## Machine Learning Pipeline

- Data Ingestion
- Data Cleaning & Transformation
- Model Training # Model Selection
- Testing & Validation # Model Selection
- Deployment

## AI Approach

### Start with Business problem and not Data

- Busines Problem
- Definition, value, stakeholders, priority, investment
- Data
- Right data to solve the business problem
- Availability
- Privenance
- Security
- Coverage
- Augmentation
- Annotation
- Model Building
- Feature Extraction
- Hypetrparameters
- Tuning
- Selection
- Benchmarking
- Deploy & Measure
- Business value measurement
- AB Testing
- versioning
- Business Process Integration
- Active Learning & Tuning
- Bias mitigation
- Ground Truth & Success Monitoring
- Version Control

## Project Statement

- Improve conversion and sales as per the example
- Clear distinction of data that is required
- Data specifying similarity of images

### Summarised into

- What problem are we trying to solve?
- How does AI add value?
- What data do we nned?

#### Best Bytes

- Scoping of the Problem, such as the work context diagram and Use case Modeling
- How do we measure success?

## Example

- When search result had a product image which was shown by iteself that had white background that converted more than other images
- This is directly related to what is shown on the search results.

## Using AI in Buisness

- Deploy for targeted use cases [Gartner]
[https://www.kgisl.com/gss/wp-content/uploads/2020/09/KGISL-Gartner_Move-beyond-RPA-to-deliver-Hyperautomation.pdf](https://www.kgisl.com/gss/wp-content/uploads/2020/09/KGISL-Gartner_Move-beyond-RPA-to-deliver-Hyperautomation.pdf)

[https://www.gartner.com/en/newsroom/press-releases/2018-12-18-gartner-survey-reveals-two-thirds-of-organizations-in](https://www.gartner.com/en/newsroom/press-releases/2018-12-18-gartner-survey-reveals-two-thirds-of-organizations-in)

Reason: 5G edge use cases and 5D deployment to several purpose driven solutions

- Business problem before data [PWC]
[https://www.pwc.com/us/en/services/consulting/library/artificial-intelligence-predictions/big-data-roi.html](https://www.pwc.com/us/en/services/consulting/library/artificial-intelligence-predictions/big-data-roi.html)

[https://www.fieldservicenews.com/blog/before-data-analytics-think-problem-to-solve](https://www.fieldservicenews.com/blog/before-data-analytics-think-problem-to-solve)

Reason: The use cases deal with historical data and is predominantly static which makes us address the problem first from a selling perspective rather than use data in the first go. Easy things first, devising a USP, we may be speculative and we do not have the evidence and reasoning to address the problem first using data so that requires an investigation.

- Success depends on the data [IBM_THINK]
[https://www.ibm.com/blogs/policy/bias-in-ai/](https://www.ibm.com/blogs/policy/bias-in-ai/)

[https://mindmajix.com/bpm-tools](https://mindmajix.com/bpm-tools)

Reason: A BPM is associated with data intensive processes. Management of outcome rather than tasks is the key. Data management and maintenance of data are key aspects that enable success.

## Measuring Success

- Using metrics

- Metrics must be Easily measurable

- Directly correlated to business performance

- Predictive of future business outcomes

- Isolated to factors controlled by the group

- Comparable to competitors' metrics

## Metrics Quiz

### Measuring NPS

NPS is a score that is the difference between promoters and detractors, the passives are not taken into consoderation

[https://www.qualtrics.com/uk/experience-management/customer/measure-nps/](https://www.qualtrics.com/uk/experience-management/customer/measure-nps/)

[https://www.qualtrics.com/uk/experience-management/customer/net-promoter-score](https://www.qualtrics.com/uk/experience-management/customer/net-promoter-score)

Using Search by ML to introduce Metrics

## Do you need AI

- An Impactful Business Problem

- Quantify the business value

- Does it have large volume of associated data?

- How much data do you have?

- Does the dataset match the problem?

- Is the dataset complete?

- Is the data annotated correctly for the ML Team?

## Need for AI example

Production systems actively learn from humans

It is best practice to incorporate real humans into training pipelines

## Things to remember

- Start with business value

- Use production data and academic data, matche reality and real world deployment

- Learning is key

## Key roles

Product Owner

- Business case owner
- Bridges frpom stakeholders to team
- Owns maximisation of product value
- Ensures that the team builds the right product

Designer

- Owns human-computer interaction design
- Visual design, information architecture, interaction design
- Useability / accessibility

Software Engineer

- Builds product infrastructure
- Problem solver in software deveopment
- Frontend/backend

Data Engineer

- Builds data infrastructure
- gets model into production
- Ensures entire pipeline can support rapid development
- Model management

Data Scientist

- Builds & selects models
- Guides the team on the state of the art technology
- Structures the problem to achieve the business metrics
- Uses data to answer business questions

Quality Assurance

- Owns quality assurance of the product
- Ensures product release is ready
- Scalability testing
- Functional testing

Development and Operations (DevOps)

- Ensures infrastructure reliability
- manages scalability and performance
- Mitigates security risks
- Ensures development and ML Team can work efficiently

## Project Management

- Business Problem
- Data
- Model Building
- Deploy & Measure
- Active Learning & Tuning

## Executing an ML Project

Scrum

- Scrum master
- Product Owner
- Team
- Product backlog / Sprint Planning Meeting / Sprint Backlog
- Finished Work
- Sprint review
- Sprint retrospective