Data

Academic area

Nature is gaining investor attention, but markets are still struggling to price its value. New research reveals why investors remain cautious, what must change to unlock opportunities, and why water is emerging as a key area to watch.

4 minute read

Off

Decentralised Autonomous Organisations rethink the role of bosses. But who really has the power?

Ying-Ying Hsieh — Wed, 15 Jan 2025 12:14:02 +0000

Decentralised Autonomous Organisations rethink the role of bosses. But who really has the power?

Ying-Ying Hsieh Wed, 15/01/2025 - 12:14

More blogs

2 November 2026 Knowledge

The corporate sustainability actions that reward investors

15 October 2026 Knowledge

Investing in nature: Why markets haven’t caught up yet

7 October 2026 Knowledge

Could working from home strain your relationship?

9 September 2026

Article at a glance

Intended to democratise organisational governance, new research from Ying-Ying Hsieh, Phanish Puranam and Noam Gelman-Hofstadter suggests power in most DAOs is rather centralised.
Autonomy is often used to mean both automation and independence. DAOs promise decisions executed by code and freedom from external control, but most still rely on human judgement and third-party platforms.
DAOs stand as an important experiment in alternative organisational governance, offering lessons for conventional companies seeking to democratise decision making and decentralise power.

DAOs are a blockchain-based alternative to traditional governance structures. Rather than a conventional hierarchy of executives and managers, DAOs aim to give members a more direct role in decision making, with some decisions executed automatically by smart contracts.

The idea is that decisions are voted on, self-executed, and recorded on a blockchain, making them visible to all members to ensure transparency and accountability at scale. This in turn reduces bureaucracy and administrative overheads, making the organisation more democratic, innovative and agile.

Take MakerDAO as an example. In March 2020, a crash triggered by COVID-19 left the organisation with serious liquidity issues. Rather than executives deciding what to do, a novel solution that surfaced informally in a community channel was debated by risk assessors, smart contract code reviewers and ordinary members, and was then approved through a formal vote.

But has the DAO model really resulted in power being shared more democratically across the board? Our research, the largest systematic empirical examination of DAOs to-date, addresses this question.

Surprisingly centralised

We found that more than 80 per cent of voting rights in the average DAO are controlled by less than 1 per cent of addresses (and multiple addresses may be controlled by single members). That means power is concentrated within a small group, though unlike in a conventional firm, this emerges through token ownership and low turnout rather than deliberate structure.

Another factor is the gap between membership and participation. We observed that the average DAO has 5,024 members, but only 709 voters and 10 proposers, with over half of all proposals receiving fewer than 12 votes. While everybody is entitled to vote, having lots of members does not necessarily mean lots of people have a meaningful say.

“While everybody is entitled to vote, having lots of members does not necessarily mean lots of people have a meaningful say.”

Whales and delegates

One key issue is the existence of ‘whales’ and ‘delegates’, i.e. those who control a large number of voting tokens. Whales hold their tokens outright, and are free to vote however they like, including on their own proposals. Delegates, by contrast, are granted voting power on behalf of other members, but this power is conditional: they can lose their delegate status if they vote against the wishes of those they represent

This distinction matters. Whales pose a greater threat to true decentralisation than delegates, because their unconditional voting power answers to nobody but themselves. And because holdings are visible on the blockchain, the position of whales may be known beforehand, which can make the result of a vote feel like a foregone conclusion. Delegates, despite holding real voting weight, rarely use it unilaterally, as going against their delegators’ wishes risks their position.

Whales can also buy more tokens to sway contested decisions. In many cases, a large share of voting tokens are locked in vesting arrangements held by founders and early investors – in the average DAO, vesting tokens equal 38 per cent of liquid holdings. When these are unlocked, voting power can be further concentrated, exacerbating the issue.

Autonomy and algorithmic governance

There is a question as to how autonomous DAOs really are. ‘Autonomy’ can mean two different things: automation, i.e. decisions executed without human involvement, and independence, i.e. freedom from outside control.

Most DAOs attempt both, but often achieve neither. We found that only around 8 per cent use on-chain governance, where votes are recorded and executed automatically via blockchain. The rest rely on off-chain voting through third-party platforms, with humans viewing results and carrying out decisions – the opposite of blockchain automation, though not completely independent of all oversight.

This offers a wider lesson beyond DAOs. It is tempting to view the trend of businesses increasingly handing decisions to AI tools as a form of self-sufficiency, in which processes run themselves free of human oversight. But if DAOs are any guide, automating decisions does not bring independence – it substitutes reliance on people for reliance on digital infrastructure and the third parties who run it.

What businesses can learn

Ultimately, our evidence suggests that DAOs deviate from the decentralised, autonomous ideal organisation design they were built around. However, this is not to say that they fail as an innovation. Rather, they stand as an important experiment in alternative organisational governance. They are best understood as a distinct organisational form in their own right – something closer to a financialised community or an algorithmic commons than either a corporation or a traditional community.

Conventional companies can also borrow elements of the DAO model, experimenting with giving employees, customers and other stakeholders a more direct voice in decision making. The MakerDAO example shows how this can work in practice – collective decisions settled through proposing and voting – even under urgent conditions that would normally require centralised, top-down control.

This article draws on findings from “Neither community nor corporation? A deeper look at decentralized autonomous organizations (DAOs)” by Dr Ying-Ying Hsieh (ImperiaBusiness School), Professor Phanish Puranam (INSEAD) and Noam Gelman-Hofstadte (DeepDAO Ventures Ltd), published In Industrial and Corporate Change.

Big Tech

Entrepreneurship

Knowledge Categories

Management

Strategy & Leadership

Management & Entrepreneurship

Academic area

Decentralised Autonomous Organisations (DAOs) promised an alternative to traditional hierarchies, but truly distributing power within them has proved more nuanced than it first appeared.

4 minute read

Off

Science’s biggest AI challenge isn’t discovery – it’s proof

Ryan Cory-Wright — Wed, 15 Jan 2025 12:14:02 +0000

Science’s biggest AI challenge isn’t discovery – it’s proof

Ryan Cory-Wright Wed, 15/01/2025 - 12:14

More blogs

2 November 2026 Knowledge

The corporate sustainability actions that reward investors

15 October 2026 Knowledge

Investing in nature: Why markets haven’t caught up yet

7 October 2026 Knowledge

Could working from home strain your relationship?

17 July 2026

Article at a glance

AI can now generate scientific hypotheses faster than traditional research systems can test them
Scientific progress still depends on verification: testing whether promising claims hold up against evidence, theory and domain expertise
AI-driven science will deliver on its promise only if we can test ideas as quickly and rigorously as we generate them

Scientific breakthroughs look sudden only in retrospect: a new theory, a medical advance, a technology that changes how we see the world. But before an idea becomes a discovery, it must survive a slower and less glamorous process of verification.

AI-driven science needs scalable verification methods. This is because AI systems can analyse vast amounts of data, identify patterns and generate new hypotheses faster than ever before. But verifying those hypotheses remains slow, costly, and highly domain-specific.

Ideas need proof

Science begins with a hypothesis: a claim precise enough to be tested, challenged and potentially proven wrong. Verification asks whether that hypothesis survives contact with what we already know by testing it against against theory, evidence, experiments, simulations, or formal reasoning. It is this process that separates genuine discoveries from ideas that merely appear plausible.

Scientific advances such as germ theory, thermodynamics and modern agricultural practices transformed society not simply because scientists proposed new ideas, but because those ideas were rigorously tested and repeatedly validated. Verification, then, is the dividing line between a promising idea and a discovery that science can trust.

The age of abundant ideas

For centuries, generating new scientific ideas was one of the most time-consuming parts of research. Over the past decades, major discoveries have become harder to achieve as scientific problems have grown more complex and interconnected.

This is what makes AI so exciting for science: it changes the speed at which researchers can move from data to possible explanations.

Machine learning systems can scan vast datasets for patterns that humans might miss, while large language models (LLMs) can help connect those patterns to possible explanations, making hypothesis generation dramatically faster.

That changes the economics of discovery. When machines can produce candidate explanations faster than researchers can test them, the bottleneck shifts from generating ideas to verifying and filtering them.

Science’s new bottleneck

In many areas of science, the pace of hypothesis generation has historically been constrained by human expertise, data collection and theory-building. AI weakens that constraint: it can propose many more candidates than existing verification systems, such as peer review, can comfortably absorb.

While the number of hypotheses has increased dramatically, verification remains comparatively slow, expensive and often dependent on human expertise. Laboratory experiments, clinical trials and peer review cannot easily be scaled at the same rate as AI-generated ideas. The result is a growing verification bottleneck.

“Generating ideas is no longer the primary constraint. Determining which ideas stand up to scrutiny is becoming the harder problem.”

Without effective verification, scientific progress risks becoming overwhelmed by plausible yet untested results. Some hypotheses may fit existing data while failing to hold up under further scrutiny. Others may appear convincing but remain disconnected from established scientific knowledge.

In other words, generating ideas is no longer the primary constraint. Determining which ideas stand up to scrutiny is becoming the harder problem.

History shows the cost of inadequate verification. Scientific claims that initially appeared convincing have later been overturned because measurements were flawed, assumptions proved incorrect, or results could not be replicated. Without rigorous verification, minor errors can compound, leading to years of wasted effort and resources.

Verification isn’t one-size-fits-all

Verification is not the same in every scientific field. In physics, it often relies on mathematical consistency and reproducible experiments. In biology, it may require laboratory testing and observation. In medicine, verification can involve clinical trials and statistical evidence.

There is no universal framework that can verify every scientific claim. What unites these disciplines is the principle that scientific knowledge advances only when ideas are tested against evidence. Although the methods differ, verification remains fundamentally driven by structured, iterative reasoning.

Trust is the new competitive advantage

AI is changing scientific discovery and may ultimately require us to rethink aspects of the traditional scientific method.

For much of human history, ideas were scarce and difficult to generate. Today, AI is making ideas abundant. The scarce resource is increasingly becoming the ability to determine which discoveries are robust enough to trust, invest in and build upon.

“In an age of abundant AI-generated ideas, trust becomes the scarce resource, and verification will be how science and business earn it.”

As AI continues to accelerate scientific discovery, the greatest opportunities may lie not in generating ever more hypotheses, but in developing better ways to verify them. That means building verification into AI-driven science itself through:

Automated checks wherever possible
Stronger links to established theory
Better benchmarks
Human oversight where scientific judgement matters

For businesses, universities and investors, the main lesson is not that AI makes scientists less important. It is that AI makes scientific judgement more valuable. As machines make it cheaper to generate hypotheses, the competitive advantage will shift to organisations that can test claims quickly, rigorously and carefully before building products, making investments or changing practice. In an age of abundant AI-generated ideas, trust becomes the scarce resource, and verification will be how science and business earn it.

Discover insights by Dr Cory-Wright

Read 'The need for verification in artificial intelligence-driven scientific discovery'.

Artificial Intelligence

Innovation

Knowledge Categories

Academic area

Analytics & Operations

AI is making scientific ideas abundant. The hard part is knowing which ones to trust.

4 minute read

Off

Ragvir Sabharwal on building strong analytical foundations for the next generation

Shamita Harsh — Wed, 15 Jan 2025 12:14:02 +0000

Ragvir Sabharwal on building strong analytical foundations for the next generation

Shamita Harsh Wed, 15/01/2025 - 12:14

More blogs

2 November 2026 Knowledge

The corporate sustainability actions that reward investors

15 October 2026 Knowledge

Investing in nature: Why markets haven’t caught up yet

7 October 2026 Knowledge

Could working from home strain your relationship?

5 November 2025

As a senior teaching fellow at ImperiaBusiness School, Dr Ragvir Sabharwal teaches probability, statistics, and econometrics to students on the BSc Economics, Finance and Data Science (EFDS) and MSc Business Analytics programmes. He also serves as Head of Year for the EFDS programme, guiding students through one of ��’s first-of-its-kind degree with a uniquely modern curriculum.

From high-dimensional data to classroom impact

Combining methods from statistics and economics, Dr Sabharwal’s research focuses on developing models that help analyse complex, high-dimensional data over time. His doctoral work in the Department of Statistics at the London School of Economics and Political Science (LSE), funded by the Economic & Social Research Council, advanced methods for estimating dynamic factor models and detecting structural instabilities in large datasets.

This work contributes both to the theoretical foundations of econometrics and to practical applications in macroeconomic and financial analysis. For example, his research demonstrates how sudden shifts in economic or financial systems can be detected in real-time, enabling models to quickly recalibrate and forecasts to remain accurate even through periods of macroeconomic instability.

Before moving into academia, Dr Sabharwal worked for several years as a competition economist at a leading economic consultancy, advising on antitrust cases in the UK and Europe. This professional experience grounds his teaching in real-world applications, giving students insight into how econometric and statistical tools are used in practice.

Dr Sabharwal has developed and delivered a wide range of courses in statistics, econometrics, machine learning, and causal inference, both in-person and online. He has taught students at BSc, MSc and PhD levels. His consistently outstanding teaching evaluations and multiple teaching awards highlight his ability to make complex quantitative concepts accessible and engaging for students from diverse educational backgrounds.

What excites him most is shaping how students approach data-driven questions early in their studies: “Helping students build strong analytical foundations right from the outset is incredibly rewarding. Rigorous early training will be invaluable in shaping how they’ll approach data-driven decisions throughout their professional lives.”

An academic journey through economics and statistics

Dr Sabharwal’s path reflects a deep commitment to both economics and statistics. He obtained two MSc degrees, one in Economics and another in Financial Statistics from LSE, before completing his PhD in Statistics there. His doctoral thesis developed new tools for handling high-dimensional time series data, tackling challenges that traditional models cannot easily address.

At ��, Dr Sabharwal combines his research expertise with his passion for teaching, shaping the next generation of data-savvy economists and analysts. By fostering strong foundations in statistics and econometrics, he aims to prepare students not only to succeed academically but also to become thoughtful, rigorous decision-makers in business, policy and beyond.

Read about the new faculty joining the Business School

Explore our latest faculty appointments

Finance and institutional resilience

Economy

Management

Knowledge Categories

Economics & Public Policy

Academic area

Dr Sabharwal’s teaching focuses on equipping students to think rigorously about data and its role in decision-making

Main image: Ragvir Sabharwal

2 minute read

Off

How to use analytics to play darts like a pro

Martin Haugh — Mon, 21 Oct 2024 11:47:42 +0000

How to use analytics to play darts like a pro

Martin Haugh Mon, 21/10/2024 - 11:47

21 October 2024

For the first time, new research from ImperiaBusiness School demonstrates how to calculate the optimal darts strategy.

From expected goals in football to biomechanics in cricket, sport is undergoing an analytics revolution. No discipline is untouched by a newfound focus on statistics, and this includes darts – a game which has undergone a rapid journey from pub activity to global primetime.

In darts, two players alternate taking turns of three successive throws. The goal is to get from 501 points to zero before your opponent, with the final throw hitting a double. To date, however, analysis of darts has focused on the individual getting from 501 to zero in as few turns as possible, which isn’t quite the same thing. Our research is the first to account for this subtle difference by modelling darts as a dynamic zero-sum game.

Strategic play

In other words, when computing a player’s optimal strategy we ought to take the opponent’s score into account. This is particularly important towards the end of a game, when a player could choose to aim for the double bullseye – a region worth 50 points that counts as a double, but is considerably more difficult to hit than other doubles.

Whether choosing this high-risk, high-reward option is optimal depends on the opponent’s score. If the opponent is likely to win next turn, it’s more likely to be worthwhile. Conversely, if the opponent has a high score remaining and is unlikely to win next turn, it might be better to take a safer option and aim for an easier double on the next turn.

By analysing data from professional players, we were able to develop a model to calculate a player’s optimal strategy for this type of situation. To illustrate this, we compared two hypothetical situations. In both situations Player A is on a score of 50 with one throw left in their turn. In the first situation Player B is also on 50 and in the second Player B is on 150.

No discipline is untouched by a newfound focus on statistics

We can calculate Player A’s win-probability as a function of where on the dartboard they target the final dart in their turn. In one situation we assessed, for example, it is much more advantageous for Player A to be aggressive and aim for the bullseye, as it gives a win probability of 46.8 per cent compared with much smaller win probabilities for any other region of the board.

In the second situation we looked at, however, a defensive approach offers a superior win percentage of 78.6 per cent if Player A aims for single 10, as this leaves them with three throws at the easier-to-hit double 20 on the next turn. This is because Player B is far less likely to win on their next turn, so Player A is likely to have another turn and therefore another chance to win.

We found that playing strategically in this way gives the typical professional player an increased chance of victory of between 0.2 per cent and 0.6 per cent in a single game (also known as a "leg"), compared to using a strategy that ignores the opponent’s score. Over a longer match (e.g. best of 35 legs), however, this can be as high as two to three per cent – a significant edge for professional players.

Skill factor

However, this does not take into account variations in skill between different players. In practice, skill levels vary with the target region (e.g. players practise targeting triple 20 more, so can hit it more easily than other triples) and the individual player (e.g. one player might be better at hitting certain regions than others).

As we did not have data for every player–target combination, we applied empirical Bayesian methods to fill the gaps, allowing us to fit a skill model for each player. We then used this to analyse real-world situations.

For example, in a 2018 match between Gerwyn Price and Peter Wright, Wright was on 18 and Price was on 20. It was the beginning of Wright’s turn and the most obvious strategy was to aim for double nine to win the leg. Instead, he aimed (successfully) at single two, then needed to hit double eight with one of his next two throws. He missed both of those, and Price then won the leg.

When computing a player’s optimal strategy we ought to take the opponent’s score into account

Commentators questioned Wright’s decision not to use all three throws to aim at double nine. However, according to our skill model, what he did was indeed the optimal strategy for him, with a win percentage of 71.4 per cent when targeting single two with his first dart versus 70.5 per cent if he targeted double nine.

Interestingly, when we applied the same situation to a different player – Michael Smith – our model recommended the double nine approach, with a win percentage of 77.8 per cent compared to 70.3 per cent. This is because Smith is more adept at hitting double nine than double eight, while the opposite is the case for Wright.

This suggests experienced players are well aware of their own relative strengths and weaknesses. Our skill model effectively makes this instinctive knowledge visible, unlocking one of the key mysteries of the game – how and why the top players play the way they do.

This article draws on findings from "Play Like the Pros? Solving the Game of Darts as a Dynamic Zero-Sum Game and An Empirical Bayes Approach for Estimating Skill Models for Professional Darts Players" by Martin B. Haugh (��) and Chun Wang (Tsinghua University).

Strategy

Academic area

Analytics & Operations

Knowledge Categories

Strategy & Leadership

Off

How to create more effective marketing strategies with data

Stephan Seiler — Thu, 13 Jun 2024 11:05:22 +0000

How to create more effective marketing strategies with data

Stephan Seiler Thu, 13/06/2024 - 11:05

14 June 2024

For a long time, academics and marketers have struggled to directly observe the consumer's "journey". But all that's changing, thanks to the availability of pre-purchase data.

Our understanding of pre-purchase behaviour changed almost overnight with the advent of the internet and the growth of online retailing. Since then, the search processes of consumers have become increasingly observable, from the virtual stores and review sites they visit to the products they browse, all in seconds from the comfort of their own homes.

Search data can even drill down into the detail of the clicks they made and the length of time they spent in making their purchase decision. This has opened the door to a far better understanding of consumer behaviour and targeted marketing strategies based on this data.

Our research takes stock of how pre-purchase data has been used by academics and marketers and highlights opportunities for its future use as the quality of data improves.

Why data matters

There are two important areas of marketing decision-making where search data is particularly useful. Firstly, it tells us a lot more about a consumer’s preferences than just observing the purchase. Most consumers compare at least a few products before making their decision and search data allows us to see specific pairings which are clearly viewed as similar products.

It also allows us to see the key product features that assume the most importance in the purchase decision and other aspects of the process, for example the outlets customers considered and to what extent price or other factors played a part. Search engine data can even show us when consumers visited review sites to aid their decision-making. This is important information for recommendation systems or pricing decisions.

We expect data quality in offline retail to increase soon, allowing managers and researchers to study offline pre-purchase behaviour

Secondly, the new data on search behaviour confirms that consumers search very few products before making a purchase, typically only two or three. This not only allows firms to shape which products consumers discover through marketing activity such as product rankings online or preferential shelf placement in a physical store, but it makes this activity crucial in bringing relevant products to a consumer's attention.

If consumers do less research before their purchase, product visibility assumes greater significance; if a firm's product isn't seen quickly, it may not be seen at all. This type of data allows marketers to put their products in front of those consumers most likely to buy them, and establishes how strongly individual marketing actions impact search behaviour, as well as purchases.

More recently, regulators have also taken an interest in such data and analysis because they worry that retailers will display their own products more prominently on their webpage. Pre-purchase data, and information on how consumers react to changes in product rankings, can show us how detrimental this type of retailer behaviour is to consumers.

Tracking consumer behaviour in physical stores

Off-line retailing presents more of a problem as it's far more challenging to track a customer's search and purchase journey through what could be a number of physical stores, reviewing comparable products, then checking their requirements are met, before making a final purchase. In the future, stores might be able to track consumers using their cell phones, but this does not yet provide accurate enough location data to see the route customers took through a store, where they stopped to browse products or which products caught their attention.

If consumers do less research before their purchase, product visibility assumes greater significance

Eye-tracking technology (both on- and off-line) can also provide important data for marketers, but this can only be done in a research environment where the customer is fully aware that their behaviour is being analysed, which in turn can skew results. However, we expect data quality in offline retail to increase soon, allowing managers and researchers to study offline pre-purchase behaviour in an equally comprehensive way as is currently possible online.

Pre-purchase behaviour has already become a key tool in the retailer's armoury and an important factor in building an effective marketing strategy. In the online world it's likely that this will become even more sophisticated as technology continues to evolve and the proportion of customers shopping online increases.

The same is true in the off-line world, as the tracking data on mobile phones improves – although it will always lag behind. Perhaps those retailers – supermarkets in particular – that straddle both worlds will benefit the most.

This article draws on findings from "Consumer search: What can we learn from pre-purchase data?", by Elisabeth Honka (UCLA), Stephan Seiler (��) and Raluca Ursu (NYU Stern).

Brand

Consumers

Marketing

Monthly newsletter

Receive the latest insights from ImperiaBusiness School

Toxic customers and how to handle them

Slighted customers can do a disproportionate amount of damage to a business – particularly in the age of social media. So, what's the best way to respond to negative reviews?

Academic area

Marketing

Knowledge Categories

Marketing

Off

The Fifth Industrial Revolution and finance: insights from Naveed Sultan

Michael Mills — Fri, 12 Jan 2024 14:01:24 +0000

The Fifth Industrial Revolution and finance: insights from Naveed Sultan

Michael Mills Fri, 12/01/2024 - 14:01

22 January 2024

Naveed Sultan, Professor of Practice at ImperiaBusiness School and former Chair of Citi’s Institutional Clients Group, shares his insights into how data and technology are transforming finance and business, and what challenges and opportunities this presents

With over 30 years of experience in global banking and finance, including as Chair of Citi's Institutional Clients Group, Naveed Sultan has been at the forefront of digital transformation in the financial sector. Now a professor of practice at ImperiaBusiness School, he sat down with us to discuss how technology is changing finance and business.

What opportunities does financial digital transformation offer to businesses?

The world is going through a major transition, which some observers are calling a fifth industrial revolution. Large parts of the economy are being reorganised, new business models, ecosystems and platforms are emerging, and lifestyles are changing.

Digital transformation, as the deployment of a range of technologies to create better economic and societal outcomes, is part of this, with data at its core. For the finance function within a business, if you can get data promptly and accurately through technology, with the right analytics, you can make better decisions and create better outcomes.

In time, these processes can make organisations more productive, efficient and accurate, and provide a greater level of confidence in their underlying activity, particularly from a compliance and risk perspective. The greater the confidence organisations have in their core processes and their operating and business models, the greater the confidence they have in doing business with clients.

How is this playing out in the financial sector?

We can split the sector broadly into two camps: new fintech players such as Stripe, Square and Revolut, and established banks. On the consumer side, there has been growing competition between the two, but on the institutional side there’s actually greater collaboration.

Fintech firms need access to banks’ customers, while banks themselves see that fintech solutions can deliver better results for their clients. This has led to incumbents in the sector taking equity positions on emerging fintech players or buying them outright.

This allows banks to offer a better, more innovative and efficient proposition to their clients. And, as fintech firms become more successful, their valuations go up and banks benefit from investing in them.

What should businesses be aware of when planning digital transformation?

Whenever new technologies come up, there is a natural tendency to deploy them in the current operating model or on top of legacy systems, without considering the underlying core processes, or the organisational, cultural and talent implications. Generally, this is unlikely to be successful.

Businesses have to first reorganise how they work, in a holistic way. They need strategic commitment from senior management, the right talent to manage data and deploy new technology, and plans in place to change culture and staff mindsets. Only when these key organisational and cultural requirements are met will a business be ready for technology-led change.

And from a society-wide perspective too, we need to start seeing digital transformation not as a technology issue, but as a policy, strategic, cultural and organisational issue. All economic actors need to operate in alignment for an optimal transition, both economically and societally.

What role are specific technologies such as blockchain and AI playing in financial digital transformation?

At the surface level, they are already increasingly important. For example, blockchain provides smart contracts, improved supply chain management and greater transparency in transactions, while AI offers better risk management, fraud detection, algorithmic trading and personalised customer experiences.

However, there is often an inflection point when technologies work in combination. So, when AI and blockchain are combined, we can see, for example, cross-border tokenisation of securities with smart contracts and enhanced fraud protection – something that combines the technologies as more than the sum of their parts.

Mr Sultan, thank you for your time.

Imperiathe expert

Professor of Practice at ImperiaBusiness School since 2023, Naveed Sultan is Co-Director of the School’s Centre for Financial Technology and co-founded the Centre for Responsible Leadership.

Artificial Intelligence

Cryptocurrency

Finance and institutional resilience

Monthly newsletter

Receive the latest insights from ImperiaBusiness School

Why leaders need soft skills in the age of AI

Emotions are not a weakness, but a strength. Learn how to harness yours to improve your leadership skills.

Centre for Financial Technology

Research Centre

Academic area

Knowledge Categories

Off

How data science can boost your chances of finding a date

Fanyin Zheng — Tue, 05 Dec 2023 14:28:42 +0000

How data science can boost your chances of finding a date

Fanyin Zheng Tue, 05/12/2023 - 14:28

6 December 2023

Online dating can be frustrating and time-consuming, but not anymore. We’ve developed new algorithms that make finding your perfect match easier and more enjoyable. Here’s how we did it.

Finding love is difficult enough without having to fathom how well dating services are doing their job of providing the perfect match. Today, this delicate task is one often left to algorithms – the “secret sauce” of hundreds of dating platforms serving millions of people around the world. But match rates across the online dating sector are often dismal. Luckily, for the many individuals who bemoan their unsuccessful sorties into the dating market, there is good news.

We’ve delved into the inner workings of algorithms used by a leading dating app and overhauled the technology – with happy results. By creating new algorithms more tailored to the individual, we’ve succeeded in increasing the number of matches by close to 30 per cent. This has wider implications for any businesses that rely upon algorithms to match customers in a two-sided market place.

How we cracked the code of dating app algorithms

Online dating is today worth $12 billion worldwide, and it’s one of the most common ways people meet their partners: in the US, some 65 per cent of same sex couples, and 39 per cent of heterosexual couples met online, research from 2018 shows. After lockdowns eased, dating app downloads soared.

But companies guard the finer details of the proprietary technology used to suggest potential dates. On the leading US platform with whom we collaborated, both users must declare they like each other for a match to be made. But most platforms don’t run every potential candidate past every other user – that would simply exhaust them. Instead, they limit the number of profiles a user sees; our partner shows a maximum of nine a day to each of its 800,000 users.

The more matches individuals have achieved in the recent past, the less likely they are to like profiles in the near future

By unpicking data held by our industry partner, we were able to understand the workings of their algorithm. This company uses profiles created by the individuals when they sign up – giving information about their age, gender, height, religion, education, location and more. They can also declare their preferences. When tastes converge, there’s a match. Note this model differs from the decentralised algorithms deployed by apps such as Tinder, which learn from users’ online behaviour.

The three keys to better dating matches

What we’ve done is to rejig the workings of this algorithm to bring in three major changes.

Firstly, we have succeeded in tailoring choices so that they are more personalised and granular. Currently, this platform applies a filter based on the user’s stated preferences and the platform’s broad assumptions – people with a degree will only be interested in others with a similar education, or of a similar height, or ethnicity, or interest in music for instance. But love may be hiding in unexpected places. Our algorithm allows for more nuance and different possibilities.

We’ve succeeded in increasing the number of matches by close to 30 per cent

We’ve also taken into account whether any "like" will be reciprocated. Half a match isn’t enough. Because our algorithm can also account for the recipient’s tastes, it makes a more judicious daily selection, making better use of the limited slots available for potential profiles, and increasing the chances of a match. Remember that individual users only see a limited number of profiles a day.

Our third innovation has been to understand and account for what we call "the history effect", which looks at a user’s recent activity on the app. The more matches that individuals have achieved in the recent past, the less likely they are to like profiles in the near future. What they experienced on the platform in their previous session will affect how they behave in the next, and we adjust for that. This may be down to feeling overwhelmed by the number of potential matches, or indeed there may be other factors at play. The more likes they get, for instance, the more attractive they might feel, and the more picky they may become.

How our algorithm performed in the dating market

Testing our algorithm in the real world is fraught with difficulties. These are real people seeking real relationships and we must tread carefully. We can’t dictate how they will react, and there’s no way to conduct a precise controlled experiment with like-for-like individuals.

But we were able to put the algorithm through its paces in a number of ways: firstly with a simulation using the platform’s data from activity of selected users based in Houston, Austin and Dallas – the three largest cities in Texas – over six months. It must be noted that all our experiments took place among heterosexual daters, who represent more than nine in 10 users in these environments. By simulating how individuals would react by using our algorithm, we were able to show a marked rise in the number of matches – enough to convince the dating platform to try it out in the real world.

Our algorithm allows for more nuance and different possibilities

And there, results were even more pronounced. We were able to test the new algorithm on individuals who used the app over a week in Houston. To gauge its performance, we compared the dating activity in very similar conditions in Dallas and Austin over the same time – and we saw a significant rise in the number of matches between Houston daters. In all, these experiments revealed a rise of at least 27 per cent.

Not only is this great news for those trying their luck in the dating market, it’s good news for companies seeking to improve their search and recommendation options. Our algorithmic experiments are broad enough to allow this innovation to be deployed in other two-sided markets, such as ride hailing apps and online work platforms. More personalised technology, better knowledge of the other party, and more oversight of recent online activity makes for success in dating, and elsewhere.

This article draws on findings from "Improving Match Rates in Dating Markets Through Assortment Optimization" by Ignacio Rios (The University of Texas at Dallas), Daniela Saban (Stanford University) and Fanyin Zheng (��).

Academic area

Analytics & Operations

Knowledge Categories

Expert Q&A: Is AI going to save us or kill us?

Evie Burrows-Taylor — Mon, 12 Jun 2023 13:50:05 +0000

Expert Q&A: Is AI going to save us or kill us?

Evie Burrows-Taylor Mon, 12/06/2023 - 13:50

12 June 2023

World Economic Forum Fellow and Visiting Lecturer at ImperiaBusiness School Dr Catherine Mulligan explores what AI means for society

A new dawn or the beginning of the end of human civilisation? Artificial intelligence (AI) has been called each of these. And in the months since the AI-powered chatbot ChatGPT was released, it’s been impossible to avoid articles about the technology’s significance and likely impact on our everyday lives, as well as the future of the human species.

Here, World Economic Forum Fellow and Visiting Lecturer at ImperiaBusiness School Dr Catherine Mulligan explores what AI means for society.

1. Some have claimed AI will have a positive impact on areas as varied as wildlife conservation, hate speech and clean energy – is this realistic?

This is not the first time we've heard technology is going to save the world: we saw it with mobile networks, broadband, the Internet of Things and blockchain. Like those, AI is a tool that can be used for good, as well as bad.

However, all digital technologies come with something that we are a little bit less prone to talk about, which is unintended consequences. AI can, and will, help us solve all the issues you mentioned, as long as we remain cognizant of the possible negative side effects – and work to solve them.

2. Where could AI make its greatest positive impact?

In an era of climate change, the greatest thing we can do with AI is reduce humanity’s environmental impact. One way that this can be done is if we use it to remove inefficiencies within industries such as the energy industry. I also believe AI can help us design and deliver more efficient solutions for education, research and industrial structures.

One thing I'm working on a lot at the moment is digital measurement, reporting and verification for carbon markets. I'm looking at how we can measure environmental impact with AI, which is essential because the models are very complex.

We're also seeing AI being used in very large-scale supply chains to reduce environmental impact, as well as to improve systems design. Another area where we’ve seen a lot of developments with machine learning, which will rapidly move into AI, is healthcare.

3. Where do you think AI’s greatest negative impact will be seen?

I'm not the sort of person who’s overwhelmed with the negative side: AI won't kill us unless we let it. Having said that, we're already seeing some of the downsides.

An obvious one is the manipulation of the social digital sphere, with things like deepfakes and the amplification of hate speech. Everyone is concerned about AI ethics, but there’s also a critical discussion to be had about whose ethics will be applied. For example, a lot of this is driven by which countries are developing the technologies, while others, who use the technology, don’t have a huge amount of say in how things are developed and implemented.

There’s a lot to consider including how to protect elections, as well as how to educate people who've never seen a world where these things don't exist. We need to think about how to explain what truth is to the next generation – as a result, AI forces us to think a lot about what truth really is.

4. Should people be worried about their jobs?

We won‘t see it immediately but jobs are going to transform rapidly. For example, you can already use ChatGPT very effectively in education, with AI able to provide a very good summary of different subjects such as leadership literature. What it isn’t able to do is build a curriculum that combines multiple disciplines together and applies them into another area such as sustainable development – yet.

So, while there will be a rapid transformation, there’s still plenty of space for humans, who will be able to evolve their careers by bringing several disciplines together effectively to solve problems – known as “boundary spanning”. This is going to challenge us to think about what it means to be human, including redefining our role in the workplace.

5. Do you think businesses are ready for the downsides of AI?

Yes and no. When you talk to senior executives behind closed doors, they all know what's going to happen as a result of AI. The really good ones will admit this to staff and work to retrain them. When I started working with one company, they had about 7,000 staff running certain parts of their system – that’s now down to 25.
I wouldn't say the business world is entirely ready, but it understands what's coming.

6.In your opinion, how should we be preparing for AI?

We could easily start to replace some of parts of the C-Suite with really good, forward-looking AI, so I think there should be honest discussions at an executive level. In the UK, most executives tend to not have digital experience: CFOs often become CEOs. Companies therefore need to get people with excellent digital expertise into their executive suite and allow them to help cope with developments in AI.

Businesses also need to consider what it means for the average employee, which means asking more than just what can be automated. Again, I think the key issue in the face of AI is boundary spanning: executives who can show they’re incorporating AI into part of their company to open up a new market, or enter adjacent markets for example, will be extremely valuable.

7. Overall, do you feel optimistic about the potential of AI?

I'm very optimistic about the future of technology but I'm increasingly worried about human beings applying it without real depth of thought.

We need to pause and think before we execute some of these digital solutions because it’s critical we get this right. The societal and the environmental implications of what we're doing are crucial – not just for companies – but for every single organisation across the world right now.

This article draws on the theme of ImperiaBusiness School’s Social Impact Day 2023 “AI for Society: A Problem or Solution”. Dr Catherine Mulligan was a panelist at the event.

Artificial Intelligence

Knowledge Categories

Off

Why did Google+ fail?

Andre Veiga — Fri, 23 Sep 2022 14:53:12 +0000

Why did Google+ fail?

Andre Veiga Fri, 23/09/2022 - 14:53

3 October 2022

Despite offering an arguably better experience, Google’s social network failed to unseat Facebook. Dr Andre Veiga examines why, and what it tells us about how social media users deal with choice

If you told students the same class was on down the hall, but with free pizza and coffee, they would up sticks and move without being asked twice. But why don’t we show the same mobility when it comes to social networks? Even if there might be a better experience on offer, people tend to stay put.

Consider the example of Google+, the tech giant’s attempt at a social network to rival Facebook. It launched in 2011, only to struggle over the years to acquire users, despite offering, as some argued, a better experience. Google finally consigned it to the corporate graveyard in 2019. This illustrates the advantage of established technology networks with the lion’s share of users: they retain the upper hand. Economists call this “incumbency advantage”.

This incumbency advantage of tech giants unsettles regulators and arguably makes for a poorer experience for individuals. If incumbent platforms know they are unlikely to lose users, they have little incentive to charge low prices and provide high quality.

How big is the incumbency advantage?

Why are individuals so reluctant to leave an established social network, even if there’s something better over the way? Some attribute Facebook’s staying power to Mark Zuckerberg’s talents, the superiority of the product, or better data. We have used a mathematical model to shed additional light on this question; our model allows us to understand how much better a new network must be before users choose to leave the incumbent platform. Or in other words, how much incumbency advantage matters.

One crucial feature of platform markets that we are considering is that people enjoy being where others are (what economists call “network effects”). Individuals enjoy using Facebook because lots of other people also use that network – people want to remain in the same network as each other.

Our model shows how these network effects make it unlikely individuals will leave an incumbent network. Imagine, for instance, a user debating whether she should switch from Facebook to Google+. Perhaps Google+ is better in terms of speed and user interface, but – as one unkind commentator described – the experience is like a tumbleweed blowing through the desert. This means she will delay moving to Google+ and continue using Facebook, knowing she can switch at any time. If all users make the same decision, no one ever switches to Google+.

Users behave like a large group of pedestrians trying to cross a busy road. In dribs and drabs or one at a time, they are unlikely to risk it. They’re far more likely to be successful if they’re inspired to cross together. Everyone ends up staying on the same side of the road, waiting for someone to take the first step. The same will hold true for moving social networks, according to our model. This reluctance to switch is perfectly rational from an individual viewpoint, but it ends up harming the group as a whole.

Guidance for regulators on social media

Surprisingly, our model shows the more opportunities there are to switch platforms, the less likely it becomes that individuals will make the move. Suppose an individual only remembers to check Google+ once a year. If she doesn’t switch at that time, and Google+ becomes extremely popular, she might miss out on the advantages of the new network for an entire year. Suppose now that individual checks every hour. She’ll then be extremely comfortable with delaying a decision since she can change her mind soon. Paradoxically, more frequent chances to migrate can increase incumbency advantage.

Our model yielded another insight: the more opportunity people have to be on several platforms at once (commonly known as “multi-homing”), the more likely they are to migrate across platforms. Multi-homing makes it less costly to take that initial step, because you can be in two or more places at once. Our model highlights the positive impact of multi-homing on competition. Regulators should keep in mind the importance of features such as data portability among platforms, or the ability to conduct business on different platforms.

Interestingly, our theory shows too that, if platforms can commit in advance to limiting numbers who may sign up, this too spurs individuals to join. Fear of missing out may be a strong motivation: there’s a risk they won’t be able to move again in the future if the new platform has reached capacity.

Regulating competition among social media platforms

Previously, authorities have acted mostly on intuition in the absence of an accurate means to assess the impact of proposed mergers and the circumstances that competitors require to thrive. Our model goes some way to providing antitrust regulators with the tools to assess and quantify the advantages held by established social platforms.

We also go some way to assessing just how much better a new platform needs to be to entice users to join. As we’ve seen in the case of Google+, offering only slight improvements in some areas just isn’t enough to win people over, given the large incumbency advantage.

Big tech companies may throw up their arms at the prospect of regulation and argue competition is “only one click away”, but our model begins to show such arguments don’t account for network effects (which lead people to delay moving platform) and reveals the competitive advantage that gives incumbent platforms.

This article draws on findings from "Should I Stay or Should I Go? Migrating Away from an Incumbent Platform" by Gary Biglaiser (University of North Carolina), Jacques Crémer (Toulouse School of Economics) and André Veiga (Imperia London).

Monthly newsletter

Receive the latest insights from ImperiaBusiness School

Restarting the future: why the intangible economy isn’t working and how we can fix it

The failure of financial institutions to keep up with the rise of the intangible economy has led to economic disappointment across the developed world. So, why has this happened, and what do we do now?