Recently, I had to address this very question. Here’s the brief I received from a client:
We have a profit and loss dashboard in Excel which we update manually each month. We want to optimise the dashboard and then use AI (Copilot) to analyse the historical financial data and make recommendations on how we can improve our financial performance.
I think my client felt they should be using AI in their business to make them more competitive and had been swayed that AI was the answer by the amount of marketing out there. I doubt they are the only organisation in this situation.
Finding out it’s not an AI problem.
Before we could even start looking at whether AI could help them, I really needed to spend a day getting to know their company, their data, and find out what problems they were trying to solve. We had some great, open, conversations about their data, their business, and the systems they use. I’m really glad we did this, because behind the colourful excel dashboard, things were not quite so rosy. I found three, key problems:
- Firstly, it wasn’t going to be easy to optimise or automate anything as-is. The data in their excel wasn’t well-structured, with lots of comments, merged cells and manual calculations. It needed a lot of work.
- Secondly, to start data mining they needed more data. The Excel contained profit/loss information, but nothing else. Their per-product based sales data was available but only via their website’s CMS, and due to a CMS misconfiguration, the CMS was not stripping out the VAT when reporting sales values. Not ideal.
- Lastly, we established that they also conducted marketing initiatives and wanted to track the efficacy of those initiatives and whether it affected their product sales.
The solution
I identified that the source data could be easily re-exported from their financial system, and I started to see an approach that would work.
Phase 1: Fix data sources – the CMS tax issue needed fixing, and historical data that incorporated the error needed to be manually adjusted.
Phase 2: Connect all data sources to power query and ensure they automatically update.
Phase 3: Build the data model – for small businesses I find it easier to build data models with the client, it’s more agile this way and clients often likes to change things. Using power pivot allows me to get useful feedback with which I can work.
The takeaway
I delivered completed phases 1 and 2, and at this point recommended we take a pause before starting to talk about AI. The client was delighted with their ability to now query and manipulate their financial data, leading to some exciting ‘light-bulb’ moments.
As it turned out, this was all they needed. The lack of visibility on key data was the problem that needed solving. For future client discussions, I’m now actively querying any AI-based requirements just to ensure that they’re not masking a more fundamental problem underneath.