1. It all starts with a PoC
Equipped with a supportive culture, skilled teams, and a study of data and AI-related risks, we are ready to start with a PoC.
2. Feasibility workshops
Given the experimental nature of AI projects, assessing project feasibility before investing time and money is crucial. Key questions include:
Is training data ready, or must it be collected first?
3. Feasibility workshops
What's the time and money investment to gather data?
4. Data attributes
Are data attributes clearly understood? Are raw attributes usable as available,
5. Data attributes
or do they need processing?
6. Data attributes
For example, is there any “lead time” feature in a supply chain database, or does it need to be derived from the invoice date to the delivery date of orders?
7. Data availability
Can data be sourced in-house, or are external attributes needed?
By now, we have estimated the development and deployment effort at a project level.
It is time to break down tasks into smaller modules, such as the effort required for data collection
8. Data availability
, preprocessing, or
9. Data availability
quality assessment.
10. Evaluating the idea
The feasibility workshop brings together cross-functional stakeholders to evaluate the likelihood of success,
11. Evaluating the idea
address challenges, and find resolutions.
12. PoC is like test-driving a car!
A PoC is like test-driving a car.
Similar to how the test drive provides an experience of the engine's horsepower, seating comfort, and other features before making a purchase,
13. PoC is like test-driving a car!
a PoC helps evaluate an idea's feasibility.
This mini-project can span a few weeks to several months, depending on how complex an idea is.
14. PoC is like test-driving a car!
Guided by the principle of “Try fast, learn faster”, it gives insights into the project's scalability.
It ascertains whether the proposed concept works and has merit that proves business value.
15. Importance of building a PoC
It identifies the scope of improvement to ensure a smoother path to product development, such as:
technical capabilities, regulatory compliance, or alignment with business objectives.
It involves showing the value proposition to stakeholders and soliciting feedback.
16. Time to conclude the PoC
Concluding a PoC involves careful evaluation and decisive actions based on the findings.
It can result in three ways – abandon,
17. Time to conclude the PoC
conditional advancement,
18. Time to conclude the PoC
and yes, it's time to scale.
Let us discuss what efforts are required to convert the "conditional advancement" decision to an absolute "yes".
19. Conditioned on data
Inadequate data quality misleads the learning process of the model and may give sub-par results.
It can be addressed with robust data validation checks.
If the data lacks diversity, it must be resolved by preparing well-represented data for all categories.
20. Conditioned on software and hardware
The AI software requires specific software libraries, frameworks, or dependencies that might be incompatible with the existing technology stack.
For example, existing legacy code in outdated programming languages would require rewriting or modification, which could be complex and time-consuming.
The PoC may also highlight the need to upgrade existing hardware to accommodate specific requirements, such as powerful GPUs, to train neural networks.
21. A clear decision
Let’s see when the business decides to abandon the PoC.
The AI model must meet the fundamental requisites, such as the required performance and accuracy, to be eligible for further scaling.
It could be because the data has no repeatable patterns for the model to learn from.
Put simply, there is no sufficient improvement over existing methods that warrants a change.
AI software is a small component integrated into the existing systems.
If it is incompatible with the organization's existing technology stack, infrastructure, or software, the costs and challenges of making the necessary changes might make scaling impractical.
If we have reached this far and addressed all such issues to conclude a PoC successfully, it is time to procure the funding to scale it further.
22. Let's practice!
As we conclude the PoC phase, let’s leverage its learnings to make informed decisions and design a roadmap for successful AI implementation.
It's time to practice.