Can a Virtual Assistant damage your organization’s image?

A simple idea like “Virtual Assistant’s (VAs) do have an impact on the users’ perception about the organization that they represent” should suffice to prompt a tight collaboration between business owners and service providers. When an organization cares enough about customers’ feedback as to invest money, time and effort in order to improve satisfaction levels via additional channels like VAs, then one might think that they will try to meet best quality standards by getting actively involved in the product’s development process inasmuch as possible. Once the final product is released, it belongs to customers.

The truth about users’ stories is that, however positive or negative their experiences may be, these will spread on the Internet without control. In fact, customers’ opinions on poor performance may impact heavily on the lifespan of web applications, especially VAs.

For those users with very specific needs, interacting with VAs whose “small talk” strategies are limited (or out of the scope of the conversation) can be acceptable to a certain degree, as long as major requests are solved somehow. This scenario can be very well exemplified by Sarah, Royal Mail’s online VA.

Royal Mail’s Sarah
• Release date: June 2008
• Platform: Web (Royal Mail’s website)
• Country: UK

Sarah provides information on Royal Mail’s online services in a fairly effective way, and this should be enough for consumers to perceive her as intelligent. However, the experience of starting a personal conversation with her can be compared to trying to get Google’s search engine answer questions like “What’s your name?”, “How old are you?”, “Are you married?”, etc. One is never going to succeed. But of course, the difference is, users all over the world know that Google is indeed a search engine; what’s deceiving about Sarah is that she is an embodied agent, with a human-like appearance and a name of her own, and, as such, one would expect her to be aware of these – otherwise obvious – personal details.

As a step further, though, what may not be as acceptable is any VA showing little or absolutely no command of the areas supposedly comprising its expertise. When something like this is observed in a VA, users’ comments, assessments and sample (negative) dialogues will most likely start spreading on the Internet right away. Lucía, an MSN chatbot associated to the Andalusian Tourism Board in Spain, failed to offer quality information on areas that she claimed to know in her welcome message.

• “Andalucía en tu Messenger” (Andalusia via Messenger)’s Lucía
• Release date: July 2008
• Country: Spain
• Platform: Windows Live Messenger (

When asked about beach recommendations for Almería, she would prompt the user to be a bit more specific about the geographical area, just to end up ignoring his/her information with a senseless remark like: “I don’t know anything about that, but let’s talk about Andalucia!”

Lucia: Conozco muchas (107) playas en Almería. Filtra más por favor.
User : Mojácar
Lucia: Oye, de eso no sé nada, ¡pero de Andalucía sé mucho!

Finally, at the other end of users’ acceptability scale (exemplifying what’s not acceptable at all) would be VAs providing misleading information, especially with such complex topics as health, drugs, sexuality and STDs. This is exactly what Dr. Robin (MSN chatbot for the Ministry of Health, Spain) had been doing for not even one week before the Ministry itself decided to discontinue its service.

Ministry of Health’s Dr. Robin
• Release date: January 2008
• Country: Spain
• Platform: Windows Live Messenger (

Dr. Robin, whose target audience was teenagers, provided misleading information on homosexuality, pregnancy and alcohol, so, after overt parents’ complaints, expressed in many different forums, most of its knowledge was discarded, and it ended up providing real basic information on minor health issues.

In many cases, then, succeeding is not a matter of standing out; but a question of not under-performing. Optimizing a VA’s performance is a collaborative process: organizations and product developers should get involved in an iterative process that includes knowledge structuring, Quality Assurance standards, acceptance tests, etc. Once online, VAs’ overall quality will be measured by customers and it is precisely these customers who will evaluate the organization and its image in the end.

Therefore… please get actively involved in your VA’s coming to birth. You might regret not doing so!

Carmen is currently working as a Knowledge Engineer in the Information Technologies Sector. From 2005 to 2011 she worked for the University of Seville's spin-off company Indisys. During this period, she was also a member of the Julietta Research Group at the university. Carmen holds a B.A. in English Language and Literature from this same university. She spent the first year of her graduate studies as a visiting student at the Department of Linguistics at the University of Pennsylvania, USA. Her prior research experiences include a position as a phonetic annotator at the University of Poitiers, France. Her interests include Non-Verbal Communication, Sociolinguistics, Discourse Analysis, Natural Languague Processing, Dialogue Systems, and Anaphora Resolution.

Leave a Reply

Your email address will not be published. Required fields are marked *