Deciding before the terms are settled

This Resource draws on QURKI’s work on technology, markets and investment, and is published alongside the new AI and Geopolitics Signal series. It brings together considerations and governing questions for boards and senior executives examining how changing conditions affect their organisation’s strategy, commitments and ability to create and capture value.

The thinking and evidence are developed in the Signals and their accompanying resources. The work is date-stamped because macro conditions and technology continue to change.

Technology and the geopolitical context it sits within are changing rapidly. AI is opening opportunities to offer new services and change how existing work is done. It is also altering what companies depend on. Governments’ decisions about technology, trade and national capability help determine which opportunities businesses can pursue.

Those decisions can change the conditions behind a business investment within months. So can changes to AI models, technology access and suppliers’ terms. Developing expertise, replacing infrastructure and earning customers’ trust can take years.

Geopolitics and technology are reshaping markets, including where value is created and who can capture it.

Each business is different, but one question is relevant to all:

Is the way we govern helping the organisation develop a successful position as its markets, capabilities, technology and geopolitical context change?

The seven themes explored here connect that question to strategy, investment, capability and risk. Each offers considerations for discussion and governing questions, with links to the Signals and Resources that develop the arguments, examine relevant examples or explain concepts.

1. A plan answers to its assumptions. A strategy answers to the market.

A three-year plan can show how the business expects to perform without explaining why it will remain well placed to compete. A strategy needs a view of how the market is changing, where value may emerge and what the organisation can bring to it. Investment and day-to-day choices can strengthen that position or narrow what becomes possible next.

Technology and political decisions can change those prospects before the effects appear in financial results. AI may make a new service viable, change its economics or give another business a route to the customer. Waiting for clearer evidence has consequences too: others may be developing the capabilities and relationships needed to participate.

Returns may take longer than expected while customers and partners make the changes needed to adopt. Policy and regulatory conditions may also still be developing. The potential value can remain substantial, and patience may be justified. What matters is evidence of a credible route to that value, a position from which the company can earn a return, and capital prepared to support the time and uncertainty involved.

Investors differ in their appetite for uncertainty, income and growth. Investor expectations and executive incentives can reinforce each other, making it harder to change course even when doing so would better serve the company’s longer-term interests.

Governing questions

• Do we have a strategy with depth beyond “a plan on a page” and a 3-year operating plan rollup? What does it assume about our macro context and our capability fit?
• Which assumptions about our market are carrying the strategy? Who in our organisation is interpreting and surfacing changes in customers, technology, competition and policy before their effects appear in operating results?
• When returns take longer than expected, what is the evidence telling us about the opportunity, progress towards adoption and our ability to earn from it? What would justify patience, further investment or a different course?
• What have we led different groups of investors to expect? Have we adequately explained the investment thesis, including why returns may take time and what evidence would support continuing or changing course? Does the capital backing the strategy match the time and uncertainty involved?
• If we held to the current plan while the market reorganised around us, what position might others establish? What would we gain by waiting to invest, and what might we give up?
• What capabilities are we prioritising to support our place in future value pools?
• If money became dearer, or the technology / AI investment cycle disappointed, which commitments could we still sustain? Which assumptions about our funding, suppliers and customers might move together? What could we change, and which contractual obligations would remain?
• Do executive incentives support decisions in the company’s longer-term interests when those decisions conflict with the current plan or near-term targets? What might those incentives discourage management from bringing to the board? Are we asking questions and creating a culture that encourage people to surface evidence that challenges the current plan?

Related papers and relevant examples

The position is taken examines how investment changes what an organisation can do next, as well as delivering its immediate purpose. It informs the questions about capability, timing and what waiting gives up. Relevant examples include Octopus Energy’s development of Kraken and Origin Energy’s later investment and adoption.

Adoption is the value thesis distinguishes a credible opportunity, a viable route to adoption and a position from which the company can capture value. It helps distinguish a case for patience from a reason to change course. Relevant examples include US chip-card adoption and UK Open Banking. Its companion case, The value can be real. Everyone can still wait., examines why participants delayed despite the collective benefit, and how changing liability changed their reasons to act.

SoFi: investing in the ability to respond follows how capabilities acquired for earlier purposes became useful in new markets. It brings together learning from customers, changing investment during the year and the continuing need to justify returns on earlier acquisitions.

The risk is shared. The race is not. examines how borrowing, committed future payments and expectations of AI-led growth connect the technology investment cycle to wider financial conditions. It informs the questions about funding, commitments and what happens if expected returns disappoint.

Huawei: the future was part of the purchase examines how political decisions changed the expected future of investments before installed equipment stopped working. TPG’s abandoned rollout and Britain’s revised security assessment make the assumptions about upgrades, future supply and assurance tangible.

2. The market is also a jurisdiction

Countries rely on technology supplied from elsewhere to run payments, communications, cloud services and increasingly AI. That access can make their economies more productive. Important decisions about prices, technological development and continued supply may nevertheless rest with foreign companies and governments. In this context, sovereignty concerns a country’s ability to make and sustain choices about the technology its economy depends on: what it can develop, where income goes, who sets the terms and whether services remain available.

Those aims sit within geopolitical competition. A government may promote its technology companies abroad while restricting where their products can be supplied. Another may seek greater control through domestic investment, local operating requirements or cooperation with other countries. The supplier’s home government may contest those choices to protect its companies’ markets and its own influence. Businesses can be caught between objectives that their commercial agreements cannot reconcile.

For a company, the consequences reach both the markets it serves and the technology it buys. Public purchasing and cooperation between countries can create demand for its expertise or help alternatives develop. Buying international capability can also strengthen what the company can do at home. What would an investment or partnership enable it to develop or decide for itself? Which decisions about updates, prices and continued supply would remain with the supplier or subject to another country’s rules? Hosting the technology locally does not, by itself, settle those questions.

For existing operations, new requirements may mean running separate systems, employing local technical teams or replacing technology used across the group. Those changes can affect capability, costs, operational complexity and liability. Recovery arrangements may also need to change if the people, data or alternative services they rely on cannot be used in a particular market. These consequences form part of the decision about how to participate, what conditions to help shape and when continued participation would become untenable.

Customers may also impose contractual restrictions on AI use. Meeting those commitments requires understanding how AI is used across the software and services involved in their work, including tools the company has not traditionally regarded as AI systems.

Governing questions

Geopolitical context and opportunity

• What are governments in our markets trying to retain or develop through their technology policies? How do those objectives align or conflict with the interests of the countries supplying the technology, and what could that mean for our business?
• Where could our expertise, purchasing or partnerships help meet demand for greater local capability or influence over supply? What commitments would that require, which standards or public policies would support it, and who might reasonably oppose the approach?

Technology choices and control

• What influence do we need over the technology on which our business depends? Would investment, partnerships or joint purchasing improve our ability to develop capability, influence terms or keep services available, and which dependencies would remain?
• Across the jurisdictions where we operate, how could changes in rules or political relationships affect the AI models, cloud services and other technology we use? Who is assessing what they mean for where data is held and processed, who can access and maintain systems, and whether shared services across countries remain workable?
• What restrictions on AI use are customers requesting, and what have we already agreed? Are the permitted and prohibited uses clear, including AI features within everyday software? Can we meet and demonstrate those commitments and maintain that confidence as our tools and suppliers’ services change?

Cost, complexity and liability

• If we had to operate technology locally, employ local specialists or replace a restricted supplier, what capabilities could we retain - and what would become harder or unavailable? What would the additional investment and ongoing costs do to the business case?
• If different markets required different systems, suppliers or teams, how much complexity would that add to our operations? Where would responsibilities become divided, and what new risks or gaps in control could arise?
• How would those changes affect our responsibilities to customers and others, our contractual protections and our exposure to liability?

Continuity and withdrawal

• Would our business continuity arrangements still work in each market? Could we use the people, data, backup systems and alternative suppliers on which recovery depends? Where would local requirements change what we could restore, how quickly and at what capacity?
• What would cause us to reconsider remaining in a market? If withdrawal became necessary, what obligations would survive, and what would transfer to another operator require? Could the same conditions constrain potential acquirers, funding or approvals?

Related papers and relevant examples

Sovereignty in a world of layered dependencies examines why countries seek greater influence over payments, cloud and AI infrastructure, and why suppliers and their home governments may contest the changes. It distinguishes income, productive capability, influence over terms and continuity, and considers what domestic investment and international cooperation can each provide. Relevant examples include Brazil’s Pix, France’s approach to digital dependence and the continuing international dependencies of European payment alternatives.

Payments: the bargain keeps changing takes a longer view of sovereignty through payments: how international networks established enduring commercial positions, and how countries have sought to change the costs, dependencies and terms that came with them. Domestic alternatives, regulation and cooperation have changed that bargain without settling it. Wero brings the continuing contest into view, while wallets, AI-led shopping and digital money create further openings for influence and value to move. The case informs questions about what today’s investments secure, what remains dependent on others and how that position might change.

Dankort: what Denmark is choosing to keep examines what it takes to sustain a domestic payments alternative when banks, merchants and customers have different incentives. It informs the questions about collective action, which rules a company favours and who would bear the cost.

AI governance landscape is a reference map for comparing selected jurisdictions and international initiatives. It distinguishes laws, proposed measures and voluntary commitments, and explains the roles and powers of the institutions involved. It helps readers locate differences to investigate across their markets; its entries are date-stamped.

3. Where value moves, and who can reach it

AI can change how customers find suppliers, compare offers and buy products or services. It can also change which businesses they need along the way. A company may reach customers directly where it previously relied on intermediaries, or find its own role bypassed. Another company’s assistant may introduce new customers while gaining influence over what they consider buying. Who understands their needs, shapes the offering and earns from the transaction can change with the route to market.

Value also develops through doing the work. Using and adapting AI can produce specialist knowledge, training material and better ways to operate. These may support the company’s next product or help a supplier develop something it can sell more widely. Cooperation can be worthwhile while the participants compete over where the resulting value goes.

Confidentiality can protect information without reserving a market. A supplier can develop competing capabilities, or combine services customers previously bought separately, without using the company’s secrets. The investment question includes what the relationship makes each participant able to offer next.

Governing questions

• How might AI disrupt our existing route to market? Which roles in serving the customer, including our own, could be bypassed or taken over?
• If AI services interpret customers’ requests before they reach us, what will still make us the choice? What can we see about how products and services are included, excluded or recommended? Would we retain enough understanding of customers to improve our offering and identify new opportunities?
• What makes access through another company’s service commercially worthwhile? How would its terms affect our costs, ability to set prices and share of the return? What customer relationship could we develop, and which regulatory or delivery obligations would remain ours?
• What is our investment creating beyond its original purpose, for us and for others? Who has the rights and means to develop those further uses, and which could become a competing service?
• Even where our information is protected, how could a supplier’s developing capabilities or control over access to customers change our position?
• When better general AI models become available, what happens to the value of the specialist work we have funded? What new offering could become viable, and what will customers keep choosing us for?

Related papers and relevant examples

The alphabet soup of agentic commerce examines how AI shopping agents, payment systems and commerce platforms compete over the route from customer intent to purchase. It informs questions about influence, liability and access to customers. The companion map, Who is assembling the agentic-commerce market?, explains the protocols and participants in the developing market.

Open models. Competing ambitions. examines what startups and established businesses develop from available AI models, and why customers, suppliers and partners can value the same work differently. Relevant examples include Cursor’s model development, Walmart’s operational investment and the competing interests around shared technology.

What the company knows examines the commercial life of organisational expertise. Airbus’s Skywise partnership shows how combined knowledge can support new services, with agreements determining who can pursue the resulting market. The Signal also considers how suppliers become more capable through customer work, and why confidentiality does not settle competitive exposure.

Walmart: when shopping starts somewhere else connects investment in product information and AI capability to shopping through outside assistants. It develops the customer experience, commercial partnerships and advertising interests behind the questions about reaching demand while preserving a worthwhile role.

4. Model selection and AI architecture

The same commercial ambition can be pursued through different AI models and system designs. A business might buy an application, connect its own software to a hosted model, or operate and adapt a model itself. It may use several models for different tasks. The choice affects what it can offer, what each satisfactory result costs and what expertise it needs to maintain the service.

Architecture is how the models, information, software controls and people work together. It determines where the company’s definitions and rules reside, what the model is asked to infer, and how a response can lead to an action. Supplying a policy to a model is different from having software enforce it. Requiring human approval is useful only if the person can understand the situation and exercise judgment.

The evidence for a choice needs to concern the system the business will actually use. A model’s name or general benchmark score does not establish how a particular configuration will perform on its work. Testing and monitoring concern the quality of results, the effectiveness of safeguards and whether the system can adapt as the business changes.

Governing questions

Purpose and selection

• What work do we expect AI to improve or make possible? On what basis have we selected these models for that work, or as foundations for further development? Who can explain the alternatives considered and the trade-offs accepted?
• What does a satisfactory result cost at the volume we expect, including computing, human review, corrections and ongoing maintenance? What evidence supports the balance we have chosen between quality, speed and cost?

Architecture and organisational knowledge

• Who develops the models, who operates them and where is our information processed? What responsibilities and dependencies follow from buying a service, operating a model or engaging specialists to do so?
• Where do the company’s knowledge, definitions and rules reside? Which meanings must be applied consistently, what is left to a model’s interpretation, and who can examine and change those choices?
• Do licences and service terms support our intended internal use, customer services and further development? What can each supplier retain or use, and what rights do we have in the work built up through use?

Authority and human participation

• What information can the AI system access, and what actions can it take through connected systems? What enforces those limits, and who can authorise a change to them?
• Which decisions or actions require human review or approval, and on what criteria? Do those people have enough context, expertise, time and authority to challenge a recommendation or stop an action? How do we know their involvement is effective?

Testing and monitoring

• How do we test and monitor the quality of advice, decisions and actions produced by AI systems in our own operating environment? Who defines an acceptable result, and what evidence shows those standards continue to be met? Are we measuring outcomes as well as outputs, and can we distinguish the system’s contribution from other causes of improvement?
• How do we test whether safeguards limit harm from errors, misuse or deliberate attacks, including attempts to manipulate the AI system, obtain unauthorised access to information or misuse its connections to other systems? What weaknesses have those tests exposed, and how are they addressed as the systems, their uses and threats change?
• Could a new team use what we have built to answer an unfamiliar business question, change a policy or develop another service? What have tests shown about the explanations, rights or specialist help it would need?

Response and correction

• How are errors and harmful outcomes detected, reported and investigated? Who can require correction, restrict use or stop the system? Who checks whether changes to models, information, connected software or working practices require further testing before use expands or resumes?
• Who holds the records needed to explain a disputed result? Can we obtain the evidence and cooperation needed from suppliers or partners to investigate it and put things right?

Liability

• If the system’s advice, decisions or actions cause harm or loss, what liability could the company face? What do supplier contracts and insurance cover, and where might exposure remain?

Related papers and relevant examples

Some assembly required explains AI model labels, hosting and operating choices, information supplied to a model versus further training, and the software needed to control actions. Its refund example, NVIDIA supply-chain example and hosting comparison show why the whole design matters. Its practical tests examine whether a team, policy, model or workflow can change. For further technical reading, AI models and organisational knowledge: selected resources explains what each selected source contributes.

What the company knows examines how a company’s distinctive understanding becomes useful in decisions, actions and new services. It distinguishes business meanings from records, and explicit definitions from the software and authority needed to apply them. Its closing discussion asks whether accumulated knowledge remains usable when a new team or business question tests it.

Open models. Competing ambitions. informs selection for further development as well as immediate use. Cursor’s account distinguishes choosing a model to adapt from choosing the strongest finished coding product. The Signal also examines how licences affect the businesses that can be developed from available models.

5. What we are buying is partly a future

A business buys technology for what it can do today and what it expects to make possible later. Its plans may depend on upgrades, continuing support or access to an AI service or other infrastructure. Those expectations are part of the investment, but their fulfilment can depend on decisions made by suppliers and governments.

With AI, the dependency can reach into the work itself. Staffing, products and customer promises may come to rely on a model’s performance and availability. Keeping records or owning work developed around a service does not, by itself, preserve the ability to use it. A replacement needs the software, connections and expertise that make the work possible.

A supplier’s wider activities or political commitments can also make a relationship unacceptable even while its service remains useful. A decision to stay or move involves evidence about performance, confidence in the relationship and what a transition would entail. The cost can include improvements deferred while a replacement is established.

Continuity concerns the work the business could sustain, not simply whether its data could be recovered. A manual fallback needs people with the capacity and knowledge to carry it out. It needs testing, as does a replacement service, against the commitments the business would still have to meet.

Governing questions

• What does our choice of technology assume about future cost, access, support and capability? How would a change affect the work and customer promises built around it?
• Who can change those terms several relationships away from us, and how would the consequence reach us?
• Which parts of our work now depend on AI models or services? What assumptions about their performance and availability, and our own people’s expertise, underpin our delivery commitments?
• How do access and support conditions differ across our countries of operation? Does the organisation’s permission to use an AI service cover the employees, contractors and partners expected to do the work?
• What could make a supplier relationship unacceptable even if its service continued to perform? What evidence would inform that judgment, and what work or future development would a transition delay?
• Under the business continuity plan, what would be the consequences of losing part of the human or technical capability on which the business depends?
  • What could continue, at what capacity and for how long?
  • What information or functionality would be lost, and what would that mean for customers?
  • Which obligations would remain, and what work would accumulate?
• What have tests demonstrated about restoring or replacing an AI service or other critical capability within the time we could tolerate? Could the same restriction, power disruption, extreme weather or shortage of expertise affect the alternative too?

Related papers and relevant examples

The terms of access can change follows how decisions beyond the immediate supplier relationship can affect continued use, support and access and why those dependencies become consequential as AI takes on more of the work. Relevant examples include ASML’s equipment servicing, Microsoft’s interruption of services to Nayara, and the distinction between an organisation’s eligibility to obtain an AI service and its people’s eligibility to use it. The companion cases, Huawei and Fable and Mythos, develop those histories in more detail.

What the company knows examines why a supplier’s political position can become consequential independently of product quality. Its NHS discussion separates acceptability, evidence of benefit and transition consequences. Rights to keep accumulated work matter alongside the practical means of using it elsewhere.

Some assembly required explains what a business can retain and reuse when a model or service changes. Its distinction between keeping accumulated work and keeping a service running helps frame replacement planning. Its tests expose dependencies in a manageable piece of work; they do not establish readiness to move an entire service.

6. Stakeholders and the effects of change

AI can make expertise available to more people, reduce difficult work and improve services. Those gains can coexist with new demands on employees, changes in customers’ access to help or costs borne by communities supporting the infrastructure. Different groups may disagree about what constitutes a good result, as well as how its benefits and burdens are shared.

Some problems concern missing information or an inaccurate representation of the work. Others concern priorities: whose time matters, which risks are acceptable and who bears the cost of meeting a target. Better information can reveal those disagreements. People may be able to correct a record without having authority to challenge the decision it supports.

The organisation also depends on people continuing to contribute and develop knowledge. Employees’ explanations and corrections can improve a system while changing their own roles and prospects. A satisfactory result can conceal the extra effort that made it possible. Whether that experience reaches decisions depends partly on what happens when someone raises it.

Governing questions

• Whose definition of a good result is built into the measures and system instructions? What benefits and burdens fall on employees, customers, partners and affected communities, and what might aggregate results conceal?
• How can affected people influence design and priorities before they are embedded in the system? Can they challenge a priority as well as correct a fact, and who can change the decision?
• What are employees, customers and partners contributing? What do they understand about how it will be used, and how could their roles, authority or prospects change? What sustains their willingness to contribute further knowledge?
• What workarounds or additional human effort make the reported results possible? Who hears about them, and can that evidence change the targets or how work is organised?
• As AI gets better at our business, what are we and our people getting better or worse at? What evidence shows that people are developing the understanding needed to question the systems and develop new work?
• How can people affected by an AI-supported decision obtain an explanation, challenge it or seek a remedy? What happens to the person who raises a problem, and how does that experience inform later decisions?

Related papers and relevant examples

What the company knows examines whose judgment shapes how organisational knowledge is used, who benefits and whether expertise keeps being produced. HCA’s scheduling work distinguishes recording staff preferences from giving those preferences influence. Research on AI assistance tests learning alongside immediate performance. The Signal also examines the incentives to contribute knowledge and why outcomes that satisfy one group may undermine another’s confidence in the organisation.

Walmart: when shopping starts somewhere else brings employee experience into the account of operational improvement. It distinguishes correcting an inaccurate estimate from questioning whether a target is reasonable, and follows the disagreement over reporting on workforce effects.

7. Acting on what we learn

An organisation can recognise a change and still be unable to act on it. Evidence may reach the wrong people, arrive too late for a contract renewal or fail to alter an approved budget. Evidence of a greater opportunity needs a route to a decision just as evidence of risk does.

The company’s own controls address only part of its exposure to AI. Suppliers and governments can recognise a danger while resisting action that would leave them behind competitors or rival countries. Investment commitments and expectations of growth add pressure to keep developing and deploying the technology. An evaluator may identify a problem without having power to require a response. The company therefore needs to examine both the evidence behind external assurances and who could act on it.

Readiness involves the time, information and authority needed to act. Waiting for certainty can consume the time needed to prepare; reacting to every change can prevent useful work from being completed. Findings from market monitoring, system tests and stakeholder experience need to reach decisions about resources and commitments while a response remains possible.

Governing questions

• Who could explain our current exposure and readiness, and how long would it take to assemble the answer? What supplier assurances, maintenance obligations or recovery assumptions remain untested?
• As markets, technology, threats and people change, who checks whether earlier assessments still hold? How do the findings reach the risk framework, investment decisions and allocation of resources?
• Where do we rely on a supplier, evaluator or public authority to act? What evidence and power do they have, and what remains our responsibility if their response is delayed or disputed?
• What evidence would justify expanding, limiting or stopping a use of AI? Who can require that response when it conflicts with a commercial commitment?
• What can management change under its existing authority, and what requires a board decision? Would our information, funding and meeting arrangements allow that decision while it could still make a difference?

Related papers and relevant examples

Post-quantum readiness is not a cryptography project examines preparation where the timing of a threat is uncertain but information has a protection life and systems take time to replace. It informs questions about maintenance, testing, decision authority and the cost of waiting. It also examines how renewal and procurement decisions can preserve or narrow future options. The accompanying Post-quantum readiness: selected resources identifies standards, guidance and practical material for further investigation.

The risk is shared. The race is not. examines the gap between recognising advanced-AI dangers and being able to act on them. It brings together evaluation failures, financial commitments and geopolitical competition, asking who can require restraint and who would bear its cost. It supports scrutiny of the incentives and authority behind external assurances.

AI governance landscape helps distinguish a safety commitment from an enforceable duty, and an evaluation programme from authority to intervene. It provides a dated starting point for examining who could act on a problem beyond the company’s own boundary.

What the themes share

A supplier renewal, a staffing choice or an investment approval can affect both today’s performance and tomorrow’s ability to compete. The questions concern how changing conditions reach those decisions: who interprets the evidence, whose interests are considered and what the organisation remains able to do.

Delivering the plan matters. So does the position the company is building while it delivers.

About the linked reading

The linked Signals and Resources develop the arguments and carry their source references. This draft draws on the QURKI review set dated 30 October 2026 and was revised on 4 November 2026. Reference maps describe conditions at their stated dates. The questions invite examination of the company’s circumstances; the supporting papers do not supply answers for every business.