Cost scaling through technology management

Over the last four years, we at Dell have cut translation cost per word in half, allowing us to support increased globalization and twice the volume at the same cost. In and of itself, that might not be overly impressive. We all know you can tweak one of the trifecta (cost, time, quality) at the expense of one or both of the others. However, if I told you that we accomplished the cost and volume scaling while keeping quality high and cutting turnaround time in half, you might find that interesting.

No discussion of cost reduction can be had without a discussion of quality. Before making any process changes or starting a cost discussion with suppliers, you need to have a solid quality baseline. If you do not know where your quality is now, you will not know if your actions have negatively impacted quality. This is crucial because for some content and businesses, quality is much more important than cost. For other content, you may already be paying too much for quality, but again, you will not know if you have reached the desired outcome if you do not know where you started. We had a robust quality process in place prior to starting this multiyear journey. As it turned out, our process was too robust, and trimming it gained us a large portion of our scaling. Our quality today is as high as it was four years ago, confirmed by our quality metrics, our stakeholders and our end customers.

In 2009, our cost structure was 25% quality, 10% project management fees, 5% desktop publishing. The other 60% involved the actual translation cost. We were getting about 20% translation memory (TM) leverage on what was primarily marketing and online commerce content.

To spend 25% of our budget on quality was too high, but not out of the range of many others who were producing





















marketing, branding and commerce content. After all, we had reached our goals in terms of quality (Figure 1). For most content streams we performed a full third-party review. We had previously eliminated internal review, shifting those resources back to core competency. Such a robust outside review seemed like the right thing to do at the time. We had given more than 100 people back some of their time and we were delivering good quality, but a statement from an industry veteran kept coming back to me: “paying another company to do what the translators are supposed to do is stupid.” I think there were some expletives in the original statement, but the message was clear. Hence, we started down a path of reduced review for high-performing languages, and while it took a lot longer than I had hoped, last year we finally stopped all third-party review, moving to a translator self-certification model, taking samples per language and per vendor each week to ensure that the quality stayed above our goal. This resulted in a $1.5 million annual savings and reduced the cost of quality from 25% to less than 10% of the overall spend. It also had a significant impact on our delivery time.

Turnaround time reductions were a function of the process changes we implemented in our quality program combined with moving to a continuous translation model versus the traditional project model. Removing quality review saved two days alone, and the remaining time savings resulted from the model change.


Tools and translation cost

There are many pieces to translation cost, with vendor rates, computer-aided translation tools, TM leverage and machine translation (MT) being the primary ones. Each had to be addressed. We gained some initial cost benefit by reducing our vendor pool from five to two. It’s a prominent trend in large enterprises across the outside services spend ecosystem. On the client side, we gain lower vendor management overhead, volume discounts and the ability to invest more time in partner relations — and honestly, this one doesn’t get the attention it deserves. The vendor gains increased revenue. It’s a win-win for those involved, but does have implications for the broader market. Once we had resized our vendor pool, we turned to rate analysis. We reviewed industry reports from Common Sense Advisory, spoke with other large enterprise clients — in aggregate, of course; no specific rates were discussed — and had an opportunity to hire in someone from the vendor side with the added benefit of strong vendor-side pricing knowledge. We negotiated roughly a 10% reduction in cost, while still leaving our partners with healthy enough margins to continue the great service they were providing.

Rate discussions are a fine line to walk. You want to get the best possible rates and be fiscally responsible, doing the right thing for your company, but cutting cost too aggressively and putting your partners in a position of having to lower their service level does neither of you any good in the long run. There are, of course, scenarios where lower cost and lower service levels are appropriate, but that was not the case here. Over this same four-year period of time, we have seen TM leverage go from 20% to over 40%, a healthy number given the rapidly changing pace of technology and technology marketing, especially in the consumer market. We did not do anything magical to increase leverage; we invested in maintaining the quality of our TM assets by having our quality assurance vendor go through the TMs to weed out inconsistencies, duplications and unused segments. We also continued to expand our internal stakeholder list and populate our TM. Altogether this was probably the easiest thing we did, and worth an estimated $2 million annually at today’s volumes. We renegotiated project management fees while negotiating translation cost, reducing them from 10% to 5% on average. This was primarily a result of the realization that with our increased volume and spend, 5% was an appropriate rate. We have not made progress on desktop publishing cost; it is on the roadmap to tackle in the 2014 fiscal year.

We started down the path of MT for a year and a half ago. We began by talking to everyone who would talk to us. We got lots of advice, mostly people telling us we could not do what we wanted to do, which was to use MT for a commerce site. We weren’t targeting support content since we had done that already, and instead we were after the browse and purchase path. The team and I have never been ones to follow conventional wisdom. We started quietly, with just two languages, French and German, and we did not tell anyone (meaning the stakeholders we were delivering content to) what we were doing until we were about ten languages in. We took the standpoint that they expect a certain product specification from us, and they should not care how we deliver, just that we do. Today all 27 non-English languages we translate into for our product content is leveraging post-edited MT. Our savings have not been huge because we still have very high quality standards, but we are saving about 10% and we are improving those savings quarter over quarter as productivity and the MT engines improve. What supposedly could not be done was not too difficult to do, with a little rethinking of the model and a willingness to take the risk.

Through an iterative process of identifying drivers, current state metrics and desired outcomes for each member of the translation trifecta (cost, time and quality), combined with assigning out initiatives to key members of the team who were uniquely equipped to tackle them, we have been able to reduce average turnaround time from seven days to three, at least for some workflows, and reduce cost per word by nearly 50%. All while scaling volume by nearly two times (Figure 2), and maintaining the high level of quality our stakeholders and end customers expect. We could not have accomplished this without a couple of guiding lights, translation quality and vendor partnership, along with a team dedicated to constant improvement and innovation.

Although we have cut cost per word substantially, we have not “saved” money. Saving money implies that you spend less. We have spent roughly the same amount each year. However, we are delivering twice the content for that same cost. We have managed to create cost scaling.