This is AI doing – applying artificial intelligence to business problems by Heikki Sassi
Download as PPTX, PDF0 likes119 views
The document outlines a strategic approach to integrating AI into business through a four-phase journey: understanding the problem, conducting experiments, development, and deployment. It discusses the importance of data quality in AI projects, the challenges of defining the right problem, and strategies for implementation, including top-down and bottom-up approaches. Additionally, it highlights the increasing significance of low-code application development in AI and analytics by 2024.
This is AI doing – applying artificial intelligence to business problems by Heikki Sassi
- 1. / Unleash your full potential
- 2. Futuriot 2 BI / Analytics Data / Cloud ERP / CRM Software Machine learning Business intelligence Data storage Data platforms Infrastructure Dynamics 365 Business Central Office 365 Fullstack
- 3. This is AI doing Applying AI to business problems
- 4. Applying AI to business problems is a journey 4 Phase 1 Understanding,Pro blem & Strategy Phase 2 Experiments Phase 3 Development Phase 4 Deployment & Use
- 5. What is the idea behind AI? 5 Machine learning AI Computer program that finds patterns & connections in data. RegressionClassification Clustering = = Decision/ action Decision/ action 1. Predictions from data 2. Insights from data
- 6. How a computer makes predictions from data? 6 Learning algorithm Prediction model Example data (input-output pairs) 1. Computer learns a model from your (historic) example data Prediction model 2. You use the model to make predictions Query with new input data Predicted output
- 7. How the learning actually works? 7 Learning algorithm 1. Model 2. Search process 3. Cost function E.g. linear regression y = ax + b An algorithm that tunes the a and b to minimize the prediction error. E.g. Mean-squared error Example data y = 1722.5x + 72302 Prediction model
- 8. You make an observation – it’s curve fitting! 8 Underfitted model Well generalized model Overfitted model✓
- 9. Finally, deep learning is no different 9 Learning algorithm 1. Model 2. Search process 3. Cost function An algorithm that tunes the network weights to minimize the prediction error. E.g. cross-correlation Example data Prediction model (cat, 89%)
- 10. What problem should I solve? 10 Problem ConceptProblem defined Understand Define Develop Concept The project starts too often from here…
- 11. Is AI the right solution for my problem? 11 Prediction Recommendation Computer vision Natural language processing 2. Problem solving requires…1. Solution cannot be coded 3. There is data Too complex to code Too dynamic to code Good quality Good quantity + +
- 12. What should be my strategy with AI? 12 Top-down Build Top-down Buy Bottom-up Build Bottom-up Buy Build Takes more time Requires new resources Riskier outcome Buy Better time-to-market Better chance of success Top-down CEO driven Strategic High business value Bottom-up Department driven Narrow value …or just wait! 12 3
- 13. What is the process for developing AI? 13 1. Problem 2. Data 3. Model 4. Deployment Business question Business goals Analytics problem Understand the data Collect the data Prepare the data Train the models Test the models Business validation Embed the model Deploy the model Maintain the model Consider the plan for deployment from the beginning Pick a priority problem Reserve time Off-the-shelve algorithms Monitor & update
- 14. That’s it! 14 Phase 1 Understanding & Strategy Phase 2 Experiments Phase 3 Development Phase 4 Deployment & Use Start with the data you have, prioritize value, and bring in experts Requires that your data is in shape Identify the problems you need to solve AI is only a part of the solution
- 15. Finally, application development is changing, so is AI 15 Gartner “by 2024, low- code application development will be responsible for more than 65% of application development activity.” Citizen data scientist Human-centric data analytics and software tools for power users
- 16. Contact for more information 16 Heikki Sassi Analytics futuriot.com Contact heikki.sassi@futuriot.com +358 50 596 9512 https://futuriot.com/case/howden-vakuutustietoa-koneoppimisella/ Check our latest AI case =>
- 18. CONFIDENTIALITY STATEMENT All information contained in this Presentation is confidential, privileged, and only for the information of the intended recipient. Neither this Presentation nor any of the information contained herein may be reproduced, used, published, redistributed, or otherwise disclosed under any circumstances without the prior written consent of Futuriot.