State of AI
State of AI
State of AI
AI has emerged as a transformative force globally with the potential of revolutionizing industries
and driving tangible impacts across various sectors. However, Latin America still lacks a comprehensive
resource to foster high-level conversations about the state of AI in the region and its potential for growth
and innovation.
In response to this gap, ALLVP is pleased to introduce the State of AI in Latin America report. Our
research team surveyed a diverse set of stakeholders, diligently gathered the latest research in the field,
and conducted in-depth conversations with experts. The result is a panoramic view of AI trends, use
cases, and potential in the region.
As we navigate a rapidly evolving landscape, let this report serve as a vital guide for enthusiasts,
innovators, business leaders and investors to understand and harness the power of AI in Latin America.
2
Agenda
3
ALLVP is on a mission to back formidable founders that solve the
hardest problems in Latin America.
5
01 AI primer.
Artificial Intelligence, Machine Learning, Deep Learning,
what is the difference?
Foundational Model Underlying models that can be fine-tuned for a variety of specific tasks.
Natural Language Subfield of AI that focuses on the interactions between computers and humans through
Processing (NLP) natural language.
Large Language Model Machine learning model that has been trained on a large corpus of text data, which aims to
(LLM) generate human-like text on the input it receives.
Supervised Learning Type of machine learning where an algorithm learns from labeled training data.
Type of machine learning that looks for previously undetected patterns in a data set with no
Unsupervised Learning
pre-existing labels and with a minimum of human supervision.
01
Handcrafted Knowledge
02
The Rise of Statistical
03
Contextual adaptation
(1950s – 1990s) Learning (1990s – 2010s) (2020s and beyond)
Good at reasoning, but no capacity to Good at learning, but struggle with Excellent at learning, reasoning and
learn or generalize. reasoning and adapting. adapting.
Rooted in rule-based systems, Grounded in statistical models, Founded on deep learning, evolving
encoding expert knowledge. learning from vast data sets. through interactions.
Manually programmed systems seek Automated systems drive a shift from Autonomous systems learn and
to imitate human reasoning. imitation to true learning. function with limited supervision.
Score in bar exam Time to reach 100M users GitHub Co-Pilot results
Percentile Months Minutes to complete task and
completion rate
$13B investment
$300M investment
Partnership
Partnership
01 Application layer
Apps with
Applications that integrate AI models into a user-facing
proprietary Apps without proprietary models
product, running their own model pipelines or relying on models
a third-party API.
Model Hubs
02 Model layer
Closed-source
Algorithms and models that process and learn from Proprietary
Foundation
data, and which power AI applications. They can be models Open-source
Models
proprietary or open-source. foundation
models
13
Sources: Leeway Hertz, a16z, Stanford
The race is on: several foundational models have emerged in
recent years and continue to evolve quickly
Language models
Sources: Jaime Sevilla et Al. (2023), Voicebot.ai, Fabian Mosele (2023). Note: Not exhaustive 14
Soaring demand for chips needed to train generative AI
systems helped NVIDIA reach a $1 trillion market cap
NVIDIA produces GPUs, chips with
NVIDIA hit a $1 trillion market cap in May ‘23
parallel processing capabilities and
NVIDIA vs. NASDAQ 5 year stock price evolution. Indexed to 100 in May 2018
high computational power critical to
train generative AI models.
16
AI has a wide range of use-cases across industries and
functions
Value proposition enhancement Operating efficiency optimization
Mobile plan Churn reduction Handling customer Billing Pricing changes Network
Telecom creation campaigns queries management impact deployment
Fraud detection
17
McKinsey estimates that Deep Learning and Generative AI can
create significant value across industries
AI’s value creation potential, by industry
% of industry’s revenues
Source: The AI frontier: Applications and value of deep learning (McKinsey, 2018), The economic potential of generative AI (McKinsey, 2023)
Note: Potential value uses the average between upper range and lower range, by industry and type of AI 18
Latin American users are embracing AI with greater
enthusiasm than the rest of the world
Respondents agreeing with the statement: “products and services using AI have more benefits than drawbacks”
% of people who agree, by country
Source: IPSOS and World Economic Forum – Attitudes towards Artificial Intelligence (2022) 19
“
Nothing in life is to be feared, it is
only to be understood. Now is the
time to understand more, so that we
may fear less.
”
Marie Curie, two-time Nobel Prize winner
Surfing the
AI tsunami
We are facing an AI tsunami, and Latin America needs to surf it. The
pace at which AI technologies are emerging and evolving presents an
unprecedented opportunity for innovation and growth in the region.
Amidst this technological whirlwind, we must not forget the fundamental
principles of business: addressing vital problems and alleviating pain
points for users. In this context, both AI-native and non-AI native
companies must harness AI as a tool to create value by streamlining
productivity and improving the value proposition.
Value creation is critical, as is value capture, and the utilization of the "AI
stack" can serve as a useful conceptual framework for identifying
Gabriel Weintraub, PhD
The Amman Mineral Professor opportunities to capture value. In LatAm, developing the bottom of the
of Operations, Information & stack, the infrastructure and tooling, may be challenging. Instead, most
Technology at Stanford companies will benefit from shifting towards the applications layer,
University potentially building upon foundational models.
21
Surfing the
AI tsunami
In this context, gaining a competitive advantage can be achieved through
proprietary data collection and maintaining control over user interactions.
It might even be worth considering going a layer deeper and developing
a foundational model specifically tailored to unique contexts, surpassing
“
the performance of generic models.
Today, 71% of startups use AI in daily operations, and 100% expect to use
AI in the future.
The main use cases for startups are: content creation (46% of startups),
interacting with users (39%), and personalization at scale (39%).
40% of startups report they are hiring teams to oversee the use of AI.
24
A third of surveyed startups were focused on AI, signaling the
rising interest among entrepreneurs
What industry is your organization focused on?
% of startups
78%
of AI startups are B2B,
highlighting the market
opportunity in helping
other businesses adopt
AI to capture value.
83%
of non-AI startups
believe AI will be
important for their future
success.
54%
of AI startups focus on
one problem at a time,
while
76%
of non-AI startups use AI
to solve 2 or more
problems
Language models
92% penetration¹
8 times higher than the
GPT-2/-3/-4 second most popular
model
Image models
33% penetration¹
2 times higher than the
Midjourney second most popular
model
Source: ALLVP State of AI Survey (2023). Note: (1) Among startups using Generative AI 28
76% of startups have the CEO or CTO leading their AI
strategy, and 40% report hiring teams to oversee the use of AI
Who leads the use of AI within your Are you hiring a team to oversee the use
organization? of AI in the organization?
% of non-AI startups % of non-AI startups
58%
of startups expect
most processes
will be AI-
embedded, up
from 7% today.
Source: OECD. Note: STEM includes 1) Natural sciences, mathematics and statistics, 2) Information and Communication 33
Technologies, and 3) Engineering, manufacturing and construction
Data privacy, regulatory compliance and cybersecurity are the
most common risks mentioned by startups
What are the three biggest risks of deploying AI?
% of non-AI startups
2x
more mentions of data
privacy, despite it not being
a deployment bottleneck
65%
fewer mentions of
organizational reputation
risks, implying they are
more open to experiments
Source: ALLVP State of AI Survey (2023), McKinsey. Note: Options with low mentions include physical safety, national security and political stability 34
“
The people who are crazy enough
to think they can change the
world are the ones who do.
”
Steve Jobs, Apple Co-Founder
Asteroid Continent:
AI in Latin America
Asteroid
Continent: All startups, everywhere, need AI.
AI in Latin I barely have had a chance to understand AI and LLMs. All I know is that
it’s about to change startups and startup building forever.
America Generative AI demonstrated to the average person not just the power
use case of everyday AI, but also the fundamental technology revolution
hidden behind closed doors, which emerged when we finally had the
capacity to compute and process billions of parameters.
“
Future of Smart Human
Fintech
Commerce Cities Capital
“
won't have these strengths, for example platforms, engaging UX, historic
data, or commercial contracts available to run their technologies.
39
The Latin American healthcare system faces challenges due to an aging
population, more prone to chronic diseases, and a shortage of doctors.
• Population aged +65 will double by 2050 (17.5% of total vs. 8% today)
Case study • 30% of population lacks access to healthcare, with most countries
struggling with insufficient doctor to patient ratios
PROSPERiA
democratizes early PROSPERiA prevents blindness via a 4-step process
detection 1. AI-powered questionnaires to identify high-risk patients
2. Screening for visual defects using visual acuity testing and a fundus
of chronic diseases camera
through AI. 3. Images are analyzed by an AI that identifies damage linked to
chronic diseases
4. Cases with suspected ocular pathologies are referred to nearby
ophthalmology clinics
43
Slang helps users learn English for specific business settings. Creating
content in an EdTech has historically been expensive because users need
tailored content with high quality and continuous updates to meet their
specific needs.
Case study Generative AI provided Slang with three capabilities that enable
personalization, reduce R&D and proofreading efforts
Slang leverages AI 1. Producing factually accurate content, even about complex topics
to scale its content 2. Producing linguistically correct content
quickly and As a result, Slang has efficiently scaled its content creation process
44
03 Corporates.
Corporates: key highlights
The main use cases for corporates are interacting with users (74%) and
optimizing sales and marketing (47%).
53% of corporates report hiring teams to oversee the use of AI, and 22%
report hiring teams of 6 or more people.
2.8x
higher adoption of
speech recognition
compared to startups
2.2x
higher adoption of
computer vision
compared to startups
58%
C-level
leadership,
vs. 76% at
startups
Corporates ramping up hiring signals a push to
catch up with startups
39%
of corporates
expect most
processes will be
AI-embedded,
compared to 58%
for startups
2.3x
more mentions of data
privacy or management,
showing more caution
2.1x
more mentions of lack of
commitment, highlighting
culture and alignment
barriers
1.5-2.2x
more mentions of data
privacy, regulatory
compliance and workforce
displacement
Source: ALLVP State of AI Survey (2023), McKinsey. Note: Options with low mentions include physical safety, national security and political stability 54
“
The only way to make sense out
of change is to plunge into it,
move with it, and join the dance.
”
Alan Watts, English author
Foundations to
implement an
AI strategy The advent of AI has created unprecedented opportunities for companies.
Robotic Process Automation and LLMs have either fully eliminated or
significantly reduced the need for humans to be involved in multiple tasks.
Machine Learning algorithms are converting proprietary data sets into a
competitive advantage everywhere from sales and marketing to supply
chain optimization. The question is no longer whether AI can help, it is
figuring out how to implement it.
The first area is data. Companies need to ensure that the data they
collect, structured or unstructured, is cleaned and made available
Julián López-Portillo wherever it is required. Furthermore, data management will be vital in
Co-founder & CEO, Senzai keeping PII and sensitive data secure, especially when integrating with
vendors or foundation models.
56
Foundations to
implement an
AI strategy
The second area is measurement. The impact of any AI implementation
should be clearly measured; this includes not only value metrics, but also
“
other operational and ESG metrics.
Content Creation processes is just a small glimpse of its enormous potential in different
fields. Regardless of the industry, AI tools can write about absolutely any
subject, from finance to health, science, and sports; its capabilities
encompass a world of possibilities.
With its ability to analyze vast amounts of data, generate insights, and
mimic human cognitive processes, AI has revolutionized the way content
is generated, curated, and distributed. Looking towards the future,
several studies predict that up to 90% of internet content will be
produced using artificial intelligence, including not only text but also
videos, images, and audio files.
However, while some view these types of tools with concern, we see
them with great enthusiasm. We believe we are witnessing a paradigm
shift that will have a significant impact on future generations. If we are to
Damian Fraser stay competitive, it is important to become familiar with these tools and
Partner, Miranda Partners engage in the conversation. As mundane tasks disappear, new jobs and
industries will be created.
58
Unlocking Creativity:
The Transformative
Power of AI in
Content Creation
“
In the case of content generation, we believe that AI will unlock human
creativity like never before, rather than replacing content creators, by
If we are to stay automating various aspects of the creative process, such as generating
competitive, it is ideas and creating initial drafts. This will allow content creators to explore
important to become new formats, including hyper-personalization, ultimately enhancing the
familiar with these tools user experience.
and engage in the It will also fundamentally change how the world perceives content, as it
conversation. As becomes a cheaper and more efficient tool to reach target audiences.
mundane tasks Now more than ever, it is essential to be open to new perspectives and
disappear, new jobs and opportunities, to learn and to become proficient in using these new tools,
industries will be which can put us one step ahead. Thanks to the rapid proliferation of
created. new AI-based tools, this is just the beginning of a new and exciting era.
59
After working with corporates, across the globe, the Senzai
founders reached the same conclusion as this report…
so they decided to solve the problems.
Case study company needed to expand its reach and streamline the order-taking
process, while maintaining a personal touch with store owners
Coca-Cola FEMSA
Coca-Cola FEMSA decided the best way to achieve their objective was
digitizes more +1M
through conversational commerce in Whatsapp. It partnered with Yalo to
stores in LatAm implement a solution that enabled
through
conversational
commerce. +1M 80% 7-9 minutes
digitized stores order conversion new average visit
growing quickly rate ordering time as reps are
through Whatsapp focused on strategic
advice
61
Intercorp is one of Peru’s largest corporations, with over 99,000
employees working in 40 companies across retail, financial services,
Case study healthcare and education. It has over US$ 9 billion in revenues, equivalent
to 4% of Peru’s GDP.
Intercorp partners On June 16th, it announced a partnership with Wharton AI & Analytics for
with Wharton to Business, an academic research center at Upenn focused innovative uses
implement the latest of artificial intelligence and data analytics for corporate settings
advances in Artificial
As part of the process, five research teams will be involved in the projects
Intelligence. for a period between 9 and 12 months
“Partnering with Wharton AIAB amplifies our ability to anticipate and fulfill
customer needs, enhancing their experiences across all interactions”
Iván Herrero, Intercorp’s Chief Data Officer
62
OpenAI partnered with Bain & Company, a global management consulting
firm, to bring AI solutions to corporates around the world. The partnership
aims to combine OpenAI’s tools and platforms with Bain’s strategic
expertise and implementation capabilities.
Case study
Two projects have already been announced.
OpenAI announced a
services alliance with
Bain & Company to
help enterprises “Create Real Magic” campaign Hopla: ChatGPT-powered
implement AI chatbot
Invited customers to submit Users request help through chat,
solutions. digital artwork through a and shopping cart is filled based
platform that combined GPT-4 on their budget, constraints or
and DALL-E capabilities menu ideas
63
04 Venture capital.
Venture capital: key highlights
35% of funds are actively helping their portfolio companies deploy AI.
65
Private investment in AI startups has boomed both globally
and in Latin America
Capital invested in AI startups, globally Capital invested in AI startups, Latin America
US$ B US$ M
18%
United
States
38%
Mexico 12%
Other
94%
countries
32%
South
America is actively investing
in Latin American
startups
Sourcing Analyzing
32% 38%
of VCs use AI of VCs use AI
”
Sheryl Sandberg, Meta COO
Evolving Paradigms:
The Role of AI in
Latin American
Venture Capital Latin American investors and operators are well positioned to ride the
new wave of AI momentum in what represents a mostly untapped
regional opportunity with global implications.
Semantix, a Brazilian data provider, raised USD 100m in 2022 – still the
biggest ticket since 2019. In Spanish-speaking Latin America, the
opportunity is equally significant and the market is still nascent. From
Carlos Ramos de la Vega 2019 to Q1 2023, the top 20 investments outside of Brazil amounted to
Director of VC, LAVCA USD75m, with only two companies raising double-digit figures: Zippedi, a
Chilean sales and inventory management platform, and Emi Labs, an
Argentine frontline recruitment automation startup.
72
Evolving Paradigms:
The Role of AI in Players in the region recognize that AI should not be treated as a new
standalone investment category, nor should it be conceived as the
Latin American business panacea. Investors are emphasizing the importance of building
Venture Capital enduring businesses around proprietary models or datasets that can
provide defensibility and a sustainable moat, without relying solely on
strategies based on application layers and cloud-sourced APIs.
“
At the same time, fund managers are aware of the renewed importance
of data privacy and IP protection among the decisive themes for the
The reach of innovation technology to become widely adopted within Latin America. Addressing
will also pose new both will be crucial to building long-lasting trust with end-consumers and
questions for investors as ensure the ethical use of any AI-enabled product, particularly with the
they recognize the impact increasing sophistication of generative AI.
of generative AI will likely The reach of innovation will also pose new questions for investors as
extend beyond their they recognize the impact of generative AI will likely extend beyond their
investment mandate and investment mandate and also challenge their financing structures
also challenge their themselves. The newest tools will enable founders to become much
financing structures more resource efficient, testing current approaches to investment
themselves. benchmarks, portfolio construction and return expectations.
73
EQT Ventures is a multi-stage, sector-agnostic Swedish VC with over
€1.2B in AUM. In 2017, it developed an AI-driven tool to source
companies, Motherbrain, which has since evolved into a co-pilot across
the full investment lifecycle.
Case study
Motherbrain aims to enhance human expertise in each stage:
EQT Ventures created 1. Sourcing: monitor data signals, proxies for growth and proxies for
Motherbrain, a data- momentum to identify hidden gems early on
assistant. from external and internal sources, leveraging the fund’s full
network
3. Value creation: helps with finding talent, M&A targets, benchmarks
and tracking competitors for portfolio companies
74
ALLVP Research Report
Appendix
76
Demographics | Startups