Explorative AI^®

What if the right option is not on the table yet?

Some decisions have no precedent to draw on. soma Discovery is built for those. Where prediction asks what is likely, Discovery asks the prior question. What is possible.

Powered by the Astut Hypothesis Engine^®. Developed at the University of Oxford's Mathematical Institute, and delivered by soma under a signed reselling agreement.

Before 7 options on the table

Whatever happened to occur to a few people, in a short meeting.

In brief

What it is for

The decision with no precedent, little relevant data, and no settled objective. Prediction has nothing to learn from. Optimisation has nothing to aim at.

The fifth paradigm

What it returns

Speed, and depth of thinking. Work that conventionally runs to months comes back in hours, without a team being assembled to produce it.

The return

Where it comes from

Developed at the University of Oxford's Mathematical Institute by Professor Peter Grindrod CBE, and built out at Astut Ltd. soma sells, installs and supports it.

Astut and the science

Next What it is for

The decision this is built for

Most of what an organisation faces has happened before, in some form, to someone. There is history to read, a base rate to lean on, and a body of practice to follow. Conventional analysis works well there, and firms already own a great deal of it.

A smaller class of decision has none of that. No precedent, little relevant data, and no settled objective, because the firm has not yet agreed what a good outcome would even look like. These decisions carry the most consequence and receive the least help. Prediction has nothing to learn from, because there is no past to learn from. Optimisation has nothing to aim at, because there is no agreed objective. So the work falls back on a small number of people, a short meeting, and whatever options happened to occur to them.

soma Discovery exists for that meeting.

Four paradigms, and a fifth

The market calls a great many capabilities by the same name. Set out by the question each answers, the distinction is clear.

Paradigm	Core question	What it is for	The limit
Predictive AI	What is likely?	Learn from historical data in order to predict.	Needs relevant history.
Optimising AI	What is best?	Search for the best solution against a defined objective.	Needs an objective and a reward function.
Generative AI	What can be produced on demand?	Generate text, code, images, audio and video.	Produces as directed, rather than exploring a strategic possibility space.
Agentic AI	How is the work organised?	Plan, coordinate and automate multi-step workflows.	Executes toward goals already set.
Explorative AI^®	What is possible?	Discover diverse, viable possibilities before deciding.	Built for the high stakes unprecedented decision, and honest about the rest.

The four established paradigms all presume something is already known: a history, an objective, an instruction, or a goal. The fifth presumes none of them. That is the whole distinction, and it is why Explorative AI is a different architecture rather than a better model.

Next How it works

How it works, plainly

Discovery runs two stages deliberately, and keeps them apart.

Stage one

Explore

The engine builds a large archive of decision options that are genuinely different from one another, rather than variations on a first idea. It does so through novelty search in an abstract space of meaning, where the size of the space, its degrees of freedom, how it connects and where it is empty are all unknown before the search begins. The mathematics is designed for exactly that condition.

Stage two

Test

Decision options are then validated, refuted or constrained using logic, domain knowledge and the explicit constraints the organisation has set. Options that cannot survive contact with the firm's own rules do not reach the table. What reaches the table is viable, and carries its reasoning with it.

This is not optimisation. Optimisation presumes an objective already exists and searches for the single best answer against it. Discovery widens the field of credible possibilities before anyone commits to what good looks like. The order matters, because an objective chosen too early quietly removes the options that would have been worth having.

What it does not do

It does not predict. Discovery will not say which future arrives, and any system that claims to is selling something else.

It does not optimise. It will not hand back one answer and call it the best.

It does not decide, and it is not built to. Its output is a field of viable options with the reasoning attached, put in front of a named person who chooses.

The human stays sovereign.

Next The return

The return is speed, and depth of thinking

Two returns, and they are plain enough to check. Work that conventionally runs to months comes back in hours. And the depth of thinking arrives without a team being assembled to produce it.

Hours, not months. Work that conventionally runs to weeks or months has come back in hours, observed across four demonstrations. The question is answered while there is still room to act on the answer.
Depth, without a team to assemble. No working group to convene, no sprint to commission, no diaries to clear. One question goes in, and a reasoned field of options comes back.
A far wider and deeper set of real options, including those a general purpose model trained on the past could never produce, because it can only recombine what has already happened.
Working that can be checked, because every option traces back to the question, the constraints and the themes that produced it.
No disruption, since Discovery sits above the analytics a firm already owns rather than replacing them.
A capability that is horizontal. The same engine serves finance, defence, energy, engineering, government and the public sector, because the shape of an unprecedented decision does not change with the industry it lands in.

Evidence

Large organisations already spend heavily to think. Most of it is committed to a calendar rather than to events.

Strategy advice was worth about $78 billion in 2025, within a wider consulting market near $250 billion. A focused strategy case runs $0.5m to $1.25m, and a four to eight week sprint $150k to $750k.

Chief executives give about a third of their time to strategy, and leadership teams twenty to thirty days each a year. The planning cycle runs the better part of a year, yet fewer than half are satisfied with it, and under a quarter say the big decisions are taken within it.

Set the return against that. What a firm spends there is the cost of assembling the thinking: a working group convened, diaries cleared, a sprint commissioned, and an answer that lands after the choices have begun to close. Discovery returns a reasoned field of options in hours, and none of the group has to be assembled to get it.

That is the whole of the case. Speed, and depth of thinking without the machinery.

Size it yourself

An outline of the return, on figures you already own.

Four questions you can answer from memory, and a few numbers your finance team already holds. Each lever starts from a figure you recognise, multiplied by the share earlier and wider thinking can touch, then by the honest share credited to Discovery rather than to luck. The case leans on the levers backed by invoices and timesheets. The softer one is credited lowest, so it lifts the case without carrying it.

Blue-sky thinking: hours a year, and the rate on that time The hours your people give to thinking rather than to delivery, fully loaded.

hours at £

Effect 40 percent · credited 75 percent, evidenced by timesheets

External spend: consultants for that thinking Strategy and options work, not delivery.

Effect 35 percent · credited 75 percent, evidenced by invoices

External spend: the data and research that feeds it Subscriptions and research services supporting decisions.

Effect 30 percent · credited 75 percent, evidenced by invoices

A cost you can retire: generative AI used for decision making Subscription tools Discovery replaces outright.

Effect 100 percent · credited 90 percent, a cost that simply stops

Revenue lost to late thinking Your own estimate of what came too late to act on.

Effect 25 percent · credited 45 percent, the softest lever, credited lowest

What Discovery costs you a year

Trim the whole case, if you would rather be more conservative

Case held at 100 percent

The workingA year

Blue-sky thinking—

External advice—

Data and research—

Generative AI retired—

Revenue lost to late thinking—

Value a year, after crediting—

Less what Discovery costs—

Benefit after cost, a year—

Return on cost—

Payback—

Five-year value—

Break even at—

Alongside, and not counted in the pound

Five operating measures, shown as evidence rather than as value.

Cycle time to a first set of optionsHours, not weeks

Time to a decisionDays, not months

Options examined per decisionHundreds, not a handful

CoverageEvery major decision, not the annual cycle

Speed to assess an inbound approachThe same week

Kept honest. This is a sizing model, not a forecast. It estimates value in play using figures you set, and it does not predict an outcome. Every default is a placeholder, and a real case is built on your own numbers. Judgement stays sovereign throughout: Discovery widens the field, and a named, accountable leader decides.

Two organisations, one voice

Organisation	Its part	What it holds
Astut Ltd	Invents and develops the engine, and holds the research behind it.	The intellectual property in the Astut Hypothesis Engine^® and Explorative AI^®.
soma	Sells, installs and supports the engine as soma Discovery, under a signed five year reselling agreement. Brings the sector knowledge, the implementation and the standing relationship.	The soma Discovery brand, and the way the work reaches the customer.

Neither organisation claims the other's work. A customer meets one brand, and one accountable relationship.

On credit and endorsement. Naming an individual academic and his company is a substantive claim of collaboration, and a different matter from an institutional endorsement. We make the first and not the second. The science originated at the Mathematical Institute and is credited to its author. No endorsement by the University of Oxford is implied or claimed.

As far as we and Astut can establish, no other organisation is building a general purpose Explorative AI capability for high stakes unprecedented decisions at executive level. That may not remain true, and we would rather say so than pretend otherwise.

Next Videos

Videos

Some of the decisions that matter most have no useful precedent

October 2026

Next White Papers

White papers

The argument, written out in full, so it can be checked rather than taken on trust.

Each paper is authored and dated, states its sources, and is candid about what is still to be proven. Three are available, as PDFs.

White paper · four pages

Explorative AI and the decision with no precedent

Where the field is going, and where Discovery already is. Sets out four established paradigms of artificial intelligence and a fifth, shows the same architecture emerging independently across frontier research, and explains why Discovery applies it to the executive decision rather than to the laboratory.

John C. Fay MBE
1 August 2026
PDF

Evidence note · two pages

What firms spend to think

Large organisations already spend heavily to think, in money and in time, and most of it is committed to a calendar rather than to events. The note sets out what that costs, band by band, with sources, and what changes when the same work returns in hours.

John C. Fay MBE
21 July 2026
PDF

Sector notes · five pages

Discovery by sector

The same argument in four industries. Financial services, defence, oil and gas exploration, and consumer products, each traced from the shock to the strategic opportunity, with what it means for every part of the ecosystem.

John C. Fay MBE
8 August 2026
PDF

Request the papers

We ask who you are because these are working documents rather than marketing material, and we like to know who is reading them. One set of details unlocks all three.

Full name Please give your name.

Organisation Please give your organisation.

Work email address Please give a valid email address.

Role (optional)

Your details are held by Soma Technologies and used only to send the papers and, if you agree above, to make contact once. They are not sold or shared. Write to john@soma-discovery.co.uk at any time to have them removed.

Something went wrong sending your details. Please write to john@soma-discovery.co.uk and we will send the papers by return.

Thank you. All three are below, and a copy of your details has gone to us.

Explorative AI and the decision with no precedent PDF · four pages

What firms spend to think PDF · two pages

Discovery by sector PDF · five pages

Next Our leadership

Our leadership

A small team, senior by design, close to the decisions it sells into.

Discovery is bought by people who carry consequential decisions, and it is sold by people who have carried them. While the category is young, a first conversation is with the founder rather than with a sales function.

Founder and Chief Executive

John C. Fay MBE

John founded soma. RAF University Air Squadron and IBM trained, he built his early career in enterprise software across the world for JBA. He then spent twenty five years in consulting, running his own practice, Apertur, which is now led by his former colleagues. He has worked across the United States, Europe, India, the Middle East and the Far East, and has delivered transformations in government, financial services, retail, technology and logistics, writing extensively on organisational transformation along the way. He was appointed MBE in 2006 for his work in voluntary leadership. He is the category creator of Organisational Judgement, and soma Discovery is the first step. soma Intelligence and soma Judgement are in development.

john@soma-discovery.co.uk

Chief Revenue Officer

Andy Sutton

Andy brings more than thirty years opening enterprise software markets across the United Kingdom, Europe and the United States, with a record of landmark wins against the largest incumbents in the field, among them SAP, Oracle and Microsoft. Trusted at chief executive and board level, he builds the senior relationships through which a new category is sold, and owns the commercial engine that converts them. He handles engagements, timing and terms.

andy.sutton@soma-discovery.co.uk

Management Accountant

Lisa Skopek

Lisa is a management accountant with twenty years of experience across several sectors, and has worked alongside the founder for fifteen of them. She prepares and owns the management accounts, together with the day to day financial operation: controls, reporting and transactions. Her work is why a young company can be run from the outset to the standard investors, auditors and customers will expect of it.

The engine behind Discovery is developed by Astut Ltd, under Professor Peter Grindrod CBE. Their part is set out in full on the science.

Next Professional Services

Professional services

Experienced people to frame the decision, run the programme and deploy the engine.

soma discovery is the world’s first Explorative AI engine for high-stakes, unprecedented decisions. It finds credible options that decision makers have not yet considered. Our professional services team helps clients set the right questions, act on the results and run the engine in their own environment.

What if the best option is not on the table?

Our services

Exploration Consultants

Senior leaders with 20 years plus of sector experience across government, financial services and critical national industries, used to working with boards and executive teams. They help you define the decision challenge the engine explores. They then work with your executives to review the output, test it against what the organisation already knows, and turn the strongest options into measurable actions for the board and the wider organisation.

Define the decision challenge with the executive team
Set scope, constraints and success measures
Review the option archive with the accountable leaders
Turn chosen options into actions, owners and measures
Brief the board and support the decision through governance

Programme Management

A dedicated programme manager runs the engagement from start to finish. They own the plan, risks and decision points, keep sponsors and users informed, and make sure each exploration informs the next. Where a client moves from trial to production, the programme manager carries the evidence and lessons from one phase to the next.

Mobilisation, plan and governance
Measures of effectiveness agreed at the outset
Risk, issue and dependency management
Regular reporting to the senior responsible owner
Phase reviews and the case for scaling

Technical Consultants

Technical consultants onboard the Astut Hypothesis Engine in the form that suits your security requirements and scale of use: multi-tenant, single tenant or on-premise. They work with your architecture, security and identity teams so the engine meets your standards from day one.

Deployment planning and accreditation support
Identity and access integration
Connection of proprietary data sources
User administration and training
Agreed release schedules and ongoing support

How an engagement runs

An engagement has four stages. Programme management runs across all of them.

1

Define the challenge

Agree the decision, what is at stake and what a good outcome looks like.

Led by Exploration Consultants
2

Explore

The engine generates and tests hypotheses beyond the options already under discussion.

Powered by Astut Explorative AI®
3

Decide

Leaders review the output against their own knowledge and choose which options to pursue.

Exploration Consultants with your executives
4

Act and embed

Chosen options become actions with owners and measures, and the capability is built into the organisation.

Programme Management and Technical Consultants

Deployment options

Most organisations start on a hosted trial and then move to the option that fits their security and scale. Our technical consultants plan each move so that work done in the trial carries forward.

soma hosted, multi-tenant

Immediate provisioning. Suited to trials and early use.

soma hosted, single tenant

A dedicated environment, typically ready within a week of a pilot, with the option of proprietary data sources.

Client-hosted cloud

Production use, planned around your cloud host and identity protocols.

Client-hosted on-premise

Production use within your own infrastructure. Tasks, costs and timelines agreed before work begins.

Why professional services

Explorative AI is new. The results depend on the question the engine is given and on how carefully the output is reviewed. Our consultants bring the sector experience to do both well.

They also build the same skills in your own teams, so the organisation becomes more capable of making well-judged decisions where there is no precedent. We call this capability Organisational Judgement.

Start with one decision

Most clients begin with one live decision and a short trial. It shows whether the engine widens your options before any larger commitment.

Arrange a conversation

Next Start a conversation

Challenge Us

Bring one decision.

Discovery is easiest to judge on a real question rather than a description. Bring one decision that matters and has no obvious precedent, and we will explore the field around it with you.

Write to us

What helps, before we talk

The decision or question that is live for you at the moment.
Who else tends to be in the room when it is weighed.
The constraints you would want any option to survive.
How much room there still is to act.

A sentence on each is plenty. If earlier sight and a wider option set would not change what you do, we will say so and step back.