The use cases of machine learning to real world problems keeps growing as ML/AI sees increased adoption across industries. However, there are certain core use cases that add lot of value for organizations and you’ll often find them being implemented in banks, healthcare, manufacturing, product companies or by consulting organizations as well.
Let’s tour of some of the core use cases of ML/AI by domains and verticals.
Marketing and Sales ML Use Cases
Let’s go over main ML use cases in Marketing and Sales function.
1. Predicting Customer Life Time Value (CLTV)
Goal: Determine how much a customer will spend with the company in their lifetime (or given time period). Companies use this information to segment the customers for campaign reachouts, upsell and inbound queries.
This is a popular project with marketing teams of organizations implementing Data science projecs..
The customers past purchase history, demographic data and product information is often used for modeling CLTV.
2. Market Mix Modeling (MMM)
Goal: Quantify the contribution of various marketing spends on sales and help optimize the marketing spend.
This is an important project as well for organizations having a considerable marketing spend. It helps bring visibility and understand the relationship about how the money you spend on various channels is helping product sales.
The inputs you will typically need are the product information, marketing spend information on various channels, both online and offline, product sales by region etc.
The results may be in the form of reports or be shared in a Dashboard since this kind of project will have repeating outputs on regular basis.
Also checkout: The industry grade MMM project course from ML+
3. Demand Forecasting
Goal: Estimate the future demand of the product for a given duration ahead (days, week, months) so that the business planning can happen efficiently. The forecasting can be of other items as well such as the estimated traffic to the website, expected footfalls in a shopping mall.
At the same time there is also the task of forecasting sales instead of Demand. It is quite similar to demand forecasting, but it also takes the conversion / buying into account and tries to forecast the actual sales.
What factors / features to use?
A number of factors may affect the demand / sales forecast such as ad spend, competitor actions, product launches / phase out, Govt actions, weather factors, marketing campaigns, discount given, holidays seasons, spend on advertisements across different channels and regions, macro economic factors such as inflation, CPI, WPI, housing starts, percent of working population in a region, zip code, average income by region, brand opinion from industry standard surveys, raw material or associated products costs.
The results may be shared in the form of reports when the audience is specific concerned individuals from management team or a dashboard when used across multiple teams or has larger audience.
Also checkout: The industry grade Restaurant Vistor Forecasting Project Course from ML+
4. Wallet Share Estimation
Goal: Estimate what percentage of individual spending a given customer buys from your company / brand?
This gives an idea of how important a given customer is for aquisition. For example, what percentage of mobile broadband bill is the person spending on companys network? This type of modeling is common in telecom ad banking sectors.
Potential useful data sources could be associated data with customer demographics, earnings, past purchases and buying pattern, frequency of product usage, lifecycle, exposure to marketing campaigns, tenure with company etc.
The result may be a scored dataset uploaded to a database, which may be accessed by various dashboards / reporting software.
5. Channel Attribution Modeling
Goal: What is the optimal channel to target a customer with certain atributes to maximize chances of conversion. Involves finding the channel that actually influenced or served as entry point for the customer.
6. Churn prediction
Goal: What is the probability that a given customer will attrite from the company (stop buying / switch to competitor)
Requires customer purchase history. The nature, details of the products purchased will help as well.
7. Buyer Propensity Modeling
Goal: Whas it the probability of a customer buying a specific product? Traditionally, this information often helps with marketing campaigns, since we do not want to spam all customers who may not be interested in the product and it risks being perceived as spam.
For this, customer info, past history, usage data is usually helpful.
8. Upgrade propensity Modeling
Goal: Who amongst your existing customers will upgrade to the new product the company is about to launch.
This is quite important because, for companies, it is usually easier to sell to people who are already your customers, than to acquire entirely new customers.
This is especially relevant when a good chunk of the product sales come from your existing customer base, in which case, this will be a critical project for the business.
9. Uplift Modeling
Goal: What is the incremental effect (or decrease) in person towards buying a specific campaign / reachout? That is, companies what to know the increase(or decrease) in probability in persuating a customer to buy for a specific marketing campaign (phone/email/ad etc).
10. Customer Segmentation
Goal: Qualitatively identify different customer groups, so you can give different treatments, both during marketing campaign outreach as well as for customer support when the customer tries to contact you back for sales or service.
11. Product Mix Modeling
Goal: What mix of products gives the lowest churn / makes highest sales. Example: A combined offer of selling laptops with a 5-year service plan lowers churn while bringing in additional revenues.
12. Recommendation Models
Goal: What other product will the customer be interested to buy? Show relevant products to customer, try to increase the browsing/loitering time or similar appropriate objective (like increase conversion / increase returning customer rate etc).
This may be understood as upsell modeling if it involves finding the probability of buying a given product given the customer has already purchased an existing product.
13. Reactivation Likelihood modeling
Goal: What is the probability of winning back a customer given a specific reachout approach(es)?
This is mosty useful for B2C companies who are always on their toes of retaining their loyal customer base, and focuses intensely on winning them back.
For example, people subscribing to a specific mobile network provider might keep switching between providers who provide a better plan. Since the cost of aquisition is substantial, companies want to focus their marketing efforts on customers who are more likely to come back.
This type of project may also be repurposed to winning merchants / service providers. For example, a marketplace like a deals website or companies like Doordash / swiggy / Urbanclap wants to have star service providers and merchants listed on their platform.
14. Adwords Optimization and buying
Goal: What adwords should the company bid for maximum conversions / leads and how much to bid?
Companies spend a good amount of their marketing budget for online ads, like Google adsense, Meta platforms etc. They have to bid for showing their ads to customers, for specific keywords that customers search for.
Because, search driven leads are typically more qualified leads as customers are searching for that product / service. For example, an electronics company want to show their shaving razor to people who are searching for ‘best shaving razor’ instead of a random person.
By finding the right adwords and bidding at the right price, can help companies to sell more.
15. Leads prioritization
Goal: Who are the leads with the highest chance of conversion?
Companies run their business by collecting leads and following up to make the sale. But then, sometimes there might be a large collection of leads and you have a much smaller sales team.
You want your sales team to focus their effort on leads that are more likely to convert. Leads prioritization initiatives can help with that.
16. Pricing Optimization
Goal: What is the right price to sell a product, considering the profitability, impact on brand image etc.
Companies give discounts to generate more sales. But the amount of sales does not increase linearly with the discount given. Sometime, certain products might not sell well when you give discounts. Yes.
Pricing optimzation helps to identify the right / most profitable price a company may sell their product.
This can also be helpful for companies that changes the price of the product with date / hour of the day, like, airline booking, cab aggregators etc. Such scenario is called ‘Dynamic Pricing’ which may very well be tackled using Multi Armed Bandits (reinforcement learning) based approaches.
17. Discount Modeling
Goal: What is the right dicount that should be given to improve sales or maximize profit. It may optionally consider the impact on profitability.
Logistics and Supply Chain
Let’ now look at ML use cases in Logistics and Supply Chain
1. Demand Forecasting
Goal: From a logistics perspective, you are more concerned with estimating the future demand of the components / parts that make up the final product.
This can be more complex than you think when the parts are used / shared across multiple end products and the demand of the end-product is highly fluctuating.
2. Fraud Detection
Goal: Automatically identify various types of frauds. There can be so many use cases.
3. Predicting delayed shipments
Goal: If incoming shipments get delayed, it can impact production as a result, companies may buy from non-regular suppliers at a much higher cost. Predicting delayed shipments can help avoid such overhead.
4. Spare parts forecasting
Goal: The requirement / demand of spare parts are often unpredictable, depending on the nature of the product. Often, this can lead to poor customer experience.
Forecasting the sales of spare parts can help avoid excess cost, run operations better and help improve customer experience.
5. Optimizing safety stock
Goal: How much safety stock to maintain for each SKU to avoid stockout? As the demand fluctuates, there is a higher risk of stockout at the warehouse which can inturn affect production.
6. Route scheduling optimization
Goal: What is the optimal route the fleet should take to minimize cost.
7. Condition monitoring of machinery and on-field products
Goal: Monitor and predict failures of equipment and attend to it before it happens.
8. Predicting time to delivery
Goal: Estimate the time to delivery accurately so the customers know when to receive the goods. This type of project is critical for food delivery apps and to online retailers.
9. Bracket Pricing
Goal: What should be the optimal lot size for purchase of parts. This is impactful, because suppliers sell at lower rates when you purchase in larger lots. But buying is larger lots can pile up inventory, leading to higer costs. So determining an optimal lot size cansave costs.
Banking and Finance
1. Credit card transaction fraud
Goal: Predict if a given transaction is fraudulent or not. This can be challenging since only a small percentage of transactions is fraudulent and flagging genuine transactions often can create a negative customer experience.
2. Loan default prediction
Goal: Predict if a given customer is going to default on a loan product or not, before even handing out the loan. Must ensure such models are non-discriminatory and free from bias.
3. Portfolio Optimization
Goal: What is the optimal allocation of funds against the various investments, given the varying risks and rates of return.
4. ATM cash level optimization
Goal: How much cash to maintain at ATMs. Avoid stock outs while not overstocking currency which would otherwise be earning interest.
5. Credit Risk Assesment
Goal: Assess the potential credit risk of entities / individuals before signing up for a banking product.
Practice dataset: Statlog data
6. Customer Churn Prediction
Goal: Predict which customers are at the risk of churning based on various characteristics and transactional behaviour.
Practice dataset: Churn for bank customers
7. Anomaly detection for Money laundering
Goal: Identify potential instances of money laundering from transactional and customer information
8. KYC verification automation
Goal: Automate KYC verification process for banks so as to be able to scale for large number of customers.
9. Algorithmic trading
Goal: Build profitable algorithmic trading strategies, test and validate it before actual use on field.
Let’s look at ML use cases in HR function
1. Resume Screening
Goal: Shortlist the most relevant resumes for various jobs
2. Employee training recommendation
Goal: Recommend training plans for employees
Goal: Develop chatbots to attend to various queries from employees
4. Churn prediction
Goal: Predict if a given employee is going to leave the job in a given time period.
5. Predicting high performance hires
Goal: Predict the performance level / rating for given potential hire.
Customer Support Use cases
1. Call Center Volume Forecasting
Goal: Forecast the call volumes in order to plan for operator availability
2. Operator scheduling and assignment
Goal: Schedule and assign the suitable operator based on the nature of query and operator skillset so as to minimize the number of touches, faster query resolution, customer satisfaction and smoother management of operations.
3. Reason code prediction
Goal: Classify the reason for the incoming customer query based on the customer characteristics, purchase / browsing history and the content of the query.
4. Wait Time Prediction
Goal: Estimate the amount of time a customer has to wait before a query gets answered / resolved. This can be important for high volume, time sensitive nature of business.
5. Automated customer query response system
Goal: Read, understand intent and respond with appropriate content for various customer queries and interactions.
Health Care Use cases
1. Predicting time to recovery
2. Disease detection from X-Rays and Scans.
3. Drug molecule development
4. Medical record transcription
5. Genomic profiling
6. Diagnostic Biomarking
Computer Vision Applications
1. Image Classification and Object recognition
- Identify food item
- Plant and flower species
- Identify crop disease
- Face recognition
- Cancer detection
- X ray analysis
- Road condition monitoring
2. Object Detection: detect presence of specific object and mark it’s location in picture.
- Identify pedestians
- Detect intruders
- Parking lot occupancy detection
3. Image Segmentation: Identify what object a given pixel belong to and draw boundaries around objects.
4. Object Tracking: Track the location of a moving object in video
- Track position of specific human in a video
- Track movement of wildlife creatures
- Traffic flow analysis
5. Super Resolution: Improve the quality, pixel density of images to make it more clear.
- Restoring old photos
- Improving the resolution
6. Special use cases:
- Blood loss measurement from child birth
Useful Datasets for Computer vision in health care:
- COVID-19 X-Ray dataset
- CT Medical Images Dataset
- Deep Lesions dataset
- MRI scans dataset
- Knee MRI
- 100k Chest X Rays
- CheXpert Radiographs
Natural Language Processing
1. Text classification
- Identify the reason code of customer query email (refund/sales/tech support etc).
- Identify product category given description
- Identify customer sentiment given review
2. Information Retrieval
- Retrieve relevant items for given search query
3. Names Entity Recognition
- Identify person, place, date, companies, phone numbers etc in a given piece of text.
4. Machine Translation: Translate text between languages
5. Parts of Speech Tagging: Identify nouns, verbs, adjectives etc for a given text. This is useful as features when you build text classifiers. Often used as an intermediate step for other algorithms.
6. Speech Recognition: Identify what is spoken and convert to text.
- Speech to text applications
- Trigger word detection
- Speaker identification
- Song identification
- Speech synthesis (Text-to-Speech)