If your data is a paper record, filed in a box, and stored in a cupboard, cabinet or archive, transform with AI will be difficult
AI is everywhere
Every business and organisation are buzzing on the AI bandwagon. And every service provider is talking up the latest AI new releases and integrations. When it comes to AI, learning models need quality data for training to be effective. It’s simple. Clean, accurate data is crucial for reliable AI outputs. AI with a bad dataset means inaccurate, misleading, and biased results. Complex decisions will require broader and wider data to deliver for businesses.
Generative AI works by using models, which are algorithms that learn the patterns and structures of a large dataset. Once they’ve learned these patterns, they can use that knowledge to produce entirely new examples that resemble the data they were trained on.
If this is your data, then AI is difficult
AI Scenarios for Finance and HR
Here are two scenarios on how generative AI trained on records could be used, the benefits and considerations:
HR Records + AI
Scenario
A company wants to streamline the performance review process for managers and ensure consistent feedback across the organisation.
Training Data
The generative AI would be trained on a large dataset of employee records, including:
- Past performance reviews with manager and employee feedback
- Employee goals and achievements
- Performance metrics relevant to different roles
- Examples of strong and weak feedback phrasing
Generation
The AI would analyse the data to identify patterns in effective performance reviews. It could then generate:
- Templates: Tailored to different job roles and experience levels, outlining key areas for feedback and providing prompts for specific examples.
- Suggested Phrasing: Examples of clear, concise, and constructive feedback language for different performance levels.
- Automated Reports: Summarising key performance metrics and highlighting areas for improvement.
Benefits
- Reduced Time: Managers save time drafting reviews with pre-built templates and suggested phrasing.
- Improved Consistency: Standardised templates ensure all employees receive fair and consistent feedback.
- Enhanced Quality: Prompts and examples guide managers in providing clear and actionable feedback.
Important Considerations
- Manager Input: While AI can offer guidance, managerial judgement and personalised feedback remain crucial.
- Employee Feedback: The system should incorporate employee feedback mechanisms to ensure fairness and address any biases.
- Explain-ability: Managers should understand the rationale behind the AI’s suggestions for better decision-making.
- Privacy – ensuring that the correct privacy and security for Personal data (PII from a GDPR perspective) are applied to how the AI stores the data and the length of time data is retained.
Financial Records + AI
Scenario
A medium-sized manufacturing company wants to optimise its pricing strategy and maximise profitability.
Training Data
The generative AI would be trained on a comprehensive dataset of the company’s financial records, including:
- Historical sales data (volume, price, product mix)
- Production costs (materials, labour, overheads)
- Customer segmentation data (buying habits, price sensitivity)
- Market research reports (competitor pricing, industry trends)
Generation
The AI would analyse this data to identify patterns and relationships between pricing, sales volume, and profitability. It could then generate:
- Price Optimisation Models: Recommending optimal pricing strategies for different products and customer segments based on predicted demand elasticity.
- Scenario Planning: Simulating the impact of various pricing changes on revenue and profit margins.
- Targeted Promotions: Identifying opportunities for targeted promotions and discounts to maximise sales without sacrificing profit.
Benefits
- Increased Profitability: Data-driven pricing strategies can lead to higher profit margins and improved financial performance.
- Competitive Advantage: AI-powered insights can help the company stay ahead of the curve in a competitive market.
- Improved Efficiency: The AI automates time-consuming pricing analysis tasks, freeing up finance and sales teams to focus on strategic initiatives.
Important Considerations
- Data Integration: Ensuring seamless integration between the AI and existing financial systems is crucial for accurate data analysis.
- Human Oversight: Financial analysts should review the AI’s recommendations and incorporate their business knowledge to make final pricing decisions.
- Market Dynamics: The AI should be able to adapt to changing market conditions (e.g., fluctuations in raw material prices) to support optimal pricing strategies.
In both scenarios, if your organisation has been doing performance reviews that just get stored in a personnel file or your historical sales data is stuck in box files in the finance team’s seven-year archive cupboard then there will be zero for AI to be trained on.
Scanning documents is an obvious step in the right direction. Categorising and indexing the digitised documents and their contained data, gives you and AI the baseline.
AI Dataset volumes
A second consideration is how much data, and this goes back to the “Garbage in” of our title. If you’ve high quality data (checked & validated) but only a tiny subset of all your records, then feeding that dataset to AI training could be the same as putting low quality rubbish data into the model.
Consider the HR scenario. If you only use the examples of strong feedback phrasing and don’t put the weak phrasing into a model, then all you’re going to get is a playback of that ‘strong’ message, without the recognition of the weak.
For the financial records, and the customer segmentation data, if you only put the customers who you ‘won’ and not those where you lost deals, then AI will be biased to think you only ever win. It will be unlikely that you’ll get improvement recommendations.
IBM discussed AI and training data bias in an article from 2023 – http://bit.ly/4ez3F2b
So, if your data is currently ‘paper’, how much do you digitise? That depends on what you need to use.
Even if you don’t train your AI tool on every detail of it immediately, you have the digital data ready to expand.
All through this you’ll need to consider how to minimise unintended bias (and as we’ve said more data can help), ensure transparency and protect privacy.
AI outweighs the cost of storage retrieval
Often if you store your records in full archival facilities, you’ll face a high cost to retrieve them for scanning, digitising and converting to useable data.
First, weigh up the costs against the benefits of AI such as smart chatbots with 24/7 availability, automation to improve efficiency, data analysis for decision making and more:
- Reduced operational time
- Greater business insight
- Reduced human error
- Enhanced productivity
- Improved competitiveness
- Better customer service and experiences
With all these factors leading to higher growth, then a return on investment becomes easier to quantify and balance against that retrieval cost.
For actual value, business size, adoption level and type of AI change the return, but studies estimate range from numerous hours of work saved, to thousands of pounds. Here’s some studies from the past couple of years:
https://connect.comptia.org/blog/artificial-intelligence-statistics-facts
https://aithority.com/machine-learning/ai-adoption-could-save-businesses-usd-35000-every-year
What’s clear, if you’re looking to transform your business with AI, you have to start with transforming your paper to data.