McKinsey Explainers

What is tokenization?
Tokenization is the process of creating a digital representation
of a real thing. Tokenization can also be used to protect sensitive
data or to efficiently process large amounts of data.

July 2024
Events of the past few years have made it clear: real thing. In Web3 applications, the token is used
we’re hurtling toward the next era of the internet with on a (typically private) blockchain, which allows the
ever-increasing speed. Several new developments token to be utilized within specific protocols. Tokens
are leading the charge. Web3 is said to offer the can represent assets, including physical assets
potential of a new, decentralized internet, controlled like real estate or art, financial assets like equities or
by participants via blockchains rather than a bonds, intangible assets like intellectual property,
handful of corporations. or even identity and data.

Web3 applications rely on a process called Web3 tokenization can create several types of
tokenization. In this case, tokenization is a digitization tokens. One example from the financial-services
process to make assets more accessible. (AI models industry is stablecoins, a type of cryptocurrency
and new modes of payments also use a process pegged to real-world money designed to be fungible,
called tokenization, both of which have little to do or replicable. Another type of token is an NFT—a
with Web3 tokenization—or each other, for that nonfungible token, meaning a token that is provably
matter. In payments, tokenization is used for scarce and can’t be replicated—which is a digital
cybersecurity and to obfuscate the identity of the proof of ownership people can buy and sell.
payment itself, essentially to prevent fraud. For
a detailed description of tokenization in AI, see As noted earlier, AI also uses a concept called
sidebar, “How does tokenization work in AI?”) tokenization, which is quite different from Web3
tokens (despite their shared name). A large
After a couple false starts, tokenized financial assets language model (LLM) used in an AI application
are moving from pilot to at-scale development. could tokenize the word “cat” and use it to
McKinsey analysis indicates that tokenized market understand relationships between “cat” and other
capitalization could reach around $2 trillion by words. (For a more detailed explanation of what
2030 (excluding cryptocurrencies like Bitcoin and tokenization means in an AI context, see sidebar,
stablecoins like Tether). Specifically, we expect “How does tokenization work in AI?”)
that organizations working with certain asset
classes will be the quickest adopters; these include The benefits of Web3 tokenization for financial
cash and deposits, bonds and exchange-traded institutions include the following:
notes, mutual funds and exchange-traded funds, as
well as loans and securitization. Larry Fink, the — Programmability. Programmability is the ability
chairman and CEO of BlackRock, said in January to embed code in the token and its capacity
2024: “We believe the next step going forward to engage with smart contracts, enabling higher
will be the tokenization of financial assets, and that degrees of automation. (For more on smart
means every stock, every bond … will be on one contracts, read our blockchain Explainer.)
general ledger.”
— Composability. This is the ability to interact with
But before we get specific, let’s get the basics down. other assets and applications on the network.

Learn more about McKinsey’s Financial — Operational efficiency. Web3 tokenization can
Services Practice. help streamline processes, automate
transactions, and more, making operations
more efficient.
How does tokenization work?
In general, tokenization is the process of issuing a These, in turn, can mean increased efficiency,
digital, unique, and anonymous representation of a liquidity, and new revenue opportunities.

What is tokenization? 2
How does tokenization work in AI?

Tokenization in AI is used to break down There are a number of tokenization tokenization algorithm. BPE starts with
data for easier pattern detection. Deep techniques commonly used in LLMs: a vocabulary of characters or words
learning models trained on vast quantities and merges the tokens, which most
of unstructured, unlabeled data are called — Word tokenization splits text into often appear together.
foundation models. Large language models individual words or word-like units, and
(LLMs) are foundation models that are each word becomes a separate token. — Morphological tokenization uses
trained on text. Trained via a process called Word tokenization might struggle with morphemes, which are individual words
fine-tuning, these models can not only contractions or compound words. or parts of words that carry specific
process massive amounts of unstructured meanings or grammatical functions.
text but also learn the relationships between — Character tokenization makes each The word “incompetence,” for example,
sentences, words, or even portions character in text its own separate can be broken down into three
of words. This in turn enables them token. This method works well when morphemes: “in-” (a prefix indicating
to generate natural-language text or dealing with languages that don’t negation), “competent” (the root),
perform summarization or other have clear word boundaries or with and “-ence” (a suffix indicating a state
knowledge-extraction tasks. handwriting recognition. or quality). In morphological tokenization,
each morpheme becomes a token,
Here’s how tokenization makes this — Subword tokenization breaks down which enables LLMs to handle word
possible. When an LLM is fed input text, it less frequently used words into units of variations, understand grammatical
breaks the text down into tokens. Each frequently occurring sequences of structures, and generate linguistically
token is assigned a unique numerical characters. Subword tokens are bigger accurate text.
identifier, which is fed back into the LLM than individual characters but smaller
for processing. The model learns the than entire words. By breaking words The type of tokenization used depends on
relationships between the tokens and into subword tokens, a model can what the model needs to accomplish.
generates responses based on the better handle words that were not Different tokenization methods may also be
patterns it learns. present in the training data. Byte pair combined to achieve the required results.
encoding (BPE) is one subword

What’s an example of tokenization What types of technologies make

in practice? Web3 possible?
Financial-services incumbents like BlackRock, As we’ve seen, Web3 is a new type of internet,
WisdomTree, and Franklin Templeton, as well built on new types of technology. Here are the three
as Web3 natives Ondo Finance, Superstate, and main types:
Maple Finance, are increasingly adopting
tokenized money market funds. In first quarter — Blockchain. A blockchain is a digitally distributed,
2024, these funds surpassed $1 billion in total decentralized ledger that exists across a
value (not much compared with total market size, computer network and facilitates the recording
but a milestone nonetheless). of transactions. As new data are added to a
network, a new block is created and appended
Immutable data on the shared ledger reduces data permanently to the chain. All nodes on the
errors associated with manual reconciliation, while blockchain are then updated to reflect the change.
24/7 instant settlement and composability provide This means the system is not subject to a single
better user experience and new revenue sources. point of control or failure.

What is tokenization? 3
— Smart contracts. Smart contracts are software bonds. Embedding operations such as interest
programs that are automatically executed calculation and coupon payment into the
when specified conditions are met, like terms smart contract of the token would automate
agreed on by a buyer and seller. Smart contracts these functions and require less hands-on
are established in code on a blockchain that human effort.
can’t be altered.
— Democratization of access. By streamlining
— Digital assets and tokens. These are items of operationally intensive manual processes,
value that only exist digitally. They can include servicing smaller investors can become an
cryptocurrencies, stablecoins, central bank economically attractive proposition for
digital currencies (CBDCs), and NFTs. They can financial-services providers. However, before
also include tokenized versions of assets, true democratization of access is realized,
including real things like art or concert tickets. tokenized asset distribution will need to
scale significantly.
As we’ll see, these technologies come together to
support a variety of breakthroughs related — Enhanced transparency powered by
to tokenization. smart contracts. Smart contracts are sets of
instructions coded into tokens issued on a
blockchain that can self-execute under specific
What are the potential benefits of conditions. One example could be a smart
tokenization for financial-services contract for carbon credits, in which blockchain
providers? can provide an immutable and transparent
record of credits, even as they’re traded.
Some industry leaders believe tokenization stands
to transform the structure of financial services and
— Cheaper and more nimble infrastructure.
capital markets because it lets asset holders reap
Blockchains are open source, thus inherently
the benefits of blockchain, such as 24/7 operations
cheaper and easier to iterate than traditional
and data availability. Blockchain also offers faster
financial-services infrastructure.
transaction settlement and a higher degree of
automation (via embedded code that only gets
There’s been hype around digital-asset tokenization
activated if certain conditions are met).
for years, since its introduction back in 2017. But
despite the big predictions, it hasn’t yet caught on in
While yet to be tested at scale, tokenization’s
a meaningful way. We are, though, seeing some
potential benefits include the following:
slow movement: as of mid-2023, US-based fintech
infrastructure firm Broadridge was facilitating
— Faster transaction settlement, fueled by 24/7
more than $1 trillion monthly on its distributed
availability. At present, most financial
ledger platform.
settlements occur two business days after the
trade is executed (or T+2; in theory, this is to
Learn more about McKinsey’s Financial
give each party time to get their documents and
Services Practice.
funds in order). The instant settlements
made possible by tokenization could translate
to significant savings for financial firms in
high-interest-rate environments.
How does a Web3 asset get tokenized?
There are four typical steps involved in
— Operational cost savings, delivered by 24/7 data asset tokenization:
availability and asset programmability. This is
particularly useful for asset classes where 1. Asset sourcing. The first step of tokenization is
servicing or issuing tends to be highly manual figuring out how to tokenize the asset in
and hence error-prone, such as corporate question. Tokenizing a money market fund, for

What is tokenization? 4
 example, will be different from tokenizing a Articles referenced:
Find more content like this on the
carbon credit. This process will require knowing
McKinsey Insights App
whether the asset will be treated as a security — “From ripples to waves: The transformational
or a commodity and which regulatory power of tokenizing assets,” June 20, 2024,
frameworks apply. Anutosh Banerjee, Julian Sevillano, and
Matt Higginson
2. Digital-asset issuance and custody. If the digital
asset has a physical counterpart, the latter — “Tokenization: A digital-asset déjà vu,” August 15,
must be moved to a secure facility that’s neutral 2023, Anutosh Banerjee, Ian De Bode,
to both parties. Then a token, a network, and Matthieu de Vergnes, Matt Higginson, and
Scan • Download • Personalize
compliance functions are selected—coming Julian Sevillano
together to create a digital representation of the
asset on a blockchain. Access to the digital asset — “Tokenizing nontraditional assets: A conversation
is then stored pending distribution. with Ascend Bit’s Brian Clark,” March 17, 2023,
Andrew Roth
3. Distribution and trading. The investor will need
to set up a digital wallet to store the digital asset. — “Web3 beyond the hype,” September 26, 2022,
Depending on the asset, a secondary trading Anutosh Banerjee, Robert Byrne, Ian De Bode,
venue—an alternative to an official exchange and Matt Higginson
that is more loosely regulated—may be created
for the asset. — “How can healthcare unlock the power of
data connectivity?,” December 9, 2021,
4. Asset servicing and data reconciliation. Once Prashanth Reddy
the asset has been distributed to the investor, it
will require ongoing maintenance. This should This article was updated in July 2024; it was
include regulatory, tax, and accounting reporting; originally published in October 2023.
notice of corporate actions; and more.

Learn more about McKinsey’s Financial Services

Practice, and check out Web3-related job
opportunities if you’re interested in working
at McKinsey.

Get to know and directly engage with senior McKinsey experts on tokenization.
Matt Higginson is a partner in McKinsey’s Boston office, and Prashanth Reddy is a senior partner in
the New Jersey office.

Designed by McKinsey Global Publishing

Copyright © 2024 McKinsey & Company. All rights reserved.

What is tokenization? 5