Learn how AI powered sports software helped event organisers

, , ,

A Sports Software Chatbot Case Study: The Fred Whitton Challenge Sportive automated assistant, advanced

We report on our AI chatbot sports software project to aid the organisers of one of the UK’s most well-known cycling events.

 

Leverageing ai powered sports software with our core Aktivebot chatbot the goal was to create an automated assistant available 24/7 to reduce time and effort needed by event organisers to respond to event enquiries whilst still providing an easy way to contact the events team if necessary.

The Saddleback Fred Whitton Challenge is a charity event in honour of the late Fred Whitton consisting of a 112-mile charity sportive around the Lake District and is arguably one of the UK’s most well known and hardest sportives with over 2000 riders and 5000 applications this year.

 

 

The chatbot sports software was used by organisers to cover the event

The Fred Whitton Challenge has been running since 1999 and as a result is extremely popular with over 4000 followers on their Facebook page where a large number of ride questions were being asked via the message me button there. We wanted the AI chatbot to assist the event organisers in answering ride and registration queries and reduce the amount of time spent answering routine questions. We also wanted to provide the ability for users to look up their time for this year and previous years.

The chatbot we created is integrated within the “Facebook Messenger app” of the Fred Whitton page and users can contact it through the private “Messages” feature of their page, or directly through the Messenger App.

The sports software project

The project brief was for The Bot Forge to create an AI powered chatbot capable of handling event enquiries 24/7 which could be deployed into the Facebook Messenger framework and utilise rich ui elements. Future deployments could be aimed at website integration.

For such a long-running event, Human Race and the Fred Whitton organisers wanted to provide the optimum user experience and still make it easy for participants to message organisers directly through the chatbot if they wanted to contact a real person by messaging them directly.

The chatbot understands human language, leveraging advanced Natural Language Processing and answers questions such as “what is the fred whitton?”, “ I’ve injured myself at the weekend I need to defer till next year”,“ when can I get my race pack?”, “ help I need the GPS files for the route”, “ Is there any way to buy a jersey post-event?”,”I want to contact an organiser”, and “when will the results be available?” The chatbot replies to a question based on it’s own programmed data or points to the specific information on the Fred Whitton Website so that it works in tandem with the website itself.

Press the play button to watch a real conversation with The Fred Whitton Chatbot

The technology

We used Google Dialogflow to provide the NLP engine and Google Firebase for the fulfilment hosting. The fulfilment or web-hook is where we were able to compute more complex answers for the AI chatbot to give to users and create the correct responses for. For example when looking up users past ride times, the web-hook was able to look up past results for users from a results database. Facebook ui elements added rich content, particularly useful when asked about merchandise details and availability; linking directly through to the official shop.

The conversations

The real challenge in creating the chatbot was leveraging natural language technology that can support the range of questions that event participants might ask: for example, all the different ways that people might ask about the route. We are helped in this process by our own Aktivebot pre-created sports events intents.

Small talk

The chatbot includes the ability to provide small talk, which is used to provide responses to casual conversation. This feature greatly improved user experience when talking to the agent.

Initial question data

Initially, we imported the pre-created sports events intents (an intent represents a mapping between what a user says and what action should be taken by the chatbot).

We then looked at FAQ data provided by the Fred Whitton steering committee and historical questions to their facebook page which gave us some invaluable insight. Using this information we were able to create the conversational scripts and then implement the conversation ability with each question matching an intent

This was an iterative process. Matching user intents to core functionality and features and training the natural language processor to understand users and handle conversation failure scenarios gracefully.

The conversational UI was then fine-tuned, with rich elements implemented where necessary.

What were the questions?

Most asked questions by participants match the questions that the event chatbot is able to answer, i.e.:

  • Questions about registration: deferring places, available places, waiting list enquiries.
  • Questions regarding merchandise: jerseys for sale on the day.
  • Questions about the ride: route details, information about closed roads, clothing enquiries.
  • Questions after the event: results, photos availability, the next ride date.
Sports Software Chatbot- Top Intents handled by conversational agent

Top Intents handled by the conversational agent

The training

The questions were often related to ride specific information. This meant that for an optimal intent matching rate, it was necessary to work closely with the event organisers to provide answers to specific questions. The capabilities of an ai sports software chatbot will improve over time, the more messaging transcript data the better so the more it’s used the better and more accurate it will get. Hence the training logs were checked multiple times a day and improvements made where necessary. By focusing on all questions answered it is possible to greatly improve the intent matching rate of the chatbot over time.

The training data was invaluable for perfecting the bot conversations. The process highlights any need for new responses as a continuous cycle of continuous learning.

The “training” of the chatbot can then be used from one year to the next. Any event detail changes can be carried out easily.

Results

The sports software chatbot was launched on 21st March with the scope constrained to Facebook Messenger with no advertising whilst the chatbot was evaluated.

Activity

The high number of participants using the chatbot can be explained by the fact that visitors still have questions that the website itself does not answer or does not answer quickly enough. The chatbot was, therefore, a great place to provide up to the minute event information, such as information about closed roads and the slight route change which resulted in one more hill showing.

The chatbot was not heavily advertised so we envisage activity levels will improve as participants get used to the chatbot as a resource they can use and other strategies to engage users are utilised.

The chatbot was answering questions on the run-up to the event and also during and after.

Success rate

The success rate of the chatbot to answer queries was overall around 60%. With more focused training over a longer period with another event in 2019 we expect this figure to rise until our aim of an 80% success rate is reached.

Sports Software Chatbot Success Rate Over the Past 30 days,

Chatbot Success Rate Over the Past 30 days

Feedback

The chatbot worked well in Facebook Messenger as its one of the preferred channels for chatbots in general. Deploying the chatbot in a chat widget as part of the website itself would undoubtedly result in more engagement and something to consider for the future.

Help intents and the handover protocol were also very successful. If a user did not get a correct response and/or wanted to get help or contact an organiser directly this worked really well. The overall feedback from users was positive. There were always some intents which the bot would struggle to match the first time which would be handled gracefully; however, due to the ability to train the chatbot, leveraging AI the correct response would be prepared for next time.

I’m impressed with the chatbot it seemed to work well. I think it is a good source of help and with it learning as it goes along it would answer lots of questions going forward. If it cannot help it still contacts the organisers where we can answer.

Carolyn Brown: Fred Whitton Challenge Steering Group — Saddleback Fred Whitton Challenge

The Fred Whitton Challenge chatbot still has many areas where it can be developed and improved, particularly by providing more integration with existing systems and utilising push notifications: this will be something carried out in the future.

Overall the success of the chatbot hightlights the benefits of deploying this type of ai sports software in sporting events and is definitely something to consider to give event organisers an advantage in a competitive market

Read about our AI Chatbot in the inBusiness Science and Technology spotlight

, ,

INBUSINESS COVERS OUR AI CHATBOT IN SCIENCE AND TECHNOLOGY SPOTLIGHT

It was great to have our AI Chatbot featured in the inBusiness magazine issue spotlight this month. You can read the feature here 

inbusiness ai chatbot article

Image: https://chambermk.co.uk/profile/inbusiness

Inbusiness is a bi-monthly publication and digital magazine created by distributed to over 3,000 business contacts in and around Milton Keynes. The June/July 2018 issue spotlight was science and technology so it was great that the editors of the magazine wanted to cover our Fred Whitton Challenge ai chatbot, particularly when the ai chatbot was created to assist the organisers of a charity ride.

You can learn more about our chatbot agency here.

We also cover further technical details about the project here.

Google give an amazing demo of their google assistant making a phone call

, , , ,

ONSTAGE AT I/O 2018 Showcasing Google Assistant

GOOGLE HAS STARTED OFF ITS ANNUAL 3 DAY I/O DEVELOPER CONFERENCE AT SHORELINE AMPHITHEATER IN MOUNTAIN VIEW, CALIFORNIA. 

In the first day, they have shown some of the amazing new capabilities of Google Assistant. One of them is being able to make phone calls on your behalf. You ask Google Assistant to make an appointment and it makes the call in the background. The demo has to be seen to be believed.

CEO Sundar Pichai played back a phone call recording that he said was placed by the Assistant to a hair salon to book an appointment.

With a voice which sounded totally natural; the person at the salon had no idea they were talking to an automated AI assistant. The Assistant even managed some small talk; dropping “mmhmmm” into the conversation.

Pichai reiterated that this was a real call using Assistant and not some staged demo. “The amazing thing is that Assistant can actually understand the nuances of conversation,” he said. “We’ve been working on this technology for many years. It’s called Google Duplex.” Pichai also made the point that Duplex was still under development and that Google plans to conduct early testing of Duplex inside Assistant this summer. Their aim is “The technology is directed towards completing specific tasks, such as scheduling certain types of appointments”

Google has a blog post with more Duplex information here which has a lot more examples of Duplex in action using different voices, for example, a really interesting one making a call to a restaurant to book a table.

Google again states that these are real-world examples.:

“While sounding natural, these and other examples are conversations between a fully automatic computer system and real businesses.”

This post also does a good job of highlighting some of the real complexities of having a conversation successfully. With many sentences having different meanings depending on the current context. In the same conversation early on the assistant also handles misinterpretation when the person called mentions a table number taken from what she has misheard. Google Assistant seems to handle this perfectly.

This looks set to be groundbreaking technology:

For users, Google Duplex is making supported tasks easier. Instead of making a phone call, the user simply interacts with the Google Assistant, and the call happens completely in the background without any user involvement.

We are looking forward to seeing more of it in summer and using the technology in our projects.

With Google also announcing their rebranding of its Google Research division to Google AI. The move shows how Google has increasingly focused R&D on natural language processing and neural networks.

It looks like Google are setting their sights on being the world’s biggest artificial intelligence (AI) company. 

Facebook has removed its pause on the app review process

, ,

In the wake of the Cambridge Analytica scandal Facebook announced late in March that it was pausing its app review process, which meant developers were no longer able to launch new apps or chatbots on the Facebook ecosystem.
It was an abrupt halt and although temporary was not ideal for any developers planning to unleash a shiny new Facebook Messenger chatbot into the wild. We were lucky ourselves that we had got one of our own bots The Fred Whitton Challenge Chatbot live a few days earlier

True to their word the pause has only been a few weeks so its great news to hear that Facebook has re-enabled the app review process so that new chatbots can now be connected to pages and set live.

App Review is Back

Today we are re-opening our app review process. The process has changed a bit as we now require business verification for apps that need access to specialized APIs or extended Login permissions. Apps that ask for basic public profile or additional permissions, such as birthday or user friends, are not subject to business verification.

 

You can read Facebook’s official statement here.

The point has to be made however that this is a helpful reminder to not be reliant on a single platform. Particularly as Facebook has a habit of changing things. Facebook executives Campbell Brown and Adam Mosseri have also stressed the idea that publishers, at least, should not be too tied to Facebook while speaking at Recode’s Code Media conference. If users find the whole experience too unsettling then Facebooks answer is “If anyone feels that this isn’t the right platform for them, then they should not be on Facebook,” Brown told Recode.

Here at The Bot Forge its good news and we are pleased to be able to get cracking building some fantastic new bots.

The Non-Technical Guide to Popular AI Terminology

, ,

AI Terminology Cheatsheet

Artificial Intelligence is talked about everywhere these days. In the news, media and extensively in science. We mention it a lot on our website and blog after all this technology is at the core of what we do at The Bot Forge.
You may well have encountered some of the different terminology used. But what do developers and technologists really mean when they use these terms? Having a simple understanding of some of the more frequently used terms can be useful when thinking and talking about your chatbot strategy. This AI terminology cheatsheet aims to help you understand; no technical knowledge required!

  1. Algorithm

    An algorithm is a formula for completing a task. Wikipedia states that an algorithm “is a step-by-step procedure for calculations. Algorithms are used for calculating, automated processing and data processing and provide the foundations for artificial intelligence technology.

  2. Artificial Neural Network

    Artificial Neural Networks or ANN are artificial replicas of the biological networks in our brain and are a type of machine learning. Although nowhere near as powerful as our own brains they can still perform complex tasks such as playing chess, for example AlphaZero, the game playing AI created by Google.

  3. Artificial Intelligence

    AI research and development aims to enable computers to make decisions and solve problems. The term is actually a field of computer science and is used to describe any part of AI technology of which there are 3 main distinctions (1)

  4. Autonomous

    Autonomy is the ability to act independently so software which can complete tasks on its own is autonomous for example systems which manage self-driving cars.

  5. Big Data

    Big data describes the large volume of data – both structured and unstructured – that floods through a business and its processes on a day-to-day basis. In the context of AI big data is the fuel which is processed to provide inputs for surfacing patterns and making predictions.

  6. Chatbots

    I think we have mentioned these once or twice! A chatbot is a conversational interface powered by AI and specifically NLP. They can be text-based, living in apps such as Facebook Messenger or their interface can use voice-enabled technology such as Amazon Alexa.

  7. Cognitive

    Cognitive computing mimics the way the human brain thinks by making use of machine learning techniques. As researchers move closer towards transformative artificial intelligence, cognitive will become increasingly relevant.

  8. Deep Learning

    Also known as a deep neural network, deep learning uses algorithms to understand data and datasets. Deep Learning is a subfield of machine learning concerned with algorithms inspired by the structure and function of the brain called artificial neural networks. Deep Learning techniques have become popular in solving traditional Natural Language Processing problems like Sentiment Analysis.

  9. Entity and Entity Extraction

    Entities are also sometimes referred to as slots. An entity is used for extracting parameter values from natural language inputs. Any important data you want to get from a user’s request will have a corresponding entity.  Entity extraction techniques are used to identify and extract different entities: Regex extraction, Dictionary extraction, complex pattern-based extraction or statistical extraction. For example, if asked for your favourite colour you would reply “my favourite colour is red”. Dictionary extraction would be used to extract the red for the colour entity.

  10. Intelligent Personal Assistants

    This term is often used to describe voice-activated assistants which perform tasks for us such as Amazon Alexa, Google Assistant, Siri etc instead of text-based chatbots.

  11. Intent

    An intent represents a mapping between what a user says and what action should be taken by your chatbot. A good rule of thumb is to have An intent is often named after the action completed for example UserProvidedColor.

  12. Machine Learning

    Probably used by you every day in Google search for example or Facebooks image recognition. Machine learning allows software packages to be more accurate in predicting an outcome without being explicitly programmed. Machine learning algorithms take input data and use statistical analysis to predict an outcome within a given range. Machine learning methods include pattern recognition, natural language processing and data mining.

  13. Natural Language Processing

    Natural language processing (NLP) enables machines to understand human language. Machine learning is used to find patterns within large sets of language data sets in order to recognise natural language and aid machines in understanding sentiment so that they can respond correctly.

  14. Sentiment Analysis.

    Sentiment Analysis is the process of determining whether a piece of writing is positive, negative or neutral or more advanced analysis would look at emotional states such as “angry”, “sad”, and “happy”.

  15. Utterance

    An utterance is anything the user says via text or speech. For example, if a user types “what is my favourite colour”, the entire sentence is the utterance.

We hope you have found this AI Terminology Cheatsheet helpful. If you want to talk about your chatbot project contact us at The Bot Forge

Comment if you think I’ve missed any terms out which should be on the cheatsheet

Shimano Chatbot

,

We like this example of a chatbot from Shimano which was launched in Facebook Messenger. Facebook users can find information about their products, events, technical documentation and a link through to a dealer locator. We particularly like the product recommendations functionality as it demonstrates how a smart recommendation can drive sales.