From Strings to Things.

How Data Scientist in @myob are transforming online search.

Hi, my name is Aamir Mirza, a Data Scientist and (Artificial intelligence) AI researcher working at @myob Australia. In this blog post my focus is to highlight some of the efforts being made to take product and functional search to the next level. This is being done in concert with Operations, Contact Center and product documentation teams.

As most users of the MYOB online help would be aware, the search feature in @myob products could use some improvement, it is based on a dated technology of string match or keyword search. This relies on the user knowing the correct keyword to search for to narrow down the results or they would need to skim through dozens of pages to find what they are looking for. This of course leads to wasted time and frustration. We saw this as an opportunity to improve search capability which would deliver accurate results and save our users time. Our motivation is driven by the question, “why can’t the search be more practical?”. In reality we ask a question and get the answer specific to the question, for example when contacting technical support. Keeping this in mind let me introduce you to some of the exciting AI powered technologies we have been building to make online search more humanlike and effective.

MYOB intelligent Suggestions.

Let’s face it, whether you are a new user or a seasoned professional, when it comes to online search it is sometimes difficult to articulate your query in a refined and effective manner. This is where the AI powered intelligent suggestion comes in, for them strings are not just a collection of keywords, the AI understands context or the sequence of words the search should be such that the underlying search engine provides meaningful results. Of course, nothing operates in a vacuum, seasoned professionals from product documentation, Contact Center management and Operations are working in tandem with Data Scientist to bring this ground-breaking technology to MYOB clients. For those wanting some visual delights here are some screen shots from the internal version being beta tested.

· Intelligent suggestions for most abstract queries.

· Completely resilient to typos. ( We understand even when not spelled right)

· OR this.

·     Tell the AI what to focus on (refined suggestion).

Still do not see what you want, you can always type your own.

Although we have trained our AI with thousands of examples, in the rare case you are unable to see what you want, you can always search free form. The AI is designed to learn from the way people search, over time it will improve and adapt to your search style. The recurring theme is to help you find helpful and relevant information faster.

We hope that this contribution from the Data Science team will help you go from strings to things that you want.

Understanding user intention.

Once we have the refined query provided by MYOB intelligent suggestion, we use AI and machine learning to understand the intent of the query .i.e. whether it is something which can be answered by product search, or the knowledge base, or it is more about expressing opinion or wanting an opinion, and in some cases even understanding if there is enough information to answer the query in a meaningful way. Based on the intent classification results we get, try to channel user in the appropriate way.

Deep learning based natural language search.

The journey does not end here, once you have picked (through Intelligent suggestions) the query most likely to answer your question, that query is then fed into our deep learning search engine to intelligently match internal knowledge-based articles most relevant to your search. These AI agents are designed from the ground up to take into account the whole context of your search and present information in the most meaningful manner. To them strings are not just keywords, they are a conversation to be answered in a meaningful way.

In the near future we hope to make technology available for both customer facing online help as well as internal staff. The whole idea is to get where you are going a whole lot faster and significantly reduce wait time to the contact center.

Towards a more exciting future.

In the near future, AI and machine learning will transform the help interaction space into an intelligent product which will not just be a passive observer or tool, but an active participant in your success.