This may be otherwise called personalization. A multi-disciplinary institution may offer between 20 to 100 courses and specializations. A given students’ eligibility and interest in less than 10% of the courses offered subject to her own academic background and career plans. It is hence imperative that the bot first queries students’ interests in streams and courses before opening the floor for free-flowing conversation. This can be achieved by a form like a gathering of interactive inputs at the start of the session followed NLP powered free-flowing conversations.

Its’ equally important to persist the above data for all future sessions with the given prospect. Considering the blocking nature of these choice components, students simply cannot be presented with these options every single session. Such choice persistence beyond the first session (24 hours in case of Whatsapp, in case of web and other media it may be configurable) also give the institution some guidance on the content choices while re-marketing.. more on this later

The most fundamental scale a chatbot is measured on is conversation coverage, that is the Artificial Intelligence engines’ ability to resolve an utterance to the correct intent. This may also be the only metric a prospective student measures the bot, and the institution that has deployed it, on.

The key to bots’ conversation coverage is continuous improvement. We at askadmissions.ai started with about 50 topics, at the time of writing this blog we have over 300 topics with multiple variants covered on our bot. We reached there through continuous improvement and some toil. While institutions that are located in the same territories that target the same students may have to train their bots on similar intents, they also tend to attract some very specific queries. For examples queries about Admission policies, Fees, specific faculties who teach certain subjects may be expressed significantly in different phrases. To start with acceptable quality and then proceed towards reaching 80 to 90% conversation resolution by your preferred NLP engine may be a pragmatic approach. Such an approach also helps the conversation designers to account for the seasonal fluctuations in traffic and intent through an academic year.

Having the right tools to gather and inspect user feedback is paramount.

When the bot fails to respond, it's not the failure of the underlying technology that annoys the users, it's the wrong handling of the failure with little exit or fallback options that leaves them frustrated. Answers and responses should be structured to make it easy for users to navigate related topics and variations. A ready list of related FAQs goes a long way in easing Intent resolution errors. Askadmissions bots evaluate each for intent as well as for semantic clues. When intent resolution fail, relevant faqs are presented instantly

Finally, human assistance as a last fallback is inevitable. Especially since college admissions chatbots operate at the point of student acquisition, institutions may want to communicate that they value each interaction enough to provide counselor assistance.