Lost in Translation: Unveiling Optimization Secrets Through Analytics

Analytics is like the superhero of the business world, swooping in to save the day by spotting all the places where time and money are going down the drain. Analytics helps companies streamline processes, trim costs, and become lean, mean, efficiency machines!

At Phrase, we take analytics seriously—it’s not just a sidekick, it’s the star of the show! That’s why we’re committed to providing robust analytics capabilities in Phrase TMS. Picture this: 7 dashboards with over 50 pre-defined charts ready to go, the ability to create your own dashboards and charts, tailored to your unique needs and tastes. And as if that weren’t enough, we’re dishing out access to the raw data with Product Data, giving you a backstage pass to the inner workings of your localization efforts. 

This spring, we enhanced our analytics capabilities by integrating data from Shared Jobs and Projects. This addition provides users with a comprehensive view of the collaborative efforts between buyers and vendors.

As we head into summer, we’re preparing to unveil a significant analytics improvement for another product in our portfolio. Keep an eye out for updates!

And that’s not all—we’re gearing up to take analytics to a whole new level. Get ready for a significant leap forward in insights and capabilities as we introduce our latest advancements in analytics. Exciting things are on the horizon!

Let’s look at an example of what we’ll soon be able to accomplish and explore one of the key parameters in the pre-translation settings.

If you’ve ever found yourself pondering what the optimal TM Threshold should be in your pre-translation settings, you’re certainly not alone. In fact, there are several signals that indicate many localization teams struggle with this very decision. Choosing the right value can be challenging, as it directly impacts translation leverage and overall quality. 

The chart below unveils a striking revelation: the most frequently employed TM Threshold in Phrase TMS is set at a modest 70. Curious as to why? Well, it turns out that this value happens to be the default setting. 

Interestingly, the only other commonly used value is a solid 100, which aligns more closely with what one might consider an optimal threshold. All other values seem to follow a pattern of being multiples of 5. Now, call me skeptical, but this doesn’t quite scream “evidence-based decision-making” to me, wouldn’t you agree?

Furthermore, upon closer examination of the TM threshold over the past four years, one might be surprised to find that it has remained virtually stagnant. Despite the significant advancements in Machine Translation technology, the “hand-off” from TM to MT seems to be stuck at the average value of 78. Well, to be fair, it has actually increased by a mere 1 point over the span of four years.

Well, it seems broken, but without analytics or data insights, who can really tell?

Let’s take a look at what we can observe today at Phrase and give you a sneak peek of what you’ll be able to see before summer ends.

The chart below illustrates the average editing time of a Segment relative to its TM Fuzzy score.

It’s worth noting that the choppiness observed toward the lower scores is due to the smaller number of averaged values. Conversely, the stability on the right side of the chart is consistent, as these scores appear very frequently in our data set (this data represents a one-month sampling, where every 100th segment is recorded).

It’s fascinating how the chart could be easily interpolated by two linear lines: one flat line positioned below scores of 94 (orange), and the other descending from the score 94 (green).

In simpler terms, it seems that TM fuzzy matches with scores below 94 don’t really boost translation productivity. What’s particularly striking is the disconnect between what this data reveals and the actual TM Threshold settings: an average of 78, with the most common value of 70.

On the flip side, a mere 1% enhancement in the quality of translations produced by the pre-translation function slashes segment editing time by a staggering 3 seconds. When you factor in the massive volume of segments processed, these seemingly small gains quickly snowball into monumental savings. In the grand scheme, extrapolating across all our customers, a single-point improvement would rescue an astonishing 110,000 hours over the past 12 months alone.

How much could your organization save? Get ready for your journey to uncover the hidden treasures of cost-saving secrets! Discover the optimal TM Threshold value for your projects, unravel the intricate correlation between editing time and fuzzy scores for your language pairs, and more. The answers to these pivotal questions are just around the corner.

Stay tuned for more. In our next post, we’ll keep exploring. 

As a teaser, what do you think the red data points in the chart represent?

We’re seeking visionary thinkers to share their thoughts and expertise on the optimization of translation workflows.