So why are project managers not “rationalized” by automation yet, and why are we seeing record numbers for new project manager hires among language service providers? asks Plunet‘s Sophie Halbeisen (pictured left).
Investing in automation does not necessarily mean replacing humans. Automation accelerates growth, and growth comes with a need for more qualified individuals. And that is a good thing! Project managers will become even more crucial in the upcoming years. For this reason, full automation should not be viewed as a one-size-fits-all-solution.
Unquestionably, automation is invaluable for an efficient translation project management, but for most types of projects we are not even close to the so called “lights out project management” — an automated project management workflow where human intervention is not necessary — and even further away from artificial intelligence taking over the jobs of our project managers.
In a survey conducted by CSA Research earlier this year, a total of 68% of language service providers had zero full automation capabilities with regard to their business and project management processes. In addition, of the 21% of language service providers that have some fully automated workflows, automation was only being used on less than 10% of their projects. So if you are not “fully” automated, you are not alone.
In this article, I want to take some time to reflect on the current state of our technology, common misconceptions in regard to automation and AI, and share my personal outlook for the future of project management.
Artificial Intelligence in the Language Industry
I want to start with Artificial Intelligence (AI), which is certainly a popular buzzword, not just in the language industry but everywhere in the world. AI means that a machine is programmed to not just follow a set of rules, but learn as humans do, essentially mimicking human intelligence to reach a goal without humans configuring each step of the way. Machine learning, a subset of AI, means that a system can improve its perception, knowledge, thinking, or actions based on experience or data, therefore getting smarter with each new data set that comes their way[1].
AI is an impressive invention and will shape the future of many industries, including ours. With regard to assistance with translations, we have seen incredible progress in quality due to machine learning over the years. Especially when the rule-based approach of MT was accompanied by machine learning, it led to better results.
However, when it comes to project management tools, we rely on rule-based automation and have not adapted AI as an alternative. There are some good reasons why. Before we get into them, let’s first get into the definitions and application possibilities of the two automation methods.
Rule-Based Automation: The Standard for Translation Project Management
Rule-based automation means that each action taken by a software is based on a rule that was configured by a human. Compared to AI, where the system is able to make a decision, each decision needs to be defined by the system admin. By tagging a project with a value such as “healthcare translation”, for instance, the system could pull a price list that is tagged with that same value and will only be shown to translators that have a background in healthcare. This also extends to more variables, such as language combinations, qualifications, availability, certifications, confidentiality levels etc. Apre-defined set of rules that reliably suggests the best matching vendor (or price list) to a project manager.
Many people call this form of automation AI because the system is seemingly using “intelligence”. However, it is not quite accurate, because a person configured these rules with their own (human) intelligence and the system is not learning or improving based on patterns of behavior. And this is what AI is about. The machine is able to predict steps because it learns from user behavior and sees patterns in data sets. Evidently, the prospect of AI seems exciting and more agile than rule-based automation, which relies on the instructions given to the system by a human.
So why do I think we are “stuck” with rule-based automation and our own intelligence for the next decade? I see the following three factors as the most significant:
Project Management Lacks Data
AI only works with massive amounts of data and the more complex a process is, the more data we need for a machine to make sensible choices. This has worked pretty well on the translation side, because the amount of data and translated content is practically endless.
Data quantity is also the reason why Google’s AI for predictive text works so well. Google’s email service Gmail has approx. 1.8 billion users, each with an average of 17,000 emails per account. This creates significant amounts of data sets that feed the Google machines. In comparison, a big language service provider might have 60–100 project managers who are managing a couple hundred of projects per month. Therefore, an LSP would need a lot of time to accumulate enough data to create reliable artificial intelligence that covers all exceptions, all variables, and individual preferences. On top of that, requirements change all the time, so once a machine has learned a certain pattern, that pattern might not even be relevant in a fast-paced environment like translation and localization project management.
High Risk for Data Bias in Project Management
In addition to the need for a sheer quantity of data, we also need high-quality data to make machine learning worthwhile, and — as always in life — getting both at the same time is very hard.
A real challenge of AI is that biased data can train machines in ways that are not intended. We have all seen the example of MT results being accidently racist, sexist and non-inclusive based on the biased data that is used to train these engines. And though the damage is mostly unintentional, it still has an impact on project quality (and society as a whole).
To use Google as an example again: Due to the lack of data quality, even image search results can be misleading despite the massive quantity of data. For the keyword “author”, only 25% of Google images are female, even though 56% of authors are actually female. If users don’t know that machines have been trained with biased data, they could presume that 75% of authors are male and, for example, base hiring choices on that perception.
The same could happen with AI in project management. With even smaller qualitative data sets, there is an even higher risk of data bias. If, for instance, 80% of your projects are run a certain way, a machine would pick up and follow these patterns, although you might make most of your profits with the other 20% of projects that require a totally different workflow. As a result, AI could learn patterns that are not helpful for your most important work and could even be harmful for your business.
The Role of Emotions and Exceptions in Project Management
Lastly, but also of significance: emotions and exceptions are the biggest enemy of automation and AI in particular. In project management, the exceptions are endless and the right way of doing things can sometimes change on the spot or be dependent on the personality in charge. Project managers have to be empathetic and great communicators, and they have to understand when to step in and avert a crisis while juggling projects that can be mission-critical. Sometimes the best solution is not the most efficient one, and in many instances, the personal connection is the reason why customers stay with a company. Studies on AI chatbots show that purchase rates go down by 79.7% when the consumer knows that the conversation partner is not a human. This shows that in businesses where building trust is important, time-saving automation could potentially harm a business when used at the wrong touch point with the customer.
Outlook for Project Management in the Language Industry
So what’s the take-away? For now, it looks like rule-based automation will be the technology behind most project management automation. In addition, it’s important to keep in mind that automation technology is evolving fast, so everyone in the translation and localization industry should incorporate it into their growth strategy and stay up-to-date with the topic.
When looking at processes for automation today, I would start with the ones that have the fewest exceptions and require the least empathy. Data handover from one system to another (e.g. from your TMS to your accounting tool) is a perfect example for a task that should be automated, as there is little to no human value to add but a very high risk of human error. A machine is better equipped for that. On the other hand, responding to customer complaints or giving a vendor feedback would be on the opposite end of the spectrum, as it requires a lot of empathy and — as everyone who ever worked in customer service can attest — can be full of surprises and exceptions!
Consequently, I would automate everything where human intervention cannot add any value. But the definition of that value can be completely different from one company to the next. Some companies in the language industry benefit greatly from time-savings due to automated vendor requests and assignments. Other companies rely on the relationships built with their freelancers in a more personal assignment approach. The same goes for quoting, where some businesses value the automation options technology offers, while others create more value by looking at each quote individually — it might even be the secret to their success.
In conclusion, automation shouldn’t replace human resources, it should allow a company to scale and grow their business. Every company in the language industry should focus on project management and workflow automation, but only if it makes sense for their specific customers and the type of services they offer. If a personal touch is your USP, it’s the last thing you should automate away.
