The double-edged sword of AI-chatbots

Why should an organization use an MSP? According to Google Bard, an experimental Google chatbot that is powered by the LaMDA large language model, there are many reasons why an organization should utilize an MSP. Here are a few of the top benefits:

• Cost savings. An MSP can help you save money by providing you with access to a team of experienced professionals who can manage your IT infrastructure for you. This can free up your in-house IT team to focus on other tasks, and it can also help you avoid the costs associated with hiring and training new staff.
• Increased productivity. By working with an MSP, you can ensure that your IT infrastructure is always up-to-date and running smoothly. This can help you avoid costly downtime and improve the productivity of your employees.

It then goes on in a similar vein to cover improved security, peace of mind, and scalability. It’s all very generalized and is certainly unlikely to make a wavering prospect decide that an MSP is the right solution for them.

Will AI-chatbots replace human resources?

However, such high-level stuff is not where many are placing AI-chatbots. Instead, they are saying that they will replace many first- and second-level support staff, removing the need for the training and maintenance of humans.

But is this right? I doubt it. Let’s start with a general AI-chatbot, such as ChatGPT or Google Bard. These have been trained on large data sets of public information. As such, they are unlikely to know an individual MSP’s platform to any depth.

Therefore, when a customer comes on and asks why something isn’t working, the best response they can expect is something very general, or something that is very wrong.

Worse still would be when a prospect comes on to an MSP’s site and asks a question such as “Which is the best MSP for <service A>?”. Based on public datasets, the answer could well come out as being a competitive MSP – it would be a very bold and brave MSP that would open itself up to such issues.

So, what can be done to make things better? Well, if you can train the AI-chatbot against a targeted data set, then it should give responses that better match what you want the AI-chatbot to give.  However, what value is the AI-chatbot in such circumstances? If a prospect goes to MSP A’s site and gets one response and then goes to MSP B’s site and gets a completely different response where it obvious that each response is tuned to the individual MSP’s business, then there will be little faith in the system.

And finding the required focused dataset to train the AI-chatbot will be no easy feat. Even long-established MSPs that have a trove of helpdesk and sales/marketing data that can be fed into the model will need to check that everything is correct and up to date. When changes happen to the platform, what new data needs to be fed in to update the model? What is the roll-back plan should the AI-chatbot start to mess up and give responses that don’t match the MSP’s needs – if the original staff have been automated out of existence, it’s unlikely that there will be any possibility of an easy fallback position.

Plus, doesn’t all this work that will be required sound just like past attempts to solve the same problem? Areas such as automated FAQs, rules-based response routing, natural language processing (NLP) and so on, have had some levels of success – and all need to have a baseline of data to work against to ensure good levels of suitable response to the customer.

However, such data sets can be highly focused and easily managed. For example, if a customer wants to reset their password, identifying that as the requirement and providing a simple response is (relatively) easy.  Sure, things get more involved where a customer asks something like “Why isn’t the app working?” – but AI-chatbots are unlikely to be any better at answering this than any other system.

MSPs: investigate AI-chatbots, but delay implementing

While it is likely that AI-chatbots will continue to improve rapidly, the current hype over how they will kill off millions of jobs soon is likely to be a bit over the top. While I recommend that MSPs invest time into looking at how AI-chatbots work and where they may fit in to the MSP’s future, I hesitate to recommend implementing them yet.

It is likely that the first real implementations will be like IBM Watson’s AI systems were – highly focused on small subsets of data targeted at specific areas (or, as with Watson, on subsets of a specific area). Watson, in its original form, failed as IBM’s marketeers pushed its supposed capabilities too hard, and those who had created and were evolving the platform could not keep up with meeting expectations.

It is likely that many AI-chatbot systems will suffer the same problems. While they are fine for public use, where a question will tend to be open and a generic answer will not be a problem, commercial concerns trying to use them to drive sales or customer satisfaction may well find them of minimal use at this time.

Photo: PopTika / Shutterstock