We are excited about the opportunities presented by Answer Bot and continue to lean into training and optimizing the model. However, we have come across a major difficulty that we think is easily solvable:
- When a customer located in one country sends a request via email, they might receive content that is intended for users in another country. This is discernible via checking the Locale of the user (e.g. en-us) and the Locale of the article that is offered to the user (e.g. en-gb).
We think this can be resolved by enabling "strict locale matching" on an Answer Bot ticket. Here's how this works out for us:
- Our Help Center is set up to support many countries, many of which will speak the same common language (example: the United States and the UK).
- However, our content will differ between countries. For example, eCommerce return procedures may differ slightly between the US and the UK
- Due to the volume of articles we must manage, we have found it best to target our articles to end users in different countries by creating one article, then modifying the content for each locale. Example: "How to return an item from your order" may exist as one article, but there will be two translations: en-us and en-gb. Both translations are technically written in English, but the en-us content will contain the US return procedures, while the en-uk content will contain the UK return procedures.
- The end user experience on the Help Center is thus: when the user is viewing content with "English (US)" selected, viewing the "Returns" article will display the US-centric information. When the user is viewing content with "English (GB)" selected, the en-gb translation will be shown and the user will be exposed to the UK-centric information.
This allows us to serve one link globally to our customers about our return processes. By omitting the locale subdirectory in the URL, we allow Zendesk Help Center to determine the user's location based on browser locale and serve a personalized experience. It is a very scalable and flexible approach and an example of Zendesk's "beautifully simple" design.
Unfortunately this does not play well with Answer Bot. Answer Bot detects the language of the requester automatically, and then serves any relevant articles it can find, in any related locale. This means that an end user in the US might get a response with UK-centric information (in fact, it is happening).
The simple solution to our problem (from our admin perspective) is this: allow for strict locale matching in the Answer Bot trigger. This means that if an end user has "English (US)" set as their locale in the end user profile, only matching articles with an en-us translation will be returned, and only the en-us version will be offered to the end user. This could be set by default, or offered as a checkbox as part of the AB trigger configuration (similar to how label restrictions are set today).
After raising this concern in a support ticket, the support team offered the solution of applying article Labels in Guide to relevant articles in each locale (e.g. label UK articles with uk_customer and US articles with us_customer), creating a unique trigger to target each requester language (e.g. "Serve Answer Bot Article to US Customer" and "Serve Answer Bot to UK Customer"), and then restricting the articles served by that trigger based on the aforementioned labels.
Unfortunately this approach won't work for us, because labels are stored at the article label and not at the translation label. We have explained above why we scale better by creating translations of a single article for each locale, instead of creating new articles for each locale. We cannot at this stage change our Help Center structure to accommodate Answer Bot (nor would we want to because of the other benefits realized with our current structure).
We know that in many cases it is helpful to leverage the NLP language detection inherent to Answer Bot. However, allowing the admin to configure how AB chooses the locale would maximize the return we gain from the solution. We are in a position to set the requester language with a trigger based on a "Received at" email value in our Zendesk, meaning that we can always detect the user's location with a very high degree of confidence based on the email address the user submitted the ticket to. As a result, by placing the Answer Bot triggers after these language-setting triggers, we would significantly enhance the result of our Answer Bot offerings by enabling a strict locale match on each of those triggers.
Please sign in to leave a comment.