Presentation given at roundtable discussion "(Re)mixing Workflows: Localization Processes Forum" on October 29, 2014 at the Localization World Conference in Vancouver, Canada.
2. +
Overview
Project Management in an ‘Agile’ world:
The ‘old’ days and the new reality
Content development and deployment is changing
The challenge for LSPs
Automation of processes is key
Technology example: expressIt
2
Localization World Vancouver 2014 29 October 2014
3. +
Out with the Old -- Waterfall
In with the New -- Agile
Software development and localization
Major and minor version releases
Help files in shrink-wrapped software boxes
CD-ROMs duplicated and distributed
Brochures/Manuals
Recalls for errors very costly
Software development: origin of Agile
Key driver is digital distribution: web pages, PDFs, eLearning, etc.
Instant distribution
Updates are easy and inexpensive
“The era of multi-month mega-projects is over”
3
Localization World Vancouver 2014 29 October 2014
4. +
Content development/deployment is
changing
App and website content is never ‘final’
“It doesn’t have to be perfect, it just has to be done.”
- get it out there, then improve it
Smaller, faster initial deployments
Rapid and continuous updates and corrections
For some clients (but not all), ‘perfect translation quality’ isn’t
critical because errors can be fixed quickly.
4
Localization World Vancouver 2014 29 October 2014
5. +
The challenge for LSPs
From: Traditional project management
Gantt charted, weekly team meetings, major milestones
To: “Air traffic control” project management
One PM handles 10-20+ small projects simultaneously
You can’t wait until the “next weekly meeting” to fix/discuss it
Issues:
Fewer words in each request PM expense out of proportion
No “minimum fee” how to cover/reduce costs?
Small jobs eaten as part of larger engagement
What happens when small requests become a significant
feature of client relationships?
5
Localization World Vancouver 2014 29 October 2014
6. +
Conundrum
Translation takes too long: Use MT
Is time to translate truly the issue for small projects?
The smaller the update, the less speed benefit from MT
Where are we actually losing time?
Automated
Standard
6
Localization World Vancouver 2014 29 October 2014
7. +
Automation of processes is key
At Elanex, our primary response was complete shift in
perspective on project management
Project management tasks in localization are highly repetitive
By automating PM activities at
the task level, we allow our
human PMs to focus
limited time and attention
on problem-solving
7
Receive files
from translators
Keep project on
Schedule
Project
Manager
(typically
available only 8
of 24 hours)
Send files to
translators
Prepare files for
transmittal
Identify
resources, costs
Develop project
schedule
Perform
analysis (word
count, etc.)
Receive files
from customer
Send files for
proofing
Receive files
from editors
Build final TM
Assemble files
for customer
Send files to
customer
Close project,
billing, etc.
Localization World Vancouver 2014 29 October 2014
8. +
expressIt: Fully Automated PM
Successful for Elanex
Project timelines reduced as much as an order of magnitude
One ‘exception-handler’ PM can monitor 50-100 projects
simultaneously
Project managers don’t miss drudgery of admin work normally
generated by large numbers of small projects
Successful for Clients
Able to place orders instantly (24/7), without waiting for their usual
account manager (or any human) to be available
Receive standard high-quality work product from human translators,
much faster than ‘normal’
Clients love the lower pricing for small requests
No ‘rush fees’
8
Localization World Vancouver 2014 29 October 2014
Traditional software development, and by extension localization process for that software was based around ‘major’ and ‘minor’ releases of the product to end users. Localization project managers worked to localize content like Help files, manuals and product boxes that were shipped as shrink-wrapped, physical product. Over time, much of the paper shifted to digital media like CD-ROMs and DVD-ROMs, but even those had to be duplicated and shipped from the factory to retail locations or direct to end users. Prevention of errors was critical because a recall or update could create huge unexpected costs.
Agile development process arose from software development and coding but have evolved into an approach for content development as well. As the medium for written words has shifted from paper to the computer screen, the power of the internet has also fundamentally changed the economics of content distribution. When your brochure or your entire manual is online as web pages, both the initial distribution and updates can be deployed instantly, and at virtually no incremental cost.
The shift in the economics of distribution is changing the way that content is developed, and the Agile approach to software development fits very will in the new model of content creation. No longer do we need a very small number of major/minor releases. In fact, content never needs to be “finalized’ at all. Content developers (and app developers, since a lot of content is now packaged or routed through apps) are focused on getting a ‘minimum viable product’ out there as quickly as possible, and then launching into cycles of continuous improvement.
What this means for us as Localization Service Providers is that the initial ‘big’ project is no longer as big as it once was…AND we have to be able to support an ongoing cycle of small, rapid updates and corrections. In many cases, it’s not just a matter of volume but this affects our clients’ attitude about quality, as well. At least for some clients, quality is less important at least at first, because it’s so easy and inexpensive to fix errors after they are identified.
For the project manager, as well as LSPs in general, this creates a new challenge that is fundamental to operating the business. We tend to have a large number of project managers who are well versed in the traditional tools of project management: Gannt charts, weekly meetings, major milestones, presentations to management, etc. But rapid cycles of small, incremental changes are a completely different beast for the project manager. The timelines from start to finish for a single project may only last days, or even hours. If a problem comes up, like a translator is suddenly unavailable because they have fallen sick, or the source text has some problem that the translator can’t resolve, the project manger can’t wait until the next weekly meeting to discuss and resolve it. They have hours or sometimes even minutes to get things back on track if they want to have any hope of keeping their project on schedule and on track. A single project manager is probably also juggling 10-20 of these types of small projects at the same time.
We often refer to this as “Air Traffic Control” project management, and we’ve found that the PMs who are really good at this, and genuinely enjoy the high-intensity of this type of work, are often not even the same people who are good at the traditional, long-time horizon mega-projects.
For the LSP as a business, one of the big challenges is that larger numbers of requests with fewer words typically knock PM expenses out of proportion with the revenue of those projects. We know that clients really don’t like to hear the words “minimum fee”, but those costs have to be covered. Historically, LSPs have generally gritted their teeth and absorbed the costs like “23 words into 42 languages” because they were part of a much larger project that would still be profitable anyway.
But what happens when these small requests become the business your client needs you to perform for them on a regular basis?
How do we increase speed and decrease costs? Well, this industry has naturally and mostly looked at machine translation as an obvious solution. Quality of output is increasing and it’s finally becoming feasible in many cases. But the big conundrum is that the smaller the project is, the less MT actually helps because there is less translation to do. It turns out that the problem isn’t really the speed at which the translator works!
A number of years ago we did a ‘time and motion’ study to look at where the elapsed time in a typical translation project really goes. We know that translation cost is typically the majority of the overall cost of a project, and it seems like it consumes the majority of time in the project, as well. It turns out that that’s not actually the case: The vast majority of the time in most projects is downtime, where nothing is actually happening at all. Downtime is mostly the time that the next project is sitting in queue waiting for the PM to finish work on the other 3 or 4 or 10 projects they are working on, the time they spend in meetings, the time they spend at lunch or after hours when they aren’t working at all. The email or task ticket or whatever it is, sits there and waits for them.
If you can automate the tasks that the PM does, it turns out that total project duration for the same project can easily fall to 1/10 of the total time for an un-automated project. And that is assuming no machine translation at all—as you can see in the bars, the total time for both translation and editing by human professional translators is exactly the same.
When we realized this, our primary response was a complete shift in our perspective on project management. Project mangers are smart, highly-qualified people. But it turns out that a huge amount of the work they do is highly repetitive and a lot of it is simply clerical, like keeping records, organizing and storing files, etc. These activities are the sort of thing that should be automated. With the application of some artificial intelligence and appropriate management of the data that is available, even things like translator team selection can be automated without sacrificing quality. Even a traditional PM’s Interactions with linguists tend to be made efficient through the use of templates. By using those same templates and simply automating the press of the ‘Send’ button, we can reduce the downtime around team formation, as well.
By automating each standard PM task, we’re not replacing PMs but instead are making them very, very efficient. We don’t want to have them bogged down with repetitive, simple, time-consuming tasks. What we really need their time and attention for are the complicated things, the unexpected problems that come up doing the course of any project. With appropriate automation, PMs become primarily exception-handlers and problem solvers, handling the problems that computers and algorithms can’t.
At Elanex, we call our fully automated workflow engine “expressIt”. It has been up and running for several years, and so far this year has handled over 5,000 individual translation jobs. From our point of view it’s great because it makes our PMs so much more effective.
As most PMs know, you can’t imagine how many ways a localization project can go wrong. Literally, you can’t imagine them all…so it’s not possible to create algorithms to handle every contingency. The expressIt system is monitored by PMs, and the system notifies those PMs when something unexpected happens. The PM steps in, diagnoses and resolves the problem, and then returns the workflow back to the expressIt engine to finish and ultimately deliver the final work product to the client directly. So to be clear, we are not replacing PMs entirely—we’re simply changing their role in the workflow process to focus on the ‘hard stuff’, while allowing the computer to handle the ‘easy stuff’, which it can do much more quickly and without getting bored. And instead of juggling 10-20 projects at the same time in ‘Air Traffic Control’ mode, our PMs are able to monitor 50-100 (or even more) projects that are simultaneously managed by the expressIt system.
Our clients are very happy with expressIt because it means that they can place orders on our system 24/7, without waiting for a human being to get back to them. Their projects are kicked off and assigned to a translation team right away. They also like it because even though we rely heavily on automation, clients are receiving the full quality of human-translated, human-edited work product. They don’t have to deal with either the poor quality of untrained MT or the high cost of training an MT system for a small volume of work. And we can avoid having to charge minimum fees or small project fees for most small requests because we have eliminated the primary source of the cost that makes those fees necessary.