Key priorities driving success for Data Science and AI into 2021
A point of view on key dimensions driving Data and AI into 2021
While we enter into the decade of 2021–2030, the Data and AI is expected to observe maturity more in depth than what we have noticed earlier. The priorities for enterprises and end users are also becoming diverse and focused.
Based on my experience over last few years and readings on various references, the following is an attempt to capture narrative around key priorities in next few quarters.
- Relevancy
- Deep dive / Specialization
- Problem Identification
- “X Analytics”
- Core Machine Learning Focus and Storytelling
- Responsible / Ethical AI
- Augmented Effect
- Relevancy:
First and foremost point is “Data relevancy” and the other one is “relevance of Data Science and AI by roles”.
While it is no brainer that “Data relevancy” is important to the success of analytics, AI and any downstream insights generation process, it has always been a challenge for many and focus has been increasing around this. We should look at relevant data by industries, by processes, by need and so on.
a) To measure success or end outcome effectively, we must be cognizant about input data sources and the “data we care most” for consideration.
b) Now that most of us are into “cloud first approach”, “cloud initiative as priority” etc., it is important to note how we are managing “needed data” in cloud space for analysis purposes so that we use to best of our business focus. Huge cost saving factor for this aspect.
c) Are we asking right questions to stakeholders, partners, SMEs so that we can ensure “data relevancy understanding is met” across the ecosystem that we operate?
d) “Right data” is obviously the mantra compared to “More data”. While “more data” may be helpful, it has to be validated before arriving at that conclusion.
When there are programs / capsule courses or equivalent upskill programs being drafted, it should be carefully designed to address relevant audience and roles for upskilling or educating in data and AI space. For example: “Data Science for Managers” type of certifications / courses are going to be different w.r.t. “Data Science for Leaders”, “Data Science for CXOs” etc. Roles such as CEO, COO, CMO, CDO, CAO, CIO, Business Unit Heads, Sales Heads, product Owners, Data Science Leaders etc should look at it in “different lens” and it is driven by specific objective in mind.
2. Deep Dive / Specialization:
Going into depth and have a focus on at least 1–2 techniques to start with, helps. Progressively, multiple techniques can be specialized for greater control based on execution experience. (e.g. these could be Regression, Classification, Clustering, Forecasting, Recommendation, Association, Neural Networks, Natural Language Processing, Reinforcement Learning etc.) The importance of “Specialization” against “Generalization” is nothing new and it attracts how practically things are approached. Understanding the science behind AI/ML, mathematics behind computations and the metrics to measure success is critical. Hence we need to spend time to do that diligently.
Specializing various ingredients of Data Science like any other field, will take time. It is more experimental and we are observing this during last few years from research based methods, applied AI approaches and so on. We can not just “spray and pray” that our method or solution is optimal unless we prove it with comparable solutions and practical evidences.
3. Problem Identification:
Identification of problems is always critical. When we deal with innovations / patents, we focus on the “problem” part and not the solution. When we understand the business need, formulate a problem statement and how it may benefit or create impact, at that time we are in control of “what” aspect of it. The “how” part or solution approaches will come later.
If we look at the “Business Understanding” and “Data Understanding” phases of CRISP-DM framework, we realize the importance of how problem statements, problem formulation, converting business problems into data problems and then understanding it, are becoming more and more critical for Data Science success.
This is where we determine what approaches or methods to solve. For example, in Data science, the problem categories could be anything as follows (not exhaustive):
- Identify patterns
- Determine classes or categories
- Predict outcomes
- Display correlations
- Detect anomalies
- Provide recommendations
It is always better to understand the business objective from the context of the problem identification and ensuring it adheres to SMART-DS(Specific, Measurable, Attainable, Relevant, Timely — Data Science) type of a goal.
4. “X Analytics”:
Business function wise theme driven analytics (e.g. Customer Analytics, Risk Analytics, Marketing Analytics, Retail Analytics, Healthcare Analytics, Operations Analytics etc.) will continue to be more impactful as we are noticing from many predictions from analyst forums and practitioners in the sector. There are just high level examples, these potentially go much deeper based on a particular business theme and/or use case theme that we focus around.
5. Core Machine Learning Focus and Storytelling:
Data Visualization focus and getting into “insights” that are quicker, faster, helps augment decision taking in regular AS-IS tasks are going to be critical. There are key aspects to check from data visualization perspective.
Please refer to my article here on “essence and principles of data visualization in storytelling journey”. Data preparation, Feature engineering, Feature selection, Different aspects of Core machine learning from supervised, unsupervised learning standpoint will continue to demonstrate value.
6. “Responsible AI / Ethical AI”:
Responsible / Ethical AI is getting mainstream. We can deep dive into various aspects to see how they are impacting or will impact.
Dimension1 (Risk in deployment):
- Bias
- Fairness
- Unethical aspects or unfair usage
Dimension2 (Regulatory aspects):
- Region specific needs (e.g. GDPR etc.)
- Laws and regulation protocol may be required to address bias
Dimension3 (Clarity as much as possible):
- What is happening from Step 1 to Step N
- Features used during the Feature Engineering process
- Any information related to feature importance or “top N features”
Dimension4 (How to approach Bias in Data and AI):
- Gather more diverse datasets
- Explore to include labels from a wider range of judges
- Monitor output of models, experiments, algorithms
- Focus on small categories and edge cases
If we refer to the Ethical Institute, then we can see 8 machine learning principles driving this. These are as follows: Human augmentation, Bias evaluation, Explainability by justification, Reproducible operations, Displacement strategy, Practical accuracy, Trust by privacy, Data risk awareness etc.
7. Augmented effect:
Augmenting Data Science / AI with other emerging technologies is expected to bring more value (e.g. AI with Cloud, AI with IoT, AI with AR/VR, AI with Blockchain, AI with Cyber Security, AI with 5G etc.).
As per various study and multiple analysis, almost 1/3rd of CXOs are concerned about the speed in which technology is changing today and hence they want to stay ahead in the game to keep pace. “Fostering innovation” and “execution with smart approach” are critical for success and gaining competitive advantage. When we augment emerging technologies with AI, it has an effect on all parts of the business including IT.
Apart from these, cloud focus is emerging and trying to impact every sphere. Applicability of AI/ML/DL solutions on cloud and cloud native will continue to intensify. Additionally, MLOps is becoming very important for success from a model monitoring, model management perspective and also from managing model drift/data drift standpoint.
Disclaimer: The postings here are personal point of views from my experiences, thoughts, readings from various sources and don’t necessarily represent any firm’s positions, strategies or opinions.