Real-time speech translation can eliminate the dependency on (near) native speakers as customer contact employees. This is particularly interesting for small languages ​​such as Dutch or Scandinavian languages, where employees are scarce and therefore relatively expensive. The technology has the potential to become the next game changer in customer contact.

 

Jakob Westgren is head of IT product and partners at facility contact center Transcom and is based in Stockholm. He is responsible for the delivery of (new) IT applications. Westgren wants to make it clear that real-time translation of speech and text are two fundamentally different things. Text channels work more asynchronously, where latency is not an obstacle. It is precisely this latency that makes real-time speech translation more complex. More about that later.

'High cost languages' in 'low cost countries'

“Real-time speech translation is an important and strategic part of Transcom’s IT portfolio. Transcom’s challenge is to find the right talent at the right time and at the right price. Real-time translation removes the limitations of language regions. It makes our talent pool much larger: you can work with ‘high cost languages’ in ‘low cost countries’.”

The latency issue

There is considerable scepticism in the market about the real impact of real-time speech translation. There are doubts about the impact of latency – the time it takes for the customer to receive a spoken response. And also, the quality of the translation is said to be too dependent on cultural differences.

Westgren is realistic: “If you ask a developer if this technology works, they will probably answer with a resounding 'yes'. If you ask the same question to a consumer, you might get a different answer.”

The complete story

The latency problem is getting smaller and is also no longer primarily caused by the technology. According to Westgren, the technology is fast enough. Converting what the customer says into text, translating that text and presenting that translated text to the 'real-time translation agent' (RTT agent) that is reading along: all that can be done in 0,12 seconds, says Westgren. Moreover, with advanced translation software, the translation is continuously adjusted while the customer continues to speak. The main cause of latency at the moment is still that for a complete translation (and therefore a complete story for the agent) the customer must have finished speaking. So it takes some time before the RTT employee can start his or her spoken response.

Translate frontrunners in real time – Transcom has a multi-vendor strategy, including in real-time translation. The company uses technology from two leaders in real-time translation: Cognigy and Parloa. In addition, the company uses multiple partners that specialize in conversational AI. Some focus solely on real-time translation, while others focus on a broader agent experience toolset. “We are constantly exploring the best technology for our customers based on their specific needs and requirements.”

Camouflaging latency

“It’s clear that the customer experience with real-time speech translation is not as good as a conversation with a native speaker. But we have found ways to make it almost as good and natural. Cultural differences can be bridged through training. And latency is addressed by adding artificial sound from keyboard strokes or background noise to the dialogue. This gives the customer the impression that the person on the other end of the line is busy with the customer’s question. Generating short interjections that agents can select also contributes to this. For example, callers hear: ‘Just a moment, I’m checking something for you’ or ‘just a moment’. You also hear these kinds of interjections when you talk to a native agent; it makes the conversation very natural and distracts from the latency.”

voicebots

Another way to deal with latency is to use a voice bot. The live agent then becomes the escalation channel, which also works on the basis of real-time speech translation. “If a handover to a live agent is required, it helps if you manage the customer’s expectations in advance,” Westgren explains. “For example, by clearly indicating that real-time speech translation is being used and that there will be a small delay between question and answer.”

And finally, the latency problem can be reduced by supporting the RTT agent with 'quick replies' that can be spoken automatically. The tool is connected to the knowledge base and provides answer suggestions to agents. Thanks to this feature, agents do not have to spend time searching for answers, which significantly improves the handling time, according to Westgren.

For the time being, the use case is decisive for success

As with AI in general, real-time speech translation also requires finding the right use case. Westgren gives an example: “Logistics service providers often work with codes that consist of long strings of numeric and alphanumeric characters. Working with such codes does not work well with real-time speech translation, as we have learned: if something goes wrong, the string has to be repeated, which costs too much time and is annoying.”

Another key to success is specific agent skills. Skills such as multitasking, being able to read quickly and work quickly with answer suggestions and knowledge snippets, as well as a good understanding of cultural nuances, are more important to RTT agents than excellent language skills.

Impact of RTT on Transcom's offshore strategy

The promise of RTT – to place ‘expensive languages’ in low-wage countries – has not yet led to any shifts in Transcom’s business, Westgren confirms. But RTT is already having an impact on Transcom’s offshoring strategy and will certainly have an impact on future investments in offshore locations, Westgren assures.

“Dutch is an expensive language from an offshore perspective. The same goes for Scandinavian languages. Normally you would look for offshore destinations for these languages, but in the near future RTT will become the leading language. Both Dutch and languages ​​from the 'Nordics' are quite mature when it comes to the use of AI.”

At Transcom, the first changes will be visible mainly at the level of individual clients in the coming period. Even though a small part will go to an alternative location via RTT in the coming period, the majority will remain with native speakers or near-native speakers for the time being, Westgren expects. He therefore expects that RTT will lead to an expansion of activities in the coming period rather than to the relocation of existing activities.

“But in a few years, if RTT is operating seamlessly, you could serve Swedish customers on a large scale from Africa. That will definitely have a significant impact on our onshore and offshore locations,” Westgren says.

Real-time Translation at Transcom in practice – “Our English speaking agents in Cairo support customers from TUI in the Netherlands and Sweden for a year now with real-time speech translation. TUI uses this tool to handle peaks in traffic, for pre-trip customer queries that can lead to additional revenue: extra baggage, seat changes, upgrades. Customers are given the option to be connected to the team in Cairo using RTT, instead of waiting in line for more than 4 minutes. In this way, TUI does not lose up- and cross-selling opportunities due to customers hanging up prematurely and customers receive support faster.” This results in major improvements in average call duration and customer satisfaction, according to Transcom.
A second example is supporting customers who use Bird shared scooters. “Our English-speaking agents in India have been supporting Bird’s customers in the Netherlands and France with real-time translation since early summer this year. Keeping churn low, retaining customers and building additional revenue is key for Bird. Bird uses RTT for all calls. These are questions related to Bird’s services, from offering questions, technical support and app support to marketing campaigns. Agents are presented with the RTT option as a choice. RTT enables Bird to grow in all markets with consistent costs and quality of service.”

(Ziptone/Erik Bouwer)

Also read: Real-time speech translation, the deep dive. What to consider during implementation.

Case, Featured, Technology Cognigy, Transcom