Analysing Sentiment Scores of Mission and Purpose Statements of Dutch Banks
When simple experiments yield surprising results
I enjoy experimenting and innovating with frontier technologies. Over the years, I’ve come up with a bunch of simple and useful ‘micro-innovations’ or ‘life-hacks’. Some examples:
- Automating the surf forecast by programming a Python script that sends me a daily notification (via Telegram chatbot) when/if the surf is going to be pumping (based on my personal preferences per surf spot);
- Programming a Python script that sends me a dinner option (via Telegram chatbot) every day at 4pm, picking from a list of my favourite recipes;
- Performing sentiment analysis (analysing polarity and subjectivity) on the discography of one of my favourite music artists.
Sometimes I stumble upon interesting and/or surprising results while experimenting. Which is what happened last week when I looked into Google’s Natural Language API.
But before diving into the experiment and results, let’s start with some theory.
A brief introduction to Natural Language and sentiment analysis
Google’s Natural Language API uses machine learning to reveal the structure and meaning of text¹.
One of its applications is sentiment analysis, which allows you to generate a sentiment score for a string of text, with scores ranging from -1 (very negative) to 1 (very positive).
Classification of sentiment scores is as follows:
- -1 to -0.25 = negative
- -0.25 to 0.25 = neutral
- 0.25 to 1 = positive
Google’s Natural Language also generates a magnitude score. The magnitude score refers to the strength of the sentiment, regardless of score, ranging from 0 to infinity.
A simple experiment: analysing mission statements of Dutch banks
Google offers a free ‘Try the API”, so I conducted a simple experiment: analysing the sentiment scores of mission statements of major banks in the Netherlands (N=12)— If I couldn’t find a mission statement, I used the purpose statement.
Due to the ‘positive’ nature of mission and purpose statements, I hypothesised that all banks would score in the 0.25 to 1 range (positive sentiment).
List of mission (6) and purpose (6) statements:
- ABN AMRO: Banking for better, for generations to come (purpose)
- Achmea Bank: Everyone should have the financial opportunities to lead a comfortable life, now and in the future (mission)
- Aegon: Help people achieve a lifetime of financial security (mission)
- BNG Bank: Everything we do revolves around making social impact. Instead of maximising profits, our priority is to maximise the social impact of our activities (purpose)
- ING: Empowering people to stay a step ahead in life and in business (purpose)
- NIBC: Create a sustainable franchise for the future, so we can continue to make a difference for our clients by focusing on their most decisive moments in business and in life (purpose)
- NN Bank: We help people care for what matters most to them (purpose)
- NWB Bank: Help our clients create added value for society as a robust and sustainable bank for the public sector (mission)
- Rabobank: Growing a better world together (mission)
- Triodos: Make money work for positive, social, environmental and cultural change (mission)
- Van Lanschot: Preserve and create wealth for our clients in a sustainable way (purpose)
- Volksbank: Banking with a human touch (mission)
After analysing all 12 statements using the Natural Language API I created a bubble plot. Output:
The bubble plot shows that 10 out of 12 statements present positive sentiment scores (>0.25).
Volksbank presents the ‘lowest’ positive score (0.5) and Achmea Bank presents the ‘highest’ positive score (0.9). Most banks’ statements present very positive sentiment scores (>0.7).
Two banks present neutral sentiment scores: Triodos (0) and BNG Bank (0.2).
No banks present negative sentiment scores.
Interestingly, there is no difference between the average sentiment scores for mission statements (0.63) and purpose statements (0.63).
Overall, magnitude scores are equal to sentiment scores: only BNG bank presents a slightly higher magnitude score (0.4) compared to sentiment score (0.2). This is why in the plot, bubble sizes get bigger with a higher sentiment score.
Google’s descibes that ‘the magnitude of a document’s sentiment is often proportional to the length of the document³’. Mission and purpose statements are typically short, which possibly explains why magnitude scores are pretty low.
I hypothesised that all banks would score in the 0.25 to 1 range (positive sentiment). And most banks did. Except for two: BNG Bank and Triodos.
What’s most interesting here is that Triodos’ mission statement does not present a positive sentiment (score of 0). As neutral as they come!
If there’s one bank in the Netherlands that I’d expect to deliver a powerful, positive mission statement it would be Triodos.
They are widely recognised as a ‘purpose-driven’ company and listed as the most sustainable Dutch bank in the Sustainability Index 2021.
So maybe it’s time for Triodos to align their ‘neutral’ mission statement (make money work for positive, social, environmental and cultural change) with their purpose-driven corporate image and sustainable operations.
Which is kind of funny, because usually it’s the other way around: we want companies (and banks in particular) to actually deliver on their ‘greenwashing’ mission and purpose statements.
In Triodos’ case, a more powerful message seems justified — no need for ‘greenblushing’ (walking the walk, but being too shy or lacking confidence to talk the talk²). Spread the word!
Frontier technologies such as Google’s Natural Language provide insights that enable companies to stand out and deliver a powerful message.
You might consider mission and purpose statements as tiny elements in marketing a purpose-driven company, but the power and weight of words should not be underestimated.
Keep in mind that tiny elements add up and that a (marketing) chain is as strong as its weakest link. Consistency is key.
In the end, most people probably don’t decide which bank they will put their money into based (solely) on a mission or purpose statement.
Likely, factors such as financial incentives, products, services and news coverage play a bigger role (assumption).
But for companies looking to deliver a consistent marketing message across the board, sentiment analysis and other NLP-techniques should not be overlooked.