Skip to content
  • Technology & data

Building ‘The Chatbot’ - Our experience with GenAI


by Antony Heljula

Learn how we harnessed to power of Generative AI to build our very own chatbot.

Generative AI (GenAI) is providing huge opportunities for organisations to streamline and improve their processes. This isn’t a new thought and has been said in countless articles over the past few years. But at TPXimpact, we have first hand experience of implementing this technology and seeing the great results it can produce, thanks to our new chatbot.

Since its emergence, GenAI is something that we’ve been keen to put in place. After all, we’re a digital transformation organisation who can support our partners with implementing these technologies, so it’s important that we can demonstrate our capabilities and expertise. Our people also wanted to see us make the most of GenAI and lead by example in this space, which was an additional driver for us, so we got to work. 

Making GenAI benefit everyone

With a clear goal in mind, our next step was to decide where we could use GenAI to make the biggest impact. I have previously discussed how HR teams can benefit from these solutions, and following discussions with our People Team, it was clear that creating a chatbot with AI capabilities would be of huge benefit to everyone. 

The TPXimpact People Team, like most HR departments, receives a large amount of questions via different channels, many of which cover similar topics and require repetitive responses taken from policy documents. The idea of using a chatbot to automatically respond to and answer these questions had been discussed before, but as this would need a large amount of FAQs to be drafted, inputted and continuously updated to work, this was never progressed. 

However, the emergence of GenAI allowed us to revive this idea, as implementing it would allow the bot to find information in policy documents automatically and by itself, making the process far simpler. In tech circles, this is known as Retrieval Augmented Generation (RAG).

Building the people’s chatbot

To create a platform that met the needs of our people team and those coming to them with questions, we first decided to use low code and no code solutions. Harnessing ‘out of the box’ bots and GenAI technologies that we just needed to programme meant we were able to quickly develop a functional tool within a month, much quicker than if we started from scratch. 

Once we had a chatbot in place, the next challenge was getting it to be able to read through policy documents and identify the necessary information to answer questions. To do this, we uploaded over 35 policy documents into the bot and taught it to understand certain business jargon and abbreviations, such as "TPXimpact", "TPX" and "company" all mean the same thing.  We then ran a series of internal tests and demonstrations for these functions, which it came through successfully. 

Finally, we put the chatbot through a series of testing phases with different stakeholders to ensure everyone’s needs would be met and any kinks could be ironed out. This helped us identify and input additional features, such as auditing and logging processes to capture information. Throughout these phases there wasn’t a single instance of the platform answering a question incorrectly and it was only unable to answer a small number of questions due to a lack of information, not errors, which filled us with confidence that this project would be a success. 

Following testing, we discussed running a few more trials, but came to the conclusion that progress is better than perfection with GenAI, and that we should launch the tool straight away, while always looking to update and improve it.

Throughout this entire process, one thing that was central to our work was ensuring users’ needs would be front and centre. One example of this is that we made it so questions could be asked anonymously. This is because people may have sensitive, personal things they need answers on, and not being able to do this discretely may put them off finding vital information that could help them. Anonymity is something that I believe should be central to all AI projects

‘The Chatbot’ will see you now

In January 2024, our chatbot (officially named ‘The Chatbot’) went live and we’re already starting to see the benefits. The platform has already received over 430 questions, with 55% of these being answered successfully, freeing up valuable time and resources for our People Team. At the same time, no questions have been answered incorrectly and no hallucination or inappropriate responses have been given, ensuring the tool can be used safely by everyone. 

Now that The Chatbot is up and running, our work doesn’t stop here. In line with our goal of continuous improvement, we have identified 15 areas for content optimisation, which we will implement to improve the bots capabilities, while we have also included Marketing FAQs into the platform, helping even more teams benefit from its abilities. 

Using GenAI and building The Chatbot has been a great project to undertake and taught us so much, all while allowing us to demonstrate our capabilities and, most importantly, support our people. We are already working with other organisations to help them do the same, and we can’t wait to further share our experience and knowledge so that everyone can reap the rewards of this incredible technology.

Antony Heljula's avatar

Antony Heljula

Technology Director

Contact Antony

Our recent insights

Transformation is for everyone. We love sharing our thoughts, approaches, learning and research all gained from the work we do.


Common misunderstandings about LLMs within Data and Analytics

GenAI and LLMs have their benefits, but understanding their limitations and the importance of people is key to their success.

Shaping product and service teams

How cultivating product and service teams to support the needs of the entire product lifecycle can ensure brilliant delivery.

Revolutionising transport: a systems thinking approach

Changing the way we think about transport is key to improving services and outcomes