The world's most
accurate AI predictions

Mantic is a world-class technical team on a mission to solve the next AI grand challenge: predicting global events with superhuman accuracy, and deploying this capability to power radically improved decision-making in business and government.

Book a demo

Example predictions

From an ongoing tournament against human forecasters

  • Business
    Revolut Banking License
  • Global Affairs
    BRICS Expansion
  • Economics
    UK Jobs Growth
  • Conflict
    Iran Leader Appearance
  • Politics
    Taiwan Recall Vote

Will the restrictions on Revolut's UK banking license be removed before July 26, 2025?

Read Analysis

Will a new country be invited to BRICS at the 17th BRICS Summit?

Read Analysis

On July 17, 2025, what will be the revised change in the UK payroll jobs data for May 2025?

Read Analysis

Will Iran's supreme leader, Ali Khamenei, make a public appearance in Iran before July 14, 2025?

Read Analysis

How many members of the Legislative Yuan will lose recall elections on July 26, 2025?

Read Analysis
Previous
Next

what mantic does

A new kind of foresight

Despite decades of progress in computer modelling, the best predictions in the messy world of human affairs still come from top human forecasters, who use reasoning and judgement.

Mantic is building machines that can predict like an expert superforecaster but with digital speed and scale.

Mantic has an edge on topics where a purely data-driven approach is infeasible or insufficient.

We’ve developed our system to make medium term predictions (1 week - 1 year out) about geopolitics, business, policy, technology, and culture.

RoundPrize Money
Q4 20244th / 40
Q1 20251st / 34
Q2 20254th / 54

In the $120,000 Metaculus AI Benchmark Tournament, we’ve consistently ranked in the top 10% of entrants, winning the top prize money in Q1 2025.

Our latest system establishes a new state of the art, benchmarked on 348 questions from Q2 2025:

0 = competition average
Mantic Metaculus Scaffolding
Benchmark: Metaculus AIBQ2

A generalist model, ready for specialization:

Forecasting is a skill. Scientific research shows that superforecasters, using open source information, can outperform trained intelligence analysts and other domain experts.

Inspired by this finding, we’ve built our prediction engine to tackle questions from any geography or industry, without requiring private data. It dynamically decides what information to find and how to model the problem, like a top human forecaster would.

Working together with clients, we can add further layers of specialisation and assurance. We can validate our prediction engine on the client’s area of focus, and we can onboard new sources of information to ensure the predictions are well-informed.

Our Offerings

We’re ready to serve predictions to clients however is most useful:

Tabular
Predictions

A regular stream of predictions in a structured format

examples:
  • Probability of terror attack in each country globally
  • Probability of CEO departure at each Fortune 500 company
Useful for:
  • Insurance underwriting
  • Supply chain risks

Deep
Forecast

On-demand reports

How it works:

The user specifies a topic, the system researches the relevant issues, formulates forecasting questions, then makes many predictions and writes a report to explain them.

Useful for:
  • Commercial due diligence
  • Strategy consulting

Custom
Dashboards

Track bespoke predictions over time

Key examples:

The user identifies which predictions they want to track, and the system notifies them when the probability goes up or down and why.

Useful for:
  • Trading
  • Government affairs
  • Corporate risk and strategy

Every prediction comes with a rationale: historical analysis, recent developments, and the key arguments.

About us

Mantic is a startup founded in London in 2024, on a mission to solve forecasting and radically improve decision-making.

We raised £3 million in our pre-seed funding round, led by Episode 1, with participation from the US trading firm DRW and a range of angel investors including leading researchers at Google DeepMind and Anthropic.

We are supported by Google Cloud and AWS, delivering the compute we need to train class-defining models.

We aim to match, then exceed, the accuracy of human superforecasters. We will drive adoption of this capability across the private and public sector, unlocking better-informed decision-making.

Our Team

Ben Day

CTO & co-founder

Before Mantic Ben was Head of Research at Foresight Data Machines, developing AI for optimizing steel production. Ben has a PhD in machine learning from the University of Cambridge, where he researched meta-learning and graph neural networks. Alongside this, he was a Research Consultant at Relation Therapeutics on AI for drug development.

Toby Shevlane

CEO & co-founder

Before Mantic Toby spent 2.5 years at Google DeepMind, where he worked as a Senior Research Scientist. There, he co-led a team within the Gemini effort that designed experiments to test Gemini’s dual-use capabilities. Before that, Toby’s PhD research at the University of Oxford was about the governance of LLM release decisions and potential misuse risks from AI.

Matthew Aitchison

Member of technical staff

Previously, Research Engineer at Google DeepMind. PhD from ANU in reinforcement learning.

Scott Jeen

Member of technical staff

Previously, PhD at University of Cambridge on reinforcement learning.

Vlad Bogolin

Member of technical staff

Previously, researcher at University of Oxford; Senior Engineer at MariaDB and FlowX.AI. PhD in AI from the Romanian Academy.

Max Clark

Member of technical staff

Previously, software engineer at Citadel. Highest dissertation mark at University of Cambridge Computer Science undergrad for paper on AlphaGo.

We're hiring

Join us to solve the next AI grand challenge.

We’re assembling a world class team. If you’re energized by hard technical problems with potentially massive impact, we want to hear from you.

Open roles
  • Data & Model Improvement
    AI Engineer
    Camden, London | £100k/yr
  • LLM Agents
    AI Engineer/Researcher
    Camden, London | £100k/yr
  • Product
    Full-Stack Developer
    Camden, London | £100k/yr
  • Operations
    Operations Manager
    Camden, London
Send your resume

Will the restrictions on Revolut's UK banking license be removed before July 26, 2025?

Background

Revolut gained a provisional UK banking licence in July 2024. To become a fully-fledged bank, Revolut must complete a “mobilisation phase” which was expected to take up to 12 months. Until then, Revolut is restricted in the size of customer deposits it can hold, and doesn’t benefit from FSCS protection.

Predictions

Both Mantic and the Metaculus community started around 65% in mid-May. Mantic had researched the examples of Monzo and Starling which both completed successfully in 12 months.

Mantic became increasingly skeptical the full licence would be granted in time, falling to 5% by mid-June, whereas the community remained more optimistic at 52%. Mantic’s skepticism was partly based on the scale and complexity of Revolut, which made it different from e.g. Monzo in 2017. Mantic’s prediction was also influenced by a statement from Revolut’s CFO anticipating approval ‘later this year’.

  • Although challenger banks like Monzo and Atom had their restrictions lifted in under 12 months, I give these precedents less weight. Revolut's greater scale, product complexity, and past compliance issues suggest a more prolonged review by the Prudential Regulation Authority (PRA).
  • Statements from Revolut's CFO in April 2025 anticipate receiving full approval 'later this year,' which points to a timeline beyond the July deadline.

Outcome

No approval. Revolut faced regulatory delays, reportedly owing to the scale of its business.

Will a new country be invited to BRICS at the 17th BRICS Summit?

Background

BRICS is a forum for coordination between major emerging economies, founded by Brazil, Russia, India, and China. The growing influence of the BRICS on the global stage is viewed as in tension with US-led Western power.

Predictions

The Metaculus community hovered around 70% for much of June, motivated by the recent history of BRICS expansion efforts. The community prediction then dropped significantly when a commenter found evidence that Brazil (the host) was not in favour of new invitations.

Mantic’s automated research had already found this out. Mantic also took a longer-term view of the history, which gives a lower rate of new invitations. Moreover, Mantic reasoned that BRICS’s new ‘partner country’ framework would reduce pressure for new invitations. Therefore, Mantic stayed around 20% throughout.

  • Historically, invitations for full BRICS membership are infrequent. Out of 16 summits held since 2009, only two have resulted in such invitations (in 2010 and 2023).
  • The bloc has recently undergone a major expansion, adding five members in 2024 and Indonesia in 2025. Key members, including the host Brazil, have signaled a preference for consolidating these gains before another round of enlargement.
  • The creation of a 'partner country' framework in 2024 provides a formal way to engage with aspirant nations without granting them immediate full membership. This may reduce pressure for new invitations.

Outcome

No new members invited.

On July 17, 2025, what will be the revised change in the UK payroll jobs data for May 2025?

Background

The UK’s Office for National Statistics had released jobs data in June showing 30.17m payrolled employees, a drop of 109k from the month before.

Would the economic data continue falling?

Predictions

The Metaculus community put most weight on another fall, predicting -1.3k jobs as a median, and only a 3% chance of a >20k increase.

Mantic was more uncertain and put greater weight than the community on a large increase (15% chance of a >20k increase).

  • There is considerable uncertainty, with my 50% confidence interval ranging from a downward revision of 6,700 to an upward revision of 14,400.
  • The initial data showed a loss of 109,139 jobs, the largest monthly drop since May 2020. Historically, large negative estimates in this series are often revised upwards as more complete data is processed. ING Bank analysis supports this tendency.
  • The Office for National Statistics (ONS) has flagged the May 2025 estimate as "more uncertain than usual". This warning, combined with recent data quality issues at the ONS, elevates the probability of a significant revision in either direction.
  • Recent revisions provide conflicting signals. The March 2025 figure was revised upward, reducing job losses. Conversely, the April 2025 figure was revised downward, increasing job losses.

Outcome

+132k jobs, above expectations.

Will Iran's supreme leader, Ali Khamenei, make a public appearance in Iran before July 14, 2025?

Background

Iran and Israel were engaged in military conflict from 13-24 June. During this time, Khamenei was hiding out at a secret location. There was ongoing speculation about his status, including his health and whether he had fled the country.

Predictions

Mantic’s predictions fluctuated throughout June with a median of 70%. Mantic’s automated research found that Khamenei usually appears in public 2-4 times per month. It decided there was an outside chance that health issues would prevent a public appearance, but that the most likely outcome was a TV appearance from a secure location in Tehran.

The Metaculus community also fluctuated but was around 20-30% from 20 June onwards. Commenters highlighted the security concerns of a public appearance.

  • Historically, Khamenei appears in public 2-4 times per month. This pattern suggests a high probability of at least one appearance in any 28-day period.
  • The recent Israeli strikes on June 13 have created strong, competing pressures. The regime has a powerful incentive to counter rumors of Khamenei's death by showing him in public.
  • The probability is significantly lower than the historical baseline due to risks from the attack. There is an outside chance he was injured, is too unwell to appear, or that security concerns will prevent any public events.

Outcome

Khamenei appearance (5 July):

Attended a religious event in Tehran.

How many members of the Legislative Yuan will lose recall elections on July 26, 2025?

Background

In the Taiwanese legislature, the KMT was the largest party and broadly more pro-China than the DPP. Under the constitution, voters can trigger elections to recall their legislator. On 26 July 2025, there were recall votes for 25 KMT legislators. If 12 were recalled, the DPP would have a majority.

Predictions

The Metaculus community’s median estimate was that around 7 legislators would lose the recall vote (4-10 at 50% confidence interval). Mantic predicted fewer losses: 4 (2-7) reserving 10% for none.

Mantic cited the high legal threshold for the recall to pass (25% of all registered voters in the district), and the historical rarity of successful recalls (it only found a single case).

  • I estimate that 4 of the 24 legislators will be recalled, with a 50% confidence interval of 2 to 7 successful recalls.
  • The primary barrier to success is a high legal threshold; for a recall to pass, votes in favor must exceed 25% of all registered voters in that legislator’s district.
  • This mass recall campaign is unprecedented in Taiwan’s history. It is driven by civic groups with the full support of the ruling Democratic Progressive Party (DPP).
  • Historically, only one sitting national legislator has ever been successfully recalled in Taiwan, making the outcome of this large-scale effort highly uncertain.

Outcome

0 legistlators recalled.