Process Quality Management in Localization

Now I would like to add one more article "in Localization" series. The topic is about lean six sigma implementation in translation and localization. As I became familiar with the topic, it opened new horizons for me, and it caused me to question some of my methods and approaches. The topic is too broad to cover everything in an article. Before moving forward, I would be glad to hear feedback from the readers who have experience in the implementation of Lean Six Sigma in their localization and translation processes.

When I first started to study Translation and Interpreting Studies at Bogazici University, the very first thing I learned was the "meaning" of translation: It is an end product and it also refers to the process of creating the end product. While focusing on the end product, professionals and companies can miss the process. However, quality work from an unqualified process is either coincidence or dependent on individual effort. So, the reputation can easily slip away.

Two years ago, our company and my manager offered a great opportunity for me to get familiar with Lean Six Sigma and Business Excellence. A close colleague of mine - the Business Excellence initiator encouraged me to study and gain experience on the topics further and luckily I had the opportunity to receive coaching from a black belt master who is very eager to teach and to apply the topic in different industries who suffered from translation quality as a buyer in the automotive industry in his previous occupation. I am thankful to them.

Before going into details and my thoughts, I would like to share the key terms: Standards, Lean, Six Sigma, Lean Six Sigma (LSS) and Continuous Improvement Process. Before understanding the key concepts, one should understand how key concepts complement each other. Without establishing standards, you cannot apply LSS methodologies or you cannot evaluate how well your standards are without LSS methodologies. 

The translation industry is already of a standard: ISO 17100. The standards help us do the same thing in the same way in every trial and set the basic quality requirements. It is believed that the quality of work improves if you standardize your processes. It frames the education that a translator should receive, a process that should be followed in every project, and professionals in the industry. However, it does not measure effectiveness and efficiency. So, this point remains open. For some people, LQA is used as a tool. However, I will share my concerns later based on some what-if questions.

Terminology Basis

I will not go into the history of Lean and Lean Six Sigma Methodology.

Firstly, I would like to explain Lean, Six Sigma and Lean Six Sigma. Although they are key terms, all of them are separate business processes and production approaches.

Wikipedia’s definition of Lean quoted from Womak and Jones’ book of “Lean Thinking: Banish Waste and Create Wealth in Your Corporation” clearly sets the limits for the term: "precisely specify value by specific product, identify the value stream for each product, make value flow without interruptions, let customer pull value from the producer, and pursue perfection.” (Womack and Jones 1996 p10)

Six Sigma is a set of techniques and tools to improve quality through process improvement. "Lean Six Sigma unites tools and techniques from Lean and Six Sigma methodologies to produce real results. It uncovers process waste, reduces non-value adding activity, and increases organizational capability." Six Sigma defines a level of quality for your business processes. This is called the sigma level. The highest level in Lean Six Sigma is “6”. A process with 6 sigma level produces 3,4 defects when the process is run 1.000.000 times, which means 99,99966% error-free process. 

Who is TIM WOODS?

Lean Six Sigma methodology focus on the removal of wastes. It is believed when wastes and redundant tasks, activities are removed, a process can be lean and cost-effective. TIM WOODS is an acronym for wastes that are aimed for removal via LSS methodologies.

T: It stands for transportation. It is not something we experience at translation. Maybe we can eliminate time lost for transportation of in-house translators

I: The letter represents the Inventory. Inventory is redundant and not received by the customer. For example, if we use an online system, we should consider archiving some old data which may lead to the slow-down of the system.

M: M stands for motion. Luckily, translation world has been digitalized for more than a decade. However, redundant email communication can be categorized as motion. So instead of using a query system accessible by all translators, if you prefer to communicate with each translator one by one, it is motion.

W: The letter is shortened for Waiting time, which is the time you keep your resources reserved and your resources cannot do any work. This is rather for project managers. We should be clear with our heads-up. If your resources remain with no work, then you should consider optimizing your workload management.

O: It refers to Over-processing. It occurs when work is done more than necessary. For example, your customer needs a translation just to understand the text, not for publishing. And when you prefer to follow three stages of translation: translation, editing, and proofreading, it is over-processing. Or if you add automated QA after every stage of translation, it can be regarded as over-processing.

O: Another “o” stands for Over-production. It may not fit into translation. But imagine that your customer needs QA of 13 hours of testing, but you carry out more than 13 hours. Most probably your customer may not pay more than 13 hours. In some cultures, it can be even perceived as offensive by some customers.

D: It is Defects: Mistakes and errors that need to be reworked. For example, you are managing a multilingual project. You forget to add one language. To add one more language, you have to follow the steps to create a project with one language if your system does not allow you to add one more language afterward.

S: It stands for Skills. Not using the skills and capabilities of your workers fully.

Continuous Improvement Process (CIP) Projects

The term sounds so marketing oriented. However, it is not. CIP falls under business excellence within operations management. The assumption is that there is always a place for improvement. To run a CIP, you should define a problem, measure it, analyze it, improve it, and finally implement a control plan whether it works or not. Please note that CIP projects may aim to shorten the process. But it may also serve to quality, thus may require adding extra steps such as filling checklists, etc.

Tools Learnt from Lean Six Sigma

Voice of The Customer

First, the voice of the customer (VoC) plays a key role in lean six sigma methodologies. We all do translation to earn money. So, we should listen to our customers. Output of VoC is very key and results trigger other tools.

Critical to Quality Tree

Another useful tool to device is the CTQ tree. In CTQ tree, we define need, driver and CTQ requirements. CTQ requirements should be measured. Needs is most of the time fuzzy. Sometimes customers say, “bad translation” – ah sorry, recently we have popular idiom among customers: “google translate”.

A famous example of the CTQ tree is the coffee quality tree

Image credit: https://www.latestquality.com/lean-articles/page/11/

We cannot measure “goodness” of the coffee, but we can measure how good it is via CTQ requirements: Acidity, temperature, richness, price. We can do a small survey and ask what our customers’ expectations from good quality of coffee”. Likewise, we can ask “what good translation is”. This will help us to understand what our customers need. Once I remember to have visited one of our key customers in my previous companies after negative feedback. During the meeting, they showed us their changes and it turned out that the customer wanted their list of terms instead of generic terms. In terms of creativity, grammar, style, they were totally happy, but all fuss was about their terms. They were right. They were aiming to set the standards in their market.

Kano Model

Another tool is the Kano model. Kano model is named after the inventor, Dr. Noriaki Kano. He defined three levels of customer expectations that have a direct impact on customer satisfaction. There are two dimensions: satisfaction (vertical axis) and achievement (horizontal axis).

Image credit: https://asq.org/quality-resources/kano-model

Expected needs are “must” requirements. They do not have to be spoken. They are sine-qua-non. For example, grammatical sentences are a must-have of translation. No customer should instruct translators to write grammatically.

Normal needs are extracted from customer surveys. They are spoken. Customers tell these to us. For example, usage of expected terms is normal needs (or it depends.)

Exciting needs are also called delighters. For example, you share translation memories, customers can track the translation status, customers do not have to copy and paste into word/excel to have the content translated, you just deliver in original format to eliminate the re-creation process. These can be delighters.

Kano model is used to prioritize the customer’s needs. First of you need to focus on expected needs. Delivering delighters without expected needs will not make customers happy.

SIPOC

SIPOC stands for Supply, Input, Process, Output and Customer. You draw your process at a granular level, map inputs, outputs and process stages

Please note that the customer here does not have to be stakeholder paying for the output. The customer is the recipient of that output. So actual customers (paying) can be suppliers as well. For example, in a team, a manager requesting a report can be the customer. On the other hand, the customer of the team manager can be a higher level of manager whom the manager will give a presentation to. The reviewer can be both a supplier in front of the project manager and a customer of the translator. So, the provider of input is regarded as a supplier and the recipient of output is regarded as a customer.

SIPOC is generally used to map the process, identify the stakeholders and roles, inputs that may have an impact on the final product and outputs that lead to the final product.

Defining Value

In LSS, another approach I find useful is to define value-added, value-enabling and non-value-added steps in the process. The definition of these steps helps us to detect the wastes, look for the automation and simplification opportunities in the processes.

Value added steps are the steps that are recognized by the customer. Usage of correct style and terminology, providing translation within budget are value added activities and they have an impact on the perception of the customer. Value enabling (non-value added but required) activities help us to carry out value added activities. For example, proofreading is value enabling for customers (buyers who are not the translation department). They only need proper and correct translation. They are not interested in how you achieve this. Non-value-added activities fail to meet one or more of the following criteria:

1.      It transforms the product (output has a different shape, form, content…)

2.      Step is not rework step (output is flawless once the step is finalized)

3.      Customer cares (or willing to pay for it)

So, translating with tool A or B does not have an impact on customer decisions or transforms the final product. Delivery of regular translation requests via email or platform can be also non-value add activity. However, the example should lead you to think that the concept is a barricade in front of improvement. The selection of the tool can be converted to a value enabling activity. For example, delivery via email triggered by a project manager or from the system does not have any impact on customer perception. So, automating this task through a system will be the removal of one non-value-added task.

Cause and Effect Matrix

Cause and Effect Matrix (C&EM) is one of the most effective tools in LSS methodology. And it is easy to facilitate.

In one column, you list all the inputs within the process after SIPOC. And you add CTQ requirements and outputs for each column. Each column head (output or CTQ) has a weight of 1 to 10. And then you start evaluating the impact of input on each output and you score 1, 3, 6 and 9. 1 is negligible impact and 9 is critical. And after the evaluation of each input, you focus on the most important inputs.

As you will see above, we assign scores for the impact of each input on each output. And score shows us the critical inputs. First, it is a robust tool because it helps you prioritize the important inputs to focus. Additionally, you must bring all team members in one room for discussion. Please note that you should not ask opinion of individual members. You should facilitate the discussion. So, it is also a communication tool.

All the tools explained so far are robust and effective tools. It visualize the problem and solution. However, the article has not so far guided you where or at what stage you can use them. In the following section, I will introduce steps to follow for CIP projects. I will help you map the tool and the stages of LSS.

Root Cause Analysis and Corrective and Preventive Action

There are several Root Cause Analysis methodologies such as Fishbone and 5 Whys. Root Causes Analysis facilitates the communication and brainstorming. It should not be carried out individually but as a team.

Each root cause analysis session starts with the definition of a problem, which is also known as the problem statement.

Fishbone is also known as Ishikawa Diagram named after Kaoru Ishikawa. The problem is the head of the fish. And you list the causes under Equipment, Process, People, Materials, Environment and Management. 

Image credit: By FabianLange at de.wikipedia - Translated from en:File:Ursache_Wirkung_Diagramm_allgemein.svg, GFDL, https://commons.wikimedia.org/w/index.php?curid=6444290

5Whys is another robust tool. You are expected to ask at least 5 why questions until you reach to the root cause. You may question how you will understand the root cause. Last time when you give a reasonable answer or the first time your answer to why is unreasonable shows the root cause. You can keep on asking why until you reach to “no answer”.

Non-Conformance Cost (NCC) and Defect Forms

First, to err is human. What a correct phrase? We all do mistakes and learn lessons. Everyone can make mistakes, pilots, doctors, drivers, mothers, fathers and teachers. Some mistakes (or defects in LSS terms) are irreversible and some are not. The point is to learn afterward. In aviation, after every crash, pilots get safety bulletins about the incident or accident. Companies should encourage to log their defects, these are not to finger point anyone, but to learn as a community. The material cost of the defects should be calculated. All relevant stakeholders should discuss defects. If a defect is frequent, this may imply that there is a problem in your process. If the same person makes the same mistake continually, then we should focus on why they make. Punishment will just terrorize people and they may do more mistakes. The reason for mistakes can be various: The need for training, lack of onboarding, workload, insufficient skills, and so on. For example, if the person lacks relevant skills, we should consider taking a good look at our recruitment or assignment process. Or if the problem is lack of onboarding, then changing the people will not help. What is worse, we will lose the knowledge base within the organization. So, we can design a simple form where we log defects, escalations, stakeholders, jobs to investigate further and to see how much we lose from recurrent or impacting defects. Sometimes we even lose our customers.

Failure Mode and Effect Analysis (FMEA) and Risk Priority Number

The tool helps us anticipate what might go wrong. It can be used during process design and for an on-going process and output of FMEA is risk priority number. It is also used to prioritize the possible defects.

You should enter input and then failure mode and its effect. Failure mode is in what way step, change or feature can go wrong. For example, you decide to translate XML in your translation management system and one of the possible defects is translated XML can be non-utf8, which ends up with character corruption.

For each failure mode (wrong encoding) and its effect (character corruption), you assign Severity Score (1: the weakest-10: the highest), occurrence (how frequently it can happen – 1: least frequently and 10: highly frequent), and detectability (likelihood of detection 1: least likely and 10: highly possible).

And Severity X Occurrence X Detectability gives you Risk Priority Number. So, the higher the RPN is, the more critical is the failure mode and this way we can focus on precautions during process design or fix this for on-going processes.

Template is based on the template by GoLeanSixSigma.com

Above figure shows an example of FMEA for preparing a perfect cup of tea with tea bag.

Those are some of the tools that are utilized in LSS. However, these are not complete. There are variety of tools. LSS especially relies on statistics. So there are many, comprehensive, easily applicable statistical tools.

Steps of Lean Six Sigma Methodology

When we face a defect, we directly go to prescription. If we are lucky, it works. If not, it happens again, and we rework. Meanwhile we must deal with frustrated customers. LSS recommends following its DMAIC methodology. Path to resolution goes through 5 stages:

Define

First, you must define the problem, improvement activity, project goals, internal and external requirements. At the definition stage a project charter is filled. The project charter should include project members, period, scope and goals. At the definition stage, the voice of the customer (VoC) is prepared. VoC plays a key role and it triggers other tools and stages. As input, you should prepare Process Map. You find out critical measurable requirements of customer.

Measure

You cannot improve the things you cannot measure. That is why we find the measurable inputs and requirements. At this stage, you need to implement a data collection plan. For example, if you have not logged your cycle time or if you do not know your productivity per hour, you need to find a way of measuring these. If you do not record any of the escalations, you must implement a solution to log the escalations.

Analyze

In the previous stage, you collect the data. Now it is time to talk to the data. Or it is time for data to talk. Now you can calculate your current sigma level. To calculate your sigma level, you do not have to wait until you run your processes 1.000.000 times. There is a simple formula to calculate your sigma level. First you should calculate your “defect per million opportunities” (DPMO). DPMO = (Total Number of Defects found in sample / (Sample size X Number of Defect Opportunities) ) X 1.000.000.

Please note that the calculation of errors is a bit different than what is used in many organizations today. For example, you deliver 10 projects. One of the projects has one typo, one mistranslation, one formatting error. Most probably you will think that your success ratio is 90 percent. However, your defect per unit ratio is different. You deliver 10 projects and you have observed 3 issues. Your success ratio is 70 percent. The logic behind  is understandable and fair. It simply advises you to focus on three different issues, which may have different root causes. And if you resolve one of them, all of them will not be resolved.

While calculating your sigma level, we also found the number of defect opportunities. Defect opportunities are all the possible defects that can happen at one instance at once. For example, a project delivered can have a typo (1), mistranslation (2), formatting (3), terminology (4) and mismatching numbers (5) in one segment. In this case our defect opportunity is 5. At analysis, you may find mostly encountered issues. It requires statistical analysis. (Minitab is commonly used in LSS). You can also use FMEA.

Improve

At the improve stage, we need to make decisions about what to improve. However, we cannot do that instinctively. LSS offers many tools. Pareto is one of them. 80% of your income is driven by 20% of your customers. Or one aspect of your process is responsible for 80% of your issues. Another tool to find out where to improve is the cause and effect matrix. Yet another tool to select improvements is RPN - Risk Priority Number.

You can also implement error-proofing mechanism. You can set some preventive, proactive and retroactive stages. Real-time terminology, spelling checkers are proactive solutions. Protected tags inside a segment is a preventive solution and it does not allow translators to change the formatting codes. And quality validators we use are retroactive tools that detect the errors before it is delivered to the customer.

Control

After implementing some improvements, you apply a control plan and start measuring the effectiveness of your improvements. If you still experience the problem, go back to the previous stage, and apply another improvement plan. Sometimes it is good to apply pilot projects before investing time and money on an agreed improvement. To achive this, you have to implement a control plan, i.e. you should apply the measurement tools you apply in Measurement phase as well as additional tools defined at FMEA.

Now I want to keep my promise to share the map of the tools and stages.

Some of the tools you can utilize in your projects. However, this should not be seen as complete list of tools.

Criticism of the approach and counter arguments

When I first joined the training and was asked to adapt this work, my first reaction was that we were not manufacturing bottle cap, spare parts for cars or delivering packages. We do translation, which is subjective to some extent. However, the translation industry both on the vendor side and the buyer side tries best to set out objective standards to measure the quality of translation such as TAUS DQF, LISA LQA and MDQF. We give a scorecard to our reviewers, ask them to find issues in the translations, categorize the issues and voila! We have a score. That is a result of a need to measure the translation quality on an unbiased basis. That is exactly how manufacturing companies do as well. They randomly select some products and test them. Each translation company or department has translation workflows, processes and processes for quality and translation itself. That means we can also measure the success of our standards and processes.

My aim is not to give you a tool to find the scapegoat, but bottlenecks or processes creating scapegoats. I believe sometimes a low quality work is not a result of a linguist’s incapability, but process or source content. Most of the time even source content does not meet the standards set in style guides. Sometimes the process we follow is not suitable for the content. And considering that each translation program is special, we should set different workflows needed. Unfortunately, there is no one process fitting for all processes.

Another criticism is about 5 stages. Please note that LSS tools do not only fix the process, they are also used to design a process. When you design a process, the stages are Define, Measure, Analyze, Design and Verify. Design replaces improve and verify replaces control. And DMAIC looks long, but helps you make grounded decisions based on data and reality.

Difference between LQA and LSS

The localization and translation industry has several approaches to measure the quality. However, these approaches measure the quality of output and do not give a clue about where it goes wrong, they tell us what goes wrong. Additionally, the translation as a process today does not only include “translating”, but project management, engineering stages as well. It is an ecosystem including end customers and content owners. LQA can be a data collection tools to find the problematic inputs and outputs. Today volume of content is increasing, turn-around time is decreasing at the localization market. So, this requires paying some attention to the processes. Otherwise we will just see a quality score at the end of the project. LSS frame always helps users follow a path for excellence at operations management, which can be achieved via close monitoring of the processes.

Conclusion

This article cannot be evaluated full guideline of Lean Six Sigma. It aims to introduce some principles of the framework and some of the tools. It was originally applied for manufacturing and handling the quality of material products. However, today it is applied at various industries from IT to hospitality management. Service industries adopt the methodology to transform their business. I believe addressing issues at process level may help to resolve some translation issues in long run.