Indian Gen AI Investment report

Explore

Indian Gen AI Investment report

Application Layer

The Application Layer in AI technology stack is the part that users interact with directly. It's like the face of the AI system, where users can make requests and get responses. This layer includes things like websites, apps, and other tools that help users communicate with the AI system.

It's the layer that makes sure user requests are sent to the right AI models and that users get the right information and insights from the AI system. Behind the scenes, system could be interacting with multiple LLMs to deliver the desired outcome.

How to measure Impact of AI applications

Parameter

Metric

Description

People

#of FTEs in industry

Acts as a proxy for numerical impact. More the number of people within a function & use case, more is the potential business impact created by AI

Productivity

#of hours spent per FTE annually FTE Salary or wage per hour

Acts as a proxy for value impact. More number of hours and more value of work of an individual, more is the potential business impact created by AI

Programmability

% of manual work v/s % digitised

Digitization of an individual’s workflow is necessary for AI adoption.

Precision

Margin for error in use case Cost of an error

Use cases having low margin for error, and high cost burden of an error are areas where AI would create maximum impact

There are no rows in this table

⁠

Vertical Apps - Industry specific Apps

These type of AI applications have following features

They are trained on large data sets of the specific industry

Rigid system design to perform specific tasks in a certain manner

High precision and efficiency of output

Factors

Description

Thesis

Underlying technology

Technology in itself is not proprietary

Fine tuned for the specific industry use cases to increase efficiency and accuracy of decision making

Usually no Proprietary IP on tech

Engineering effort

High engineering effort along with industry expertise required

Once built and maximised for accuracy & efficiency, periodic upgrades & low maintenance effort

High engineering effort

Industry expertise is a must

Data Availability

Highly industry specific data

Hard to collect, clean & label

High quality structured data

Data should be Proprietary/unique

Business use cases

Industry dependent market size; but huge opportunity of industry wide vertical integration

Once adopted; difficult to switch

Needs to provide immense value on the precision parameter

Value in precision

If it works; industry wide quick adoption

Capital requirement

Initial cash infusion; can be started with seed of upto 500k to 2 mill

Quick to revenue generation

Investment size in scope

There are no rows in this table

⁠

Horizontal Apps - Function specific Apps

Factors

Description

Thesis

Underlying technology

Technology in itself is not proprietary

Fine tuned for the specific functional use cases to increase efficiency

Usually no Proprietary IP on tech

Engineering effort

High engineering effort along with functional expertise of use case required

Once built and maximised for overall accuracy & efficiency, needs to be updated to unlock every new vertical

Usually limited engineering effort

Functionality expertise is a must

Data availability

Trained on vast data set specific to functionality

Usually fine tuned on customer specific data/inputs to provide maximum value

Hard to collect, clean & label such data.

Functionality specific data needs to be proprietary/unique

Should be able to quickly fine tune on customer specific data

Business use cases

Function dependent market size; but huge opportunity of cross industry, horizontal integration

Easy to switch, usually plug and play integrations

Needs to provide immense value on the People, productivity & programmability parameters

Horizontal pivot/ expansion to different verticals possible

Easy to replicate; Distribution is the key

Capital requirement

Initial cash infusion; can be started with seed of upto 500k to 2 mill

Quick to revenue generation

Investment size in scope

There are no rows in this table

⁠

Emerging Themes

(click on the cards to open detailed view)

Themes

Potential Use cases

Drivers

Challenges

AI feasibility

Software Development

Code visualization

Bug testing

Scale testing

UI/UX testing

Architecture efficiency testing

DevOps

Automate Project management

Rapid prototype development

No-code prompt based development

Tech stack replication (custom logic implementation)

Audience is familiar with AI

Easy integration in existing tools

Structured problems & outputs

Lower costs & TAT for feature testing

Faster push for innovation compressing production timelines

Low token limits reduce the context of overall system architecture resulting in low compatibility

Potential data privacy & security concerns due to large amount of data used to train AI

Skill gap; Next generation of developers needs to upskill to learn to code with AI

High

Consumer Apps

AI apps for Events, Notifications & Messaging

Productivity apps

AI assistants

Health apps (Personalised recommendations, progress tracking, goal setting etc.)

Simplified chat based UI/UX

High value addition by personalisation

Easy to switch from existing apps

Seamless integration with other category apps with AI stack

User Data lock-in with existing apps

High costs of personalisation

Extra token utilization in taking users personal context

Data privacy issues

Possibility of inaccurate information/ hallucination

Some use cases might be a feature for the existing players in the market; with established distribution they’ve got the upper hand

Easy to replicate

Initial user acquisition would be a high cash burn

High

Moderate

Healthcare & Pharma

Clinical decision making (Diagnostics, monitoring, testing)

Automating Physician Workflow

Biotech & Pharma (Drug discovery, Predicting chemical reactions)

Large unstructured datasets, mostly incomprehensible to human brain

High precision tasks; Specialised AI can deliver lower margin of error

Time & cost taking processes; AI minimizes the time frame by huge factor bringing down the cost and manpower required

Alternate method - Trial & error

Lack of high quality structured data sets to train AI models

High integration effort & costs

Skill gap; Healthcare workforce will have to be trained to work with AI

Ethical concerns as AI might not always make most ethical decisions

Data Privacy & compliance issues

Use cases like automating physician workflow might not have a Moat or right to win

High

Moderate

BFSI

Personalisation of products

Personalised terms & offers

Intent prediction

Loan decisioning (underwriting)

Insurance claims processing

Personalised Investment strategy

Compliance management

Large datasets of past decision making & standardised processes make AI output accurate and efficient

High human bias in decision making that can be eliminated by AI

High value addition by personalisation

Huge time & cost savings on processes

Current method - Manual workforce

Data privacy concerns

Human bias in past decision making data might make the AI biased toward certain persona parameter like ethnicity or Income group

High Integration effort & cost

Skill gap; Workforce needs to be trained to work with assistance of AI

Low token limit on foundational models limit the data parameter that can be used for personalisation & decision making

High

Moderate

E-comm, Consumer Goods & Retail

Product Innovation

Real time Insight generation

Enhanced shopping experience

Automated on-scale customer service

Shopper marketing

Real-time intent recognition

Personalised product & offer recommendation

Highly targeted personalised marketing

Demand & Inventory planning

Category management

Generalised AI apps can provide great value since cost of error is moderate

Moderate integration effort with existing infra

Great value in real time action delivery; time is of high importance with rapidly changing consumer preferences

Personalisation increases conversion probability by huge margins, increasing topline

Data privacy concerns

Human bias in past decision making data might make the AI biased toward certain persona parameter like ethnicity or Income group

Lack of structured good quality public data

Skill gap; Workforce needs to be trained to work with assistance of AI

Low token limit on foundational models limit the data parameter that can be used for personalisation & decision making

Easy to replicate

Usually plug & play, so easy to switch

Some use cases might be a feature of existing established player

High

Media & Creator Industry

Video & Image generation

Text to audio generation

Real time audio dubbing tools

Text generation( Social media content creation, Email marketing, Newsletters, Scripts, etc)

Generalised AI apps can provide great value since cost of error is low

Text based interface makes learning curve negligible

Time of output decreases significantly along with effort and cost

Highly creative output

Limited compute power makes output a little costly if not used efficiently

Limited token capacity make it difficult to provide large contexts & references

AI will be a feature addition for existing players; with established distribution they have a right to win

Data privacy concerns arise from the data set used to trained to fine tune the AI

High

Transport & Logistics

Resource planning and optimisation

Predictive maintenance

Traffic detection & Route planning

Supply chain tracking & communication

Vertical integration of data value chain enabling faster more accurate data driven decision

Enhanced passenger & cargo safety

Existing low level AI & ML models are well established & provide significant value

High effort in integration and training

Low

Travel & Tourism

Personalised travel planners

Tour guide Chat bots

Security & efficiency at travel destinations

Enhanced tourist safety

Enhanced tourist experience

AI analyses multiple options & reviews resulting in Cost & time savings for travellers in planning and booking

Low value addition by personalisation

High effort in integration and training at travel destinations

Existing player can provide this as a feature

Lack of autonomous decision making makes value addition similar to research AI apps

Established travel agents/planners apps provide end to end services at similar costs

Low

Automotive

Predictive maintenance

Driver assist

Self driving cars

Automotive designing

Manufacturing operations management

Huge long term savings by AI implementation on assembly line

Enhanced driving experience & safety

Long term cost saving on vehicles via predictive maintenance

Either its value addition is low

OR High effort in integration and training

OR Very high cost of error & tech breakthrough is needed to achieve that precision in decision making in real time.

OR existing AI & ML models are well established & provide good enough value

Moderate

Low

Manufacturing

Predictive maintenance of machinery

Process management & optimization

Quality assurance & testing

Supply chain management

Assembly line automations

High long term cost savings by AI integration on assembly line

Inefficiencies in maintenance of large supply chains

Time taking, costly & low accuracy QA processes

Either its value addition is low

OR High effort in integration and training

OR Cost of tech is very high

OR existing AI & ML models are well established & provide good enough value

Low

Business functions

Human Resource

Profile shortlisting

Workflow management - Scheduling interviews, pipeline management

AI Conducting interviews on scale

KPI tracking, feedback collection etc.

Finance & Accounting

Chat interface for Finance overview

Automated accounting & analytics

Sales Ops

Revenue Operation autopilot

Automated sales pipeline management

Sales analytics

Meeting recording and analytics

Automatic CRM feed

Marketing Ops

Funnel tracking & analytics

Insights for data backed strategy

Highly repetitive and structured processes

High cost saving

Increased efficiency & accuracy of tasks

Decreased dependency on Human or 3rd party

Moderately high costs

Some use cases still need lot of human intervention (hiring)

Data availability to optimise for some use cases (Human resource, Marketing Ops)

Region agnostic, lots of competition from foreign players

Easily replicable

Existing players are launching AI as a feature; with established distribution, they have an edge

High

Moderate

There are no rows in this table

⁠

Conclusion of Investment thesis on AI Application layer

Application layer has the most tremendous potential for Indian AI apps to disrupt existing processes across industries.

These apps usually do not have a tech moat, but are fine tuned for specific need.

Accuracy of their output and overall success is tied to the foundational model they are built on top of.

These are relatively capital intensive to start with and align with the VC’s ROI horizon of 10-12 years.

Region specificity can be observed maximum at this layer.

Vertical Apps

Usually region agnostic. Possibility of ‘Built in India for the Globe’

Data is the major bottleneck; one who solves can bring precision in output and earn the right to win.

Industry expertise in the team is a must.

Very few companies emerge, once market is cracked, they’re there to stay.

Horizontal Apps

Functional expertise of the team are a must; but doesn’t give right to win/moat

Usually region specific due to fine tuned layer of regional data for quality output which can be used across different industries. Ability to cater in regional languages becomes important.

Easy to replicate so product also doesn’t give right to win/moat

Usually high competition because anyone in adjacent industry can switch and become competitor; access to funding becomes crucial in case of price war

Distribution is king + consistent user relevant product innovation gives right to win

Emerging themes conclusion

Emerging themes conclusion

Theme

Conclusion

What to look for?

Investment rating

Software Development

High value addition by AI

Very strong drivers

Challenges are manageable for most of the use cases

High development expertise of the team required (must have)

Early product acceptance by dev community (must have)

Strong GTM (good to have)

Large SAM (Usually true in this category) & SOM (must have)

Consumer Apps

High to Moderate value addition by AI

Moderately Strong drivers

Challenges are also strong

Strong GTM; sustainable distribution (must have)

Functional expertise in the team (good to have)

Large SAM & SOM (must have)

Multiple revenue streams (must have)

Healthcare & Pharma

Very high to moderate value addition by AI

Very Strong drivers

Few Challenges are also strong but solvable

Strong industry expertise in the team (must have)

Proprietary IP/ Unique data (must have)

Large SAM & SOM (usually true in this category)

Either strong product MOAT or High value, limited MOAT (must have)

More focus on vertical apps compared to functional apps like workflow automation (preferred)

BFSI

Very high - to moderate value addition by AI

Very strong drivers

Strong challenges, can be solved with time

Strong industry expertise in the team (must have)

Proprietary IP/ Unique data (must have)

Large SAM & SOM (usually true in this category)

Capability to quickly fine tuning on customer data (must have)

More focus on vertical apps compared to functional apps like personalisation (preferred)

E-comm, Consumer good & Retail

Very High value addition by AI

Very strong drivers

Very strong challenges, difficult to cut through crowd

Strong functional expertise in the team (must have)

Proprietary IP/ Unique data (must have)

Large SAM & SOM (must have)

Capability to quickly fine tune on customer/user data (must have)

More focus on functional apps compared to vertical apps (preferred)

Strong distribution/ GTM (must have)

Fast paced, user insight led innovation (must have)

Media & creator Industry

Very high value addition by AI

Very strong drivers

Strong challenges, difficult to cut through crowd, but vertical use cases might be a good way to go

Strong functional expertise in the team (must have)

Proprietary IP/ Unique data (must have)

Large SAM & SOM (must have)

Capability to quickly fine tune on customer/user data (must have)

More focus on vertical apps compared to horizontal apps like general avatar generation (preferred)

Strong distribution/ GTM (must have)

Fast paced, user insight led innovation (must have)

More focus on region specific use cases due to high foreign competition (preferred)

Transport & Logistics

Low value addition by AI

Weak drivers

Strong challenges; rigid systems; existing tech is good

Strong Industry expertise in the team (must have)

Large SAM & SOM (must have)

Capability to quickly fine tune on customer/user data (must have)

Vertical apps might provide the impact for value creation < Horizontal products can be easy to pivot (preferred)

Travel & tourism

Low value addition by AI, other than some use cases where it’s investment heavy

Weak drivers

Strong challenges

Strong functional expertise in the team (must have)

Large SAM & SOM (must have)

Strong distribution/ GTM (must have)

Ability to easily pivot (preferred)

Automotive

Very high value addition by AI in long term, very low short term addition

Strong drivers

Atleast one very Strong challenge in every use case, mostly innovation or capital heavy

Strong Industry expertise in the team (must have)

Very strong product moat (must have)

Ability to play the long game (return period large)(must have)

Manufacturing

Very high value addition by AI in long term, very low short term addition

Strong drivers

Atleast one very Strong challenge in every use case, mostly innovation or capital heavy

Strong Industry expertise in the team (must have)

Very strong product moat (must have)

Ability to play the long game (return period large)(must have)

Business functions

Very strong value addition by AI

Very Strong drivers

Strong challenges, similar as any B2B SaaS

Strong functional expertise (good to have)

Large TAM, SAM & SOM (must have)

Strong distribution/ GTM (must have)

There are no rows in this table

⁠

Gallery

Want to print your doc?
This is not the way.

Try clicking the ⋯ next to your doc name or using a keyboard shortcut (

CtrlP

) instead.