64    Replicates:  1    Fraction:  1/2
Blocks:    1    Center pts (total):    0          

  
 
Benefits of Plackett-Burman Designs:
 
·         Plackett-Burman designs are usually helpful when we have a significantly large number of factors, and it is not economically feasible to conduct even 1/4 or 1/8 factorial designs.
·         Not only does it saves time and money but provides us with vital few from trivial many which can then be optimized further via full factorial or resolution IV or V fractional factorial designs.
 
Limitations of Plackett-Burman Designs:
 
·         These are only helpful early on in a study when we have limited knowledge about the overall study and there are large number of factors to choose from.
·         Due to them being highly fractional, it is not advisable to use them for studying interaction effects.
·         Recommended to be used only when there are negligible 2-way interactions.  The results of these designs to identify main effects are not reliable when two-way or other higher order interactions are present.
 
Examples of Plackett-Burman Designs:
 
·         Clinical Research – There are several potential factors that are at a play at the initial stages of a drug development against a diesease and PB designs come to its rescue by weeding out the insignificant factors. Once the main interactions are identified, the drug resistance is measured against those by deploying a full factorial or a fractional factorial design.
 
·         FMCGs – PB designs are used in the initial stages of a product development for screening significant factors from a list of potential factors that can be further studied and optimized with an eye towards enhanced customer experience and increased market share.
 
·         Agriculture:  Plant fertilizer industry always strives to enhance the effectiveness of their plant foods by investing a considerable time and effort on optimizing the factors that are crtitical to the crop yield and quality. PB designs helps them to narrow down to these vita few from trivial many.
 
 

Both Transportation Waste and Motion Waste being two of the eight wastes of lean, may exhibit the following differences:
 
Transportation Waste
Motion Waste
1)  Any movement of material and information between process steps, workstations and plants that does not add or create value.  
 
2).  May cause information dilution resulting in loss or misrepresentation of vital facts. Damage of materials while in transit might also ensue thereby adding to further costs.
 
 
 3) Examples:
Service Industry:
a).  Interdepartmental e-mail chains cc’d to more than required individuals with unnecessary attachments.
b).  Multiple processing levels in a transaction processing unit. Processors, evaluators, QA1, QA2s etc.
c).  Multiple escalations in a call centre rather than FTRs.
 
 Manufacturing Industry:
a).  Unnecessary movement of goods from warehouse to production floor and vice versa.
b).  Ordering of raw materials from a distant vendor when nearby options were available.
c).  Delivery routes that add to inefficiencies.
 
1).  Any movement of people and equipment within process steps, workstations and plants that does not add or create value.
 
2)  Occupational injuries might ensue due to stress and strain caused by overuse of certain muscles and body parts leading to too many sick leaves and absenteeism.  Early wear and tear of machinery could also be a consequence of this.
 
 3) Examples:
Service Industry:
a).  Using way too many keystrokes and mouse clicks for simpler daily tasks rather using shortcut keys.
b).  Taking time to search for vital information across multiple folders or systems.
c).  Way too many unproductive meetings and discussions within teams.
 
 
Manufacturing Industry:
a).  Taking time to find the right tool every time the need arises.
b).  Unnecessary human and machine movements than required to finish a task.
c).  Reaching for tools that are at a distance.
 
 
Identification of Wastes Arising from Unnecessary Motion and Transport:
The first step to eliminate inefficiencies arising out of these wastes is to identify the reason behind these wastes.  Poor layout of equipment and machinery, large batch sizes, absence of updated SOPs/Standardized Work, inefficient staff, non-calibrated machineries are one of the few primary reasons for both transportation and motion wastes.  Both these wastes can be detected through effective Value Stream Mapping and Gemba Walks.
 
How to Prevent Transportation Waste:
1.            Once inefficiencies are identified, implement 5S.  This would eliminate and minimize most of the wastes cropping up due to transportation.
2.            Optimize the usage of remote assistance tools for troubleshooting rather than physical addressal of mundane technical issues.
3.            Minimize or eliminate use of paper trails.  Maintain a central repository for knowledge sharing and lesson learned.
4.            Empower team members so that they are self-organized requiring minimal supervision and strive towards developing a team with T-shaped skills.  This would maximize the FTRs.
5.            Emphasize more on cellular layouts than functional layouts.
6.            Reduce batch sizes by implementing SMEDs.
7.            Use continuous flow where possible.
 
How to Prevent Motion Waste:
1.            Implementing 5S can eliminate wastes due to unnecessary motions.
2.            Develop Standardized Work and ensure team adheres to same.  This would eliminate inefficiencies arising out of unnecessary motions.
3.            Use VA/NVA analysis within processes to eliminate NVAs thus eliminating unproductive movements.
4.            Identify opportunities for automation and use appropriate automation tools.
5.            Ensure that the individuals and machineries are calibrated to avoid unacceptable work or scrap.
6.            Ensure that the team meetings and briefings are time-boxed effectively.
 
Conclusion:  Despite having these basic differences, both motion and transportation wastes to an extent go hand in hand.  The unnecessary movement of material and information caused by transportation waste more often than not results in unnecessary movement of workers and individuals too resulting in a motion waste.

Both overproduction and overprocessing are two of the 8 wastes according to Lean.
 
Overproduction:  Producing sooner and faster than required and in more quantities than needed is overproduction.  Overproduction results in waste of time, labor and materials thereby creating too much inventory which results in extra cost.  If the overproduced product is seasonal, it would either end up as a scrap or would add to the storage cost and if it is perishable then it would have to be discarded.  Poor estimation of customer demands often results in overproduction.
 
Examples of Overproduction:
Service Industry:  a)  Huge meals in the restaurants.  b)  Creating way too many reports and records than actually required. c) More number of beds in the hospital than required.
Manufacturing Industry:  a).  Keeping labor and material on a standby.  b)  Warehouses filled with overproduced unsold goods.
 
How to Prevent Overproduction:
1). Using TAKT time to gauge customer’s demands.
2). Practicing Just-In-Time inventory management.
3). Using production levelling by both Volume and Type.
4). Doing production control by using supermarkets where continuous flow is not possible.
5). Having a production scheduling point called as a “Pacemaker Process” in place preferably towards the end of the production line that signals a pull to the upstream processes for production thus enabling a continuous flow for the downstream processes.
 
Overprocessing:  Any additional work that does not add any more value to a product than expected is called overprocessing.  Overprocessing often results in a reduced overall equipment and people effectiveness. The primary reason for overprocessing is not having a common and clear understanding on the critical to quality and critical to acceptance parameters of the end product.
 
Examples of Overprocessing:
Service Industry:  a). Many approval levels warranting quite a few signatures on a document for even smaller requests.  b). Recommending too many diagnostic tests to patients. c). Entering same data at multiple places. 
Manufacturing Industry:  a).  Using advanced machinery for a product that could have been easily produced using a basic machinery.  b)  Adding more attributes/features in a product than actually needed.
 
 How to Prevent Overprocessing:
1).  To have a clear understanding and communication on the critical customer requirements and updates.
2).  By doing VA/NVA analysis for each activity and develop a Standardized Work for the entire process. This should be done in an iterative manner for the entire process flow as more activities might get added due to changes in customer requirements.
3).  Keeping it simple by eliminating complexities that arise out of excessive documentation, instructions, manuals etc.
4). Automate the process steps where possible to reduce or eliminate the chances of overprocessing.
5). Revisiting steps 1 through 4 periodically to ensure sustenance.

Both the approaches are used to address the fluctuating demands of the customers and ultimately aim to eliminate waste, reduce lead time and achieve an overall equipment and people effectiveness.  A few comparisons are listed below:
 
Demand Levelling
Production Levelling
1).  A carefully thought-out proactive approach that influences the demand itself of the customer to arrive at a more stable and predictable demand pattern that drives a levelled production.
1). Popularly known as “Heijunka”, it is a mixture of both proactive and reactive approaches in which the production schedule, based on predicted customers’ demands, is fine-tuned in such a way that there no overburden on the systems, resources and equipment while producing products in a consistent manner.
2).  An example in manufacturing industry could be – Build-to-Order approach especially used in automobile industry to arrive at a predictable near-approximate demands.
 
In service industry, concept of happy-hours in restaurants and pubs can be used to level the demand surges in peak hours.  Another example, could be passport generation in Passport Seva Kendra, where demand is levelled by allocating appointment slots.
2).  An example here could be a shoe company producing shoe types A, B, C, and D averages weekly demands of A (5), B (3), C (2), D (2).
 
A mass manufacturer with apparent changeover challenges, interested in economies of scale would follow the following sequence – AAAAABBBCCDD – known as levelling by volume.
 
On the other hand, a lean manufacturer who want to leverage the benefits of a product type along with volume may want to follow this sequence –AABCDAABCDAB - known as levelling by type.
3). Preferable at the beginning of lean implementation to gauge the demand levels of the customers.
3).  Is mostly used towards the later stages of lean implementation once value-streams are finalized and takt time is known. A final production schedule is made visible by the use of Heijunka Box.
4).  Not feasible in situations when there is rush of demands due to emergencies, pandemic, and low-price high-volume scenarios.
4).  Less effective in situations where there are infrastructure and resource challenges to carry out SMED especially when the manufacturer intends to level the production by the product type. 
 
Vital Trade-offs:  Despite being two different concepts, they both complement each other in a variety of ways.  In situations where there are limitations in implementing production levelling due to various capacity constraints, demand levelling is done to meet customer’s demands by modifying its various product offerings and by triggering a change in the way the customers place their orders.  The insights obtained from this subsequently feeds into a production/service schedule thereby enabling Heijunka.
 
Similarly, where demand levelling is not possible, a TAKT time provides an approximation on the customer demands and drives the overall production schedule where customer’s requirements are fulfilled via small batches (levelling by volume and type), single-minute exchange of die (SMED) and standardized work.
 
Limitations:  Despite all its benefits, Heijunka to an extent walks a tightrope by trading inventory or lead times for stability and is a short-term workaround intended to smoothen the crests and valleys of customer’s demands.  Another limitation is that it is responsive only to moderate demand fluctuations.  Unusual variations in demand need more extreme measures.
 
With most of the organizations these days moving towards agility by taking steps towards creating a more responsive production system that is more flexible and could cater to varying levels of customer’s demands, Heijunka more or less limits oneself to an approach that revolves around various constraints.  You never know when the swings in demands which are perceived as a constraint in Heijunka might be perceived as an opportunity by an agile competitor that is ready to pounce on it with its advanced tools and techniques. A workforce with T-shaped skills, working in a cellular layout with an adaptive production schedule could be a starting step to exploit these constraints.

Before launching a new product or improving its existing products, companies often are in a state of dilemma when they have several features and components to choose from to have a competitive edge in the market.  This type of indecisiveness can be addressed through a Conjoint Analysis. 
 
Conjoint Analysis also referred to as a trade-off analysis is a survey-based research technique which organizations use to gauge the interest level of customers when they are presented a product with different permutations and combinations of features.  A statistical analysis is performed on the data derived from this survey.
 
Relationship with Design of Experiments:  Conjoint analysis crosses over Design of Experiments in many ways and use the same concepts of Runs (Profiles, Cards, Stimuli, & Panels), Experiments (Studies, Investigations), Factors (Attributes, Features), their corresponding Effects (part-worth score) on the response variable. The only difference is that in Conjoint analysis, it is assumed that only main effects are significant and two-way and other higher-order interactions are insignificant, so they use highly fractionated factorial designs.
 
Types of Conjoint Analysis:  The two most commonly used Conjoint analysis are as follows:
 
Choice Based Conjoint (CBC) Analysis: - Most commonly used conjoint analysis in which a customer is asked to respond to a combination or levels of features in a product also called as full profile product.  This is a preferable method when the number of attributes are up to 7-8.  For e.g. the following are the various combinations of features that a smartphone company wants responses to:
 
Attributes/Factors
Level 1
Level 2
Level 3
RAM
4 Gb
6 Gb
8Gb
Storage
80Gb
128 Gb
256 Gb
Battery
5000mAh
4000mAh
4500mAh
Display
6.4 inch
6.70-inch
7.0-inch
Camera
16-megapixel
32-megapixel
32-megapixel
Price
Rs. 15000
Rs. 18000
Rs. 20000
Expandable Storage
Yes
Yes
No
 
 
 
 
 
 
 
Adaptive Choice-Based Conjoint Analysis (ACBC):  In cases where the attributes are more than 8, adaptive conjoint analysis is performed.  In this case the subsequent set of questions in the survey is modified based on the responses to the previous set of questions.  This type of analysis helps where responses are required against too many attributes and their corresponding levels. Responses to the previous set of questions provide a guideline for customizing the next half of the questions thereby reducing the number of questions significantly and yet extracting the most valuable insights from the respondents in less amount of time before the respondents gets cognitively tired. This type of conjoint analysis is more interactive and more engaging with the customers. It consists of the following 3 steps:
 
Step 1 – Build Your Own:  For a product of interest, against each feature the respondents are asked to select their preferred level.  The cost for the selected level auto-populates.
Step 2 – Screening:  The respondents are presented with a combination of features based on their response to step 1 and asked for their preferred choice.  
Step 3 –Choice Task:  Based on their response to step 2, the respondents are provided with a set of attributes and their corresponding levels and asked for their preferred choice.
 
How are the survey results analyzed?
 
Once the survey results are populated, the data is fed into a statistical software and each level of an attribute is assigned a score based on its weighted preference called as a Part-Worth Utility Score.  Higher the score, more the chances of the feature being included in the final product.  If we sum up all the part-worth utility scores of a level for each attribute it would give us the Total Utility Score for the entire product. The table below shows an illustration for same.
 
Attributes/Factors
Level 1
Part-Worth Utility Score
Level 2
Part-Worth Utility Score
Level 3
Part-Worth Utility Score
RAM
4 Gb
1.7
6 Gb
1.7
8Gb
2
Storage
80Gb
2.0
128 Gb
2.0
256 Gb
2.5
Battery
5000mAh
2.0
4000mAh
3.0
4500mAh
2.0
Display
6.4 inch
1.2
6.70-inch
1.5
7.0-inch
1.7
Camera
16-megapixel
2.1
32-megapixel
2.8
32-megapixel
2.8
Price
Rs. 15000
3.5
Rs. 18000
4.5
Rs. 20000
2.5
Expandable Storage
Yes
2.2
Yes
2.2
No
1
Total Utility Score
 
14.7
 
17.4
 
14.5
 
 
 
 
 
 
 
 
From the above illustration, it is evident that a smartphone with Level 2 Attributes appeals the most to the customers.
 
For Adaptive Choice-Based Conjoint Analysis, a utility score is also calculated for each respondent as well along with the levels of attributes.
 
Conclusion:
 
The Total Utility Score thus arrived at by the above analysis, can be a valuable input to determine the preference market share for the product of interest. It would also provide insights into the most optimal sales offer that the company could go for. The organizations use the results of the analysis to understand the price elasticity of the product, optimal price point, demand forecasting, sensitivity to a brand name and how much a customer is willing to pay for a new service or a new feature.  A product thus developed with optimized product features and components drives its value proposition in the market.

It would depend on whether we are doing a “Screening Design” or an “Optimization Design.” 
If the criticality of the 5 factors is yet not established, then we would go ahead with a Resolution V fractional factorial design.  Out of the 5 factors, the factors with significant main effects can be further considered for a full factorial optimization design.
 
Half Factorial - Screening Design: If the significance of the given 5 factors is questionable and are not yet validated as critical Xs, we can use a resolution V design to screen out non-critical factors.  Half factorial experiments for screening are used primarily in two scenarios:
1)     The existing historical data is inconclusive.
2)     There is no historical data available at all.
 
Impact on Time, Resources, and Complexity:  A design summary of a fractional-factorial experiment with 5 factors and 2 levels with and without replication is shown below.
Without replication

With replication

A fractional factorial design even with replication would require 32 runs which is almost half of a full factorial.  Complexity also would be comparatively less as we would only be focusing primarily on the main effects and not the interactions.  Since the effort is less too, fewer resources would be deployed.
 
Full Factorial - Optimization Design – If all of these 5 factors are found to be critical, then we may want to optimize their behavior towards the response variable by conducting a full factorial experiment with replication.  The following is the design summary for same:
Impact on Time, Resources, and Complexity:  A design summary of a full factorial experiment with 5 factors and 2 levels with replication is shown below.

In comparison to a fractional factorial design, a full factorial experiment would be more time consuming as the number of runs would be more.  We would have to conduct 64 runs with replication for a full factorial experiment.  If any of these 5 factors warrant an addition of a centre point to rule out curvilinearity, then a few more runs would be added to it.  Complexity would increase as interaction effects are also to be studied along with main effects and would necessitate utilization of more resources due to a sizeable number of experiments.  We need to also factor in the Scope, Time, Cost and Resource constraints while conducting a full factorial experiment.  Most of the R&D departments with higher risk appetite usually proceed with full factorial as they have to always come up with a robust design.