Prominent Belgian speech tech users coming out at seminar

Last Thursday 60 odd people showed up at the NCI Business Center near Brussels Airport for a half-day seminar about “Speech Technology: Customer Experiences in Belgium”. The event was organised by’s Speech Technology Workgroup, currently chaired by your servant. STW seminar 'Speech Technology: Customer Experiences in Belgium', Nov 9, 2006, Diegem, Belgium

Among the - almost exclusively Belgian – audience were representatives from the following sectors: banking and insurance, (mobile) telecom, hospitals, public transport, utilities, government agencies, and academia. The event was sponsored by all major speech technology players active on the Belgian telecom & contact centre market: Acapela Group, DB Scape, Dimension Data, Genesys, Nextira One, Nuance, Quentris and Ubicall. STW seminar 'Speech Technology: Customer Experiences in Belgium', Nov 9, 2006, Diegem, Belgium

In a joint introductory address, Vincent Vanden Bossche (General Manager and Alain Rondenbosch (General Manager DB Scape) sketched the growing importance of speech vs. DTMF (touch tone) technology in the European self-service market. According to Mr. Rondenbosch, Belgian customer service departments have been slower to embrace speech technology compared to their British, German or French counterparts for two reasons: call volume (author’s note: Belgium only counts 10 million inhabitants), and the added complexity of multilingualism (author’s note: Dutch and French are Belgium’s major official languages – apart from German, which is spoken by 60 thousand people in the East. On top of that, English is widely spoken by the expat community in and around Brussels, the political capital of Europe).  However, DB Scape’s CEO also stressed that despite these additional challenges, Belgian customer service departments have exactly the same needs as their foreign colleagues: provide a better 24/7 service, at an acceptable cost level.

The major goal of the seminar was to provide a platform for Belgian end user organisations who have already taken up the speech-enabled self-service challenge. Five of them - coming from diverse sectors - were willing to share their experience, each focusing in on one or more of the following technologies: VoiceXML, speech synthesis (aka text-to-speech or TTS), speech recognition (aka ASR), and, last but not least, speaker verification (or voice authentication).

Cédric Bourgeois (Banksys) speaking at speech technology seminar on Nov 9, 2006

Cédric Bourgeois, Head of CTI Section at Banksys set the stage for the subsequent speakers with his story of how Banksys recently replaced its expensive legacy IVR with a future-proof VoiceXML platform. Banksys’ 120 contact centre agents handle 1,7 million calls annually, divided over seventeen applications, three of which are (partly) automated.

Banksys’ business requirements for moving to VoiceXML were threefold: shorten the IVR application implementation time, prepare for future market evolution, and reduce cost. On the technical side, they wanted to have a tighter control on the change process, reduce risk, use state of the art technology and revamp all of their legacy IVR applications.

To achieve these goals, Banksys opted for Genesys Voice Platform and bought speech licenses from Nuance. Banksys’ application developers used DB Scape’s Vocabase tool to migrate all legacy IVR applications to the VoiceXML platform. According to Mr. Bourgeois, Vocabase cuts development time in half compared to Genesys Studio, which comes bundled with GVP. The migration project was led by PriceWaterhouseCoopers, with NextiraOne acting as system integrator.

With 60% of the migration project’s - undisclosed – investment resources going to capital expenditures vs. only 40% to operational expenditures, Cédric Bourgeois ascribed the relatively high CAPEX vs. OPEX ratio to Banksys’ strong system redundancy requirements.

Now that Banksys has successfully migrated all legacy IVR applications to the new VoiceXML platform within time and budget, it is in a perfect position to take on the next challenges: integrate more self-service automation into their contact centre processes, implement speech recognition, and open up their systems to the business users so that they can manage their own voice portal content.

Some critique was voiced that Banksys’ migration project did not include a sufficient amount of speech technology experience to warrant their presence as a speaker in this seminar. If the organising Speech Technology Workgroup did decide to invite Banksys to present their case, it is because it is typical of the way speech technology enters many contact centres in Belgium: as a potential feature on a VoiceXML platform that is acquired in the first place to replace a dead-end legacy platform.

The important question in such an investment is whether the owners of the new platform are well aware of the business potential of its speech technology features. Banksys clearly has this vision, so their presence here was more than justified. We can’t wait to see more results from them over the coming six to twelve months.

Alexandre Gaschard (EDA) speaking at speech technology seminar on Nov 9, 2006

The second speaker was Alexandre Gaschard, CEO of European Directory Assistance. EDA operates the numbers 1212 and 1313 in Belgium, in direct competition with Belgacom’s 1207 and 1307. The company is also active on the Dutch and Irish DA market. The total Belgian DA market counts 4,3 million calls per year, that’s 4 per inhabitant, for a total value of 45 million euro.

EDA uses text-to-speech technology from Acapela Group to speak out the phone number to the caller, as soon as the human operator has looked it up. The use of TTS not only represents a time savings of 15% of the total call (including the front-end IVR part), but also allows for easier scalability when handling international DA calls, compared to the traditional method of audio file concatenation. Thanks to TTS, which Mr. Gaschard described as a “mature technology”, EDA not only reads aloud the full address, but is also able to offer extra services like cinema listings and restaurant menus. Reverse number lookup queries are even handled in a fully automatic way thanks to a pragmatic combination of DTMF-based IVR and TTS.

For EDA, the ability to quickly deploy innovative services to new markets is just as important an argument to use TTS as is cost reduction.

The efficient and effective use of TTS on the multilingual Belgian DA market does not happen just like that, however. EDA and Acapela Group have invested quite some effort in the creation of specific dictionaries containing the right pronunciation of non standard words like family, company, street and place names. To make his point, Mr. Gaschard demonstrated a simple Belgian French sentence with and without TTS tuning. The result was just as funny as convincing.

Julian Perez Sanchez (Brugmann Hospital) speaking at speech technology seminar on Nov 9, 2006

Next speaker in line was Julian Perez Sanchez, Head of Reception and Telephony at Brugmann University Hospital in Jette. He was assisted by Willy Van de Voorde, Sales Director at Ubicall, who started off with a general presentation of Voxplorer, Ubicall’s flagship product. One thing that’s special about Voxplorer is the broad coverage of the phonetic lexicon for person names, containing on average four transcriptions (= pronunciations) per name.

Mr. Perez Sanchez then provided some statistics on Brugmann’s general operator service before the installation of a speech recognition solution: about 300.000 calls per year, no less than 30% of them (that’s 100.000 a year!) abandoned, “very high” waiting times, lots of patient complaints about the phone service, and so lots of lost patients as well. No need to tell that some action was in order. By installing Voxplorer Receptionist as a front-end to the human operator service, Brugmann University Hospital was able to treat 40% of these incoming calls in a fully automatic way; the volume of calls to be handled by the operators was cut in half. Fewer operators were able to provide a better service; what’s more, stress levels went down, and team utilisation went up.

In the appointment division, the initial situation was even worse, with an abandonment level of 50% (yes, one out of two). The available lines were almost permanently busy, and if patients finally got through, the probability of getting answered was low, as staff had to deal with a growing number of patients flocking to the hospital in person to make their appointment. To solve this waiting line nightmare, the hospital installed Voxplorer Queue Manager. The automated system now leads callers to the right division, and informs them how long they need to wait.

But there was a third, yet more serious problem: when patients finally got their appointment with a doctor, they didn’t show up. This happened about 250.000 times a year. One of the reasons was that there was no way to cancel an appointment, except via the – already overloaded – appointment lines. The problem is that hospitals pay doctors to treat patients, not to sit idle, waiting for them to (not) turn up. As overbookings became more common, waiting times and frustration often also went on the rise. And as a side-effect, to make matters really dramatic, the emergency service became totally unmanageable. Ubicall’s Voxplorer PalMinder now reminds patients of their appointments via phone, SMS or mail; conversely, patients can also answer to this reminder should they want to cancel their appointment. The system can then propose another date, because it is integrated with the doctors’ appointment schedule database. Thanks to the new system, doctors’ agendas are managed in a much better way, and the waiting time to get an appointment has been considerably reduced.

Julian Perez Sanchez concluded that the installation of a complex speech-enabled phone system has not only improved phone access. More importantly, it has had a real beneficial impact on patient care as a whole. Even though patients may need some time to adapt t an automated reception based on speech recognition and synthesis, the speaker was confident that they would finally get used to it.

Yvon-Marie Willems (BASF) speaking at speech technology seminar on Nov 9, 2006

After the break, Yvon-Marie Willems, Facility Manager & Local Information Manager at BASF in Brussels, presented his dealings with EVA, Quentris‘ Electronic Voice Assistant. Peter Edel, Advanced Technology Manager at Quentris, assisted him. BASF Brussels “employs” EVA in two ways: firstly as a virtual receptionist, taking calls from external callers; and secondly as an automated attendant for her internal “colleagues” at BASF who want to reach each other by just saying a name. The human operator at BASF Brussels switches incoming calls over to EVA during lunch breaks, after hours, and in cases of an overflow.

The BASF talk stood out, because it was the only one really focusing in on human factors. Mr. Edel listed the different attributes of internal vs. external callers: conscient or not about the speech attendant, knowledgeable or not about the called company’s organisational structure, etc. The callers’ different set of expectations has a direct impact on EVA’s ergonomy, for example on the list of recognised department names: generally speaking, external callers don’t know subdepartment names.

Peter Edel further urged system administrators to treat EVA with care: a badly chosen initial prompt can easily undermine all technical work in one minute; having the system ask the right question at the right time is more important than tuning speech recognition. He also made a plea to accompany EVA’s technical setup with sufficient communication to the internal users.

Text-to-speech technology is not considered perfect, but more than “good enough” to serve in automated attendant applications. Its flexibility makes it easy – and therefore economically interesting – to quickly tune system questions, in case an ergonomic audit has shown shortcomings in that area. In the same vein, speech recognition statistics are seen as an important tool for improvement.

The main message of the talk was this: as important as speech recognition accuracy may be, there are many other success factors. Ergonomy, user involvement and integration in the existing business processes are the most prominent ones.

Philippe Mercken (Dexia Direct Private) speaking at speech technology seminar on Nov 9, 2006

The last talk of the seminar was given by Philippe Mercken, Manager Dexia Direct Private, with assistance from myself. Dexia ranks among the fifteen largest banks in the euro zone and is a world leader in public finance. In Belgium and Luxemburg, the bank offers retail services to several million customers; it has also developed a private banking business for affluent customers in Belgium, Luxemburg, France and other European countries. Dexia Direct Private (DDP) was created for active high net worth individuals: the service allows them to have trading instructions executed by a Dexia professional on the basis of a simple phone call. Dexia wanted to offer these approximately 500 demanding customers a convenient (DTMF-free) and secure method to authenticate themselves over the phone.

In October 2005, a consortium led by Dimension Data Customer Interactive Solutions took up the challenge to build a speaker verification system for DDP that would have to be operational three months later. A small team from Nuance Aachen was called in to deliver the underlying speaker verification (SpeechSecure) and speech recognition (OSR) software, to design the Voice User Interface, and to write the SRGS grammars for speech recognition. I was called in to provide general guidance to the project manager & end customer, to develop the speech application (Genesys Studio + custom VoiceXML code to integrate with SpeechSecure), and to perform usability tests. Telephony specialists from Dimension Data took care of technical design and integrated the system with Dexia’s databases and Genesys’ routing engine.

The system went in production in the course of January 2006, as foreseen. DDP professionals rang up their customers, explained the concept, and asked them to enroll by providing answers to multiple secret questions. On the basis of the customer’s voice samples, voice prints were created and stored in a secured database (all on premise, of course). Immediately after enrollment, customers started calling in, simply said their name, gave the (correct) answer to one of the secret questions and were then transferred to the DDP professional, with a clear on-screen indication of their authentication status. Mr. Mercken testified that after 6 months of operation, hundreds of customers are regularly and successfully using this system. He mentioned two success factors: the application targets a customer type receptive to this type of concept, and the quality of the briefing preceding each enrollment is high.

From my own point of view, the most important success factor has been a constant and proactive management of expectations in Dexia’s project team. Because of speech applications’ human-like nature, they tend to create strong opinions, usually in opposite directions: some people think speech applications are science fiction – meaning they will never work; others think they are simply magic and can handle any type of caller input – wrong again. It is vital for speech project managers to guide first-time users and customers in what they should and should not expect.

So speech technology is not magic, it’s (only) technology. Realism is the word.

Judgeing from the discussions I had at the afterparty, the seminar was well received, and greeted as a timely initiative. For me it was really great to see so many local people (re)gain confidence in speech technology, after years of misperceptions in both directions.

A heartfelt thank you to all speakers, sponsors and attendees!

3 Responses to “Prominent Belgian speech tech users coming out at seminar”

  1. [...] Next month, the Speech Technology Workgroup organizes its second half-day seminar on “Speech Applications: Customer Experiences in Belgium”. The first edition in November 2006 already attracted a sizeable audience, so to accommodate the expected larger turnout the organizers chose as new venue Belgacom Towers, next to Brussels North Station. The event takes place on Tuesday Feb. 12, 2008 in the afternoon. [...]

  2. [...] Two years ago, Dexia Bank of Belgium launched a speaker verification system for its private banking customers. My role in this project included general speech technology assistance to the project leader, expectation management, VUI development, technical integration of the speaker verfication engine with the application, and usability testing. [...]

  3. [...] On the 12th of February the Belgian speech technology industry gathered in the Belgacom Towers for the second edition of the seminar “Speech Technology: Customer Experiences in Belgium”. Apart from seminar host Belgacom, the event was also sponsored by Acapela, DBScape, Genesys, NextiraOne, Nuance, Quentris and The Ring Ring Company. Chairman Alain Rondenbosch of the organizing Speech Technology Workgroup was very pleased to welcome a larger audience than at the first edition in November 2006. [...]

Leave a Reply