In today’s bulletin, Charlie gives an insight into conducting a BIA interview using AI and discusses the positives and negatives of the outcome.

This week, I have been exploring AI and I wanted to share some thoughts on where I see AI and business continuity going as I learn more about its uses and limitations. I thought I would reflect on what I had learned from conducting BIAs, as well as the insights gained as part of a webinar I gave a couple of months ago on behalf of the BCI AI group entitled ‘AI for BCM – Practical Use Cases with a BIA Case Study’. I wanted to look at three elements of a BIA: whether I could devise a chatbot which would interview you to determine a number of elements of the BIA, whether AI could turn the information into a report which could be sent back to the interviewee to read before conducting the interview, and whether the information could be extracted from a number of interviews and then compiled into an Excel spreadsheet to be used for holding the BIA information. Overall, my use of AI was very successful. I don’t have a huge amount of AI skills, and I am entirely self taught. I have no coding skills and very little patience with IT if it doesn’t work.

I used Lovable and ElevenLabs to build a chatbot which would conduct a BIA interview. Over the years at PlanB Consulting, we have tried to speed up the process of gathering BIA information by using workbooks which are sent out before the BIA interview to gather information which the interviewer could read, speeding up the time spent conducting the interview. For me, it was never a good idea to ask questions where the interviewee may have an opinion, such as what is your RTO or MTPD, but to ask questions which are easy to reply to, such as what does your department do, how many staff do you use, do you work from home or which office do you work from.

I interviewed myself using the chatbot, so perhaps I was cheating a little as I knew what the answer should be, but for me it worked well. At the end of the interview I used the information from the interview transcript to produce a report which gave lots of background information which any interviewer could use as prior reading before interviewing me on some of the finer details of my department. The AI transcript was very easily turned into a format which was easy to read. For the final part, I had three transcripts for three different interviews. Could AI extract the information and format it for use in an Excel spreadsheet, where all the BIA information would be collated? It did this brilliantly and would have saved me from cutting and pasting information. One key lesson was that you have to check the outputs carefully, as AI can hallucinate information when the original interview data is incomplete.

My learning for this is that AI is brilliant at doing individual tasks, saving me lots of time. However, it creates BIAs as a very rigid process, which is how I design BIAs, but this does mean that there is no flexibility in the process which is how others might prefer it. 

The second issue was that in the interview, there was no possibility of answering secondary questions or the chatbot wanting to follow up if the questions were not answered. I got it to say after each answer ‘is there anything to add’, but if you said no, it moved on to the next question. Perhaps with more AI skills, I could have done this and made it ask deeper questions but I am sure it would have taken me days to figure out how to do this.

I have passionate arguments with my fellow consultants on whether people are happy to speak to a chatbot and whether some would even find it easier as the BIA interview could be done in their own time, there would be no social niceties, and the interview could be conducted dispassionately and finished. I think this is a time and space issue and soon we will get used to talking to machines. If some people are having romantic conversations with AI Chatbots, I am sure BIA interviews are not too far behind!

The other issue my fellow consultants have with my AI BIA ideas is that they are able to spot issues, risks, and activities which they couldn’t get with a questionnaire, and they use their experience of identifying issues which a machine wouldn’t. I agree on this, but perhaps the chatbot can gather the basics, giving them more time for the chat where deeper and more complex issues can be explored.

So, what have I concluded? AI is brilliant if you know exactly what you want to achieve and it will save you huge amounts of time, often doing cut and paste or manipulating data to get it into the format you want. This is all personal to you and if you need to make a process that is five steps to get the right result that’s fine. If you want a working model which others can use and work in a number of different ways then I believe you don’t want a home-built AI tool, but one built by a professional who understands how to put the guardrails on the task, so the data is not hallucinated or tasks performed inconsistently. They can also make the user interface and experience work well which doesn’t matter in a home-built tool.

AI is great for people who know what they want and can then get AI to do it. If you are not really sure what you are trying to achieve you are better using a prebuilt AI tool or bot. AI, in this case and lots of others, is going to favour those who have knowledge of the subject they are working in, rather than those who are new to the subject and are using AI to do a task they don’t really understand how to do and what they are trying to achieve.