When delegated AI reprices flexibility – and who ultimately pays

Shikko Nijland, CEO of Innopay: The emerging divide isn't technological access – it's delegated representation: who has an agent negotiating on their behalf, and how well it performs.

In brief

Delegated AI is not simply automating transactions. It is automating the exercise of rights. As agents begin to buy, cancel, negotiate, and dispute at machine scale, they alter the operating logic of markets. Consumer protections designed for episodic human use become inputs to continuous optimisation. Efficiency gains are real — but so is the redistribution of optionality. If optimisation gains are privately captured while adjustment costs diffuse across the ecosystem, flexibility will be economically repriced. Those with strong, aligned agents will recover that value. Those without will absorb it. The Agent Advantage Gap widens accordingly.

From digital divide to delegated divide

For two decades, digital inequality has centred on access: connectivity, devices, and digital literacy.

That divide remains.

A deeper shift is now underway. Increasingly, routine economic and administrative decisions will be delegated to autonomous agents capable of searching, comparing, transacting, cancelling, renegotiating, and disputing on behalf of individuals and organisations.

When delegation becomes the default interface to markets, the relevant question changes. It is no longer only about whether people can participate online. It becomes: who is represented by which agent, under what constraints, and in whose interest?

This is not simply faster ecommerce. It challenges the behavioural assumptions embedded in market design.

Most consumer protections — cooling-off periods, refund windows, chargebacks, free trials — were designed under an implicit premise: that rights would be exercised intermittently, constrained by human attention and effort. Friction functioned not only as an inconvenience but as a stabiliser. Pricing models assumed partial exercise.

Delegated optimisation changes that premise. And once rights can be exercised at machine speed and at scale, the economic meaning of “flexibility” cannot remain unchanged.

Human-bounded versus machine-bounded markets

In human-bounded markets, protections operate as safety valves. They correct edge cases without reshaping pricing structures.

In machine-bounded markets, the same protections can be executed continuously and programmatically. What was occasional becomes repeatable. Optionality becomes harvestable.

Early adoption may generate clear positive-sum gains: lower search costs, faster matching, and reduced friction. The distributional challenge emerges when optimisation shifts from reducing cost to competing over scarce allocation — limited inventory, priority access, ranking position, or contractual flexibility. At that margin, gains for one actor increasingly correspond to losses or higher adjustment costs elsewhere.

This does not imply misconduct. It reflects predictable behaviour when systems are optimised against explicit objectives such as price, latency, or allocation. Importantly, delegated agents may also expand inclusion in some contexts — reducing search and administrative friction for underserved populations — provided minimum quality, alignment, and contestability standards are in place.

A concrete illustration: travel bookings

Consider a refundable airline ticket. Historically, only a minority of customers would book speculatively and later cancel if prices changed. Monitoring and rebooking required effort. 

Now imagine agents provisionally booking multiple flights, monitoring prices in real time, and cancelling and rebooking automatically when cheaper fares emerge. At a limited scale, this behaviour is absorbed operationally. At the material scale, pricing models adjust — refundable fares become more expensive, flexibility is segmented into premium tiers, and non-refundable options expand. 

The protection remains legally intact. Its economic value changes. That shift is the hinge of the broader argument.

This dynamic is not hypothetical. In the human-bounded economy, the repricing of flexibility is already underway. US retail returns reached USD 890 billion in 2024, with online return rates at 19.3%. In response, two-thirds of US retailers introduced return fees in the past year. Major retailers have shortened return windows from 60 to 30 days. Restocking fees are proliferating. (NRF / Happy Returns, “2025 Retail Returns Landscape Report”) These are the early indicators of the repricing this article describes — and they emerge from human behaviour still constrained by attention and effort. Delegated agents remove both constraints.

The embedded option dynamic

When an agent can transact immediately and reverse later if conditions improve, consumer protection behaves economically like an embedded option.

Downside is capped by refundability. Upside remains open. Time increases option value.

For human actors, exploiting that optionality was rare and costly. For agents, it is rational and repeatable. Even where penalties exist, agents optimise around them, accelerating the path toward tighter equilibria. In hospitality, the dynamic is already measurable: OTA bookings with free cancellation show cancellation rates of 25–45%, and hotels are deploying predictive models to score no-show probability per reservation — a defensive response to optionality exercise that has not yet been fully automated. (Jengu, “How AI Reduces Hotel No-Shows,” 2026)

There is no universal tipping point. Thresholds depend on sector economics, reversibility costs, and regulatory design. The relevant signal is not abstract adoption, but the share of reversible transactions executed programmatically and the divergence between provisional and final demand.

Once that divergence becomes material, pricing models adjust. And because market actors adjust competitively rather than centrally, the system tends to tighten through feedback loops, not through a single visible break.

Competitive escalation: optimisation as arms race

When one actor deploys optimisation agents, others must respond. 

If consumers arbitrage refund windows, merchants introduce defensive optimisation. If platforms embed dynamic repricing to offset reversibility risk, competitors follow. If suppliers
tighten policies, intermediaries redesign products to preserve optionality.

Optimisation begets optimisation.

The arms race is already visible. Return fraud accounts for an estimated 9% of all returns, costing US retailers over USD 100 billion annually. In response, 85% of retailers now deploy AI-based fraud detection. Serial returners — 5–8% of customers — account for 35–45% of all returns, prompting policy tightening that affects all customers. (NRF, 2025) This is the competitive escalation cycle this article describes, operating in the human-bounded economy. Agent-mediated automation accelerates all sides simultaneously.

These dynamics can be amplified in concentrated platform markets, where pricing power and rule-setting are more centralised and adjustment costs can be passed through more quickly.

No single actor may intend to dilute flexibility. Yet competitive escalation shifts equilibrium conditions.

Markets internalise observable costs. Diffuse ecosystem costs — working-capital strain, dispute overhead, and repricing effects — are slower to attribute and therefore more likely to be socialised. The consequence is a gradual repricing of flexibility that is easy to rationalise locally, but significant in aggregate.

From optimisation to systemic arbitrage

In a machine-bounded regime, provisional demand becomes less reliable; inventory appears committed but remains reversible; forecasting error increases; working capital requirements rise; and dispute cycles lengthen.

Merchants and platforms respond rationally. Refund windows tighten. Deposits increase. Flexibility is segmented. Verification friction rises. Margins embed reversibility risk.

This is not market failure. It is a competitive adjustment.

The distributional question persists: who can arbitrage the repriced environment? Answering that requires following the adjustment costs as they move through the system.

The hidden adjustment cost

The cumulative effects resemble a distributed adjustment cost — a “hidden tax” in the economic sense — embedded in prices and conditions: higher base prices for flexible goods and services; narrower cancellation windows; tiered access to priority treatment; increased compliance and dispute overhead; larger capital buffers across supply chains; and additional verification friction.

Each step is locally rational. Collectively, they shift welfare distribution.

This distributed cost is not uniformly recoverable. Actors with strong, well-aligned agents can continue to extract residual optionality — adjusting timing, routing, and policy navigation. Actors with weaker or no agents absorb the repriced environment passively.

Even after equilibria tighten and margins compress, residual optionality remains unevenly exploitable. The issue is not whether repricing is efficient. It is who can arbitrage the repriced system.

The Agent Advantage Gap

This asymmetry explains the convex nature of the Agent Advantage Gap.

Advantage compounds through continuous optimisation. Disadvantage compounds through exposure to adjustment costs that cannot easily be arbitraged.

Scarcity lies not in access to “any agent,” but in access to aligned delegation, supervisory capacity, and supportive trust infrastructure.

Infrastructure defines the feasible action space. Agent sophistication determines how that space is exploited.

Where identity systems, mandate controls, and enforceable governance are robust, agents operate transparently. Where infrastructure is fragile, delegation may amplify instability. In lower-trust or lower-capacity jurisdictions, these dynamics can be amplified by cross-border platform dependence, weaker enforcement, and limited access to public trust rails — shifting both value capture and adjustment costs outward.

The gap amplifies existing vulnerabilities rather than creating new categories of disadvantage. This matters because the adjustment costs of agentic optimisation do not land on a blank slate — they land on existing capacity constraints.

Slow adjustment, not visible crisis

The defining feature of this transition is gradualism.

Markets do not collapse. Instead, flexibility is incrementally repriced; conditions tighten at the margin; fees proliferate; and access becomes segmented.

Each adjustment appears rational in isolation. Over time, however, behavioural assumptions embedded in consumer protection may shift materially. By the time the shift is recognised, the new equilibrium may already have redefined what “protection” and “flexibility” mean in practice.

The strategic question: net societal contribution

Agentic optimisation can create meaningful efficiency gains. The strategic question is whether those gains remain net-positive at the system level.

If business cases account only for direct efficiency benefits while externalising ecosystem-level costs — reversibility friction, dispute escalation, liquidity strain, repricing effects — optimisation risks eroding the trust infrastructure on which markets depend.

Sustainable competitive advantage requires internalising systemic impacts rather than diffusing them.

What leaders should monitor

Boards, regulators, and market operators should treat delegated optimisation as a structural risk surface.

Early-warning indicators are practical and observable. Reversibility intensity: cancellation rates, rebooking loops, dispute frequency, programmatic reversals. Signal degradation: divergence between provisional and final demand, forecasting volatility. Policy drift: tightening refund windows, fee proliferation, and flexibility segmentation. Distributional exposure: which segments absorb repricing versus those able to arbitrage it.

Regulatory triggers should focus on reversibility intensity and signal degradation rather than abstract adoption metrics.

Governance design principles

Mitigation should prioritise resilience, not restriction.

Treat trust infrastructure as foundational: digital identity, mandate frameworks, and auditable logs. Establish minimum quality standards for consumer agents. Embed contestability in delegated
decisions. Incorporate ecosystem externalities into strategic planning.

The objective is not to weaken consumer protection, but to preserve its economic meaning in a machine-executed environment.

Conclusion: representation defines the next divide

Agents will optimise. Competitive pressure ensures they must.

The task for leaders is not to slow optimisation, but to design institutions and market rules that can absorb it without silently diluting protection or widening distributional gaps.

In an agent-mediated economy, inequality will increasingly hinge on representation: which systems negotiate on one’s behalf — and whether those systems protect or extract.

The next divide will not simply be digital. It will be delegated.

Five strategic questions for boards and policymakers

As delegated AI becomes structural, governance must evolve from product oversight to system stewardship.

1. Are we entering a machine-bounded regime — and are we instrumenting it? What share of reversible transactions is executed programmatically? Are cancellation loops, rebooking cycles, and dispute flows tracked separately for human- and agent-initiated behaviour? At what point does provisional demand begin to distort price, capacity, and risk signals?
2. Are efficiency gains shifting into allocation competition? Where has optimisation moved from reducing friction (positive-sum) to competing over scarce allocation — inventory, priority access, ranking position, or contractual flexibility? Which parts of our model are most exposed to provisional demand?
3. Are we in a competitive optimisation cycle? If one actor deploys defensive or extractive optimisation, must others follow to protect the margin or share? Are tightening policies and repricing emerging as structural responses rather than deliberate strategy?
4. What is the net societal contribution of our agent strategy? Do our business cases internalise ecosystem-level impacts — reversibility friction, dispute escalation, liquidity strain, repricing effects — or are these diffused across the system? Are optimisation gains demonstrably net-positive at the system level?
5. Do we have guardrails before equilibrium shifts? Is trust infrastructure robust — identity, mandate controls, auditability? Are regulators and market operators monitoring reversibility intensity and signal degradation early enough to preserve the economic substance of consumer protection?

About the author

Shikko Nijland is CEO of Innopay, a payments and digital identity consultancy within Oliver Wyman (Marsh McLennan). He co-authored Everything Transaction (Netherlands Management Book of the Year, 2019) and advises financial institutions on the intersection of digital transactions, trust infrastructure, and adaptive AI governance.

Follow Shikko on LinkedIn: 
https://www.linkedin.com/in/shikko/