Implementing a lightweight quick-win pilot project to kick off conversations about AI and digital change in a 200 year old company.
Building Your AI Trojan Horse: The Humble Chatbot
It´s a conversation I´ve personally witnessed in a few large companies. It goes like this.
“We need to get into AI. Like, I´m not sure how, but, we need to be, you know…doing, AI stuff”.
It appears many companies feel this way. All the neighbours are “doing AI”, so, we better “doing AI” too, but, where to start?
The fact of the matter is, onboarding AI is not dissimilar to onboarding any other system, sales strategy or graduate recruit. It can be turned to all sorts of areas, but the best use cases often come from something very specific, measurable and repeatable.
For bigger companies, this is all the more important as part of the battle around AI adoption will be getting coworkers bought into the process too. They´re already using public LLMs for daily tasks, and probably sharing company information in the process. Building proprietary systems, retaining control of company data and steering messaging, is idea, but it´ll take a bit of time. But an outright “ban” on doing so, simply won´t work. People have personal phones. People are resourceful.
Instead then, one way of getting a company “bought in” to using AI internal systems is by building a simple tool. A tool the business could make use of on a daily basis. A tool that has usage analytics than can be assessed and lessons taken forward.
For a major construction client, one immediate thought to “let´s do AI stuff” was to build a chatbot.
For me, the humble AI chatbot is a perfect example of the challenges of AI adoption.
It only takes a conversation with a badly programmed chatbot (you know, like most of the customer service chatbots that are currently out there) to get a user wound up, frustrated, and disillusioned with the whole process. Because often these chatbots are programmed to attempt to do a myriad of tasks for a wide variety of members of the public.
However, b2b businesses are a bit different. In fact, by looking at the search history on the construction company´s website and combining this with the website´s google analytics traffic, we saw that almost all the traffic could be broken down into a few key areas
- How do I apply for a job?
- How do I join your supply chain?
- Who do I speak to about projects in X region
- Do you guys build X type of project?
Of course, the details for each of these questions could already be found on the company website, however, as is the case with many b2b companies, a traditional focus on accounts and relationship building via business development teams, rather than pushing prospective customers to the website meant the company website was less of a shop and more of a company wikipedia – Over 35 individual business points of contact, over 500 projects, and a myriad of different pages, certifications and portals for supply chains and construction. Like many a construction site – it functioned well, but it wasn´t necessarily optimised for the public.
Here then, was a use case. A simple chatbot that, programmed on company data, would answer questions along the above four categories. An MVP was built by myself, using scraped website data from around 600 key website pages, logic added around search terms via extensive prompting, and an interface built by a low-code builder.
The final MVP result was subsequently handed off to the incumbent BPO to adapt and add to the businesses’ emergent tech stack, where to this day it remains as a bubble across the entire company website, receiving dozens of queries a day.
An significant uptick in traffic to key business generating pages (contact us, case studies in particular) was noted, as was interest from customers. However, perhaps more fundamentally, search queries were logged and assessed in monthly meetings, in fact over time the tool became so effective at identifying projects based on a few keywords (Show me hospitals we´ve built near sheffield) that a third of the traffic came from the company´s own employees. Other key, surprising takeaways, which could potentially be actioned in future:
- Members of the public don´t necessarily know the difference between a residential developer – who would sell homes directly to the public, and a tier one contractor that does not. As such individuals were continually messaging the chatbot, asking for detailed information on house prices and how to buy. Passing this information in a form format to a partner developer could be a good look in public projects
- Likewise members of the public often asked for information about health centre opening hours
- Would be customers often had very specific queries, for instance around the business´ experience with specific sustainable materials. This could be used to shape future marketing campaigns.
