Where Aggregators Win with Agentic AI

Twenty hours. That is how long a typical person spends planning a two-week holiday. Finding a mortgage provider, taking out an insurance policy, switching energy providers or selecting mobile plan are equally involved. Aggregators like Booking, Expedia, Comparethemarket, and Zillow earned their place by turning this mess into one screen, one set of filters, and a clean handoff to an operator. For twenty years, that has been enough.

Personal AI agents are now set to ease that burden. They can interpret our intent, fetch prices and terms from the source, and complete the purchase from your wallet with consent. They can schedule viewings, prefill applications, and check small print against your preferences. The routine comparison work that once justified the aggregator’s spread is being automated.

So what is left for aggregators in an agentic world. More than you might think, and less than they are used to. They will not be paid for access to inventory and clicks. They will be paid for reliable outcomes when real life gets messy. Think guarantees that a trip arrives as booked. Think instant rebooking when a hotel room is un-livable at 2 a.m. because of the noisy ice machine down the hall, think facilitating an insurance claim, or negotiating a late loan repayment. Think onboarding that stitches together identity, credit checks, and policy rules across multiple operators without the customer lifting a finger.

If you run an aggregator, a bank, an insurer, a utility, or a brokerage, this isn’t a long term roadmap candidate. It is an operating choice that needs to be made soon. When agents can find and compare, what will your part be in the agent-powered customer journey that earns trust, reduces risk, and delivers outcomes customers will happily pay for.

An aggregator (e.g. MoneySuperMarket, Policygenius) brings offers from many operators into one place, standardizes how you compare them, and routes you to a provider that can fulfill. It usually does not hold inventory or carry legal responsibility for the outcome. A marketplace (Zillow, AirBnB) hosts buyers and sellers, sets trading rules, and often processes payment, so it carries more operational responsibility and takes a higher cut. Metasearch (Skyscaner, Kayak) is a search layer over many sellers that sends you out to book elsewhere rather than completing the transaction itself.

Aggregators earned their place because the early web was messy. Discovery was hard, product data was inconsistent, and trust was scarce. By normalizing information, ranking options, and simplifying checkout, they reduced time and risk for buyers while giving operators a reliable source of measured demand.

The money flows through a handful of engines. Click and action fees pay for traffic or completed applications. Commissions and revenue share are paid when a booking, switch, or policy is completed. Advertising and promoted placement sell visibility inside high-intent journeys. Subscriptions and tools monetize seller workflows and analytics. Ancillaries such as insurance or premium support add margin. Data products and benchmarks round out the mix.

The pattern shows up differently by sector. In travel, online agencies act as marketplaces that take payment and own the itinerary, while metasearch routes to airlines or hotels and earns on clicks and ads. In financial services, comparison and eligibility checks feed lenders and insurers who pay per lead, per funded loan, or per policy, with regulation shaping how much can happen on the aggregator’s site. In utilities and telecom, switching services enroll customers with a new provider for a bounty per completed switch, while post-sale service sits with the provider. In real estate, portals monetize attention and leads that agents and developers buy, but legal transfer and settlement remain with licensed professionals outside the portal.

This is the baseline the agentic era will challenge. Aggregators that only sell discovery and clicks will feel margin pressure. Those that can translate complexity, carry some responsibility, and make outcomes reliable will still earn their keep.

Agentic commerce is the AI evolution that builds on self-service ecommerce. Phones now ship with enough compute for on-device reasoning, edge models cut latency, and assistants are moving from chat to action: Apps will turn into Personal AI. As these mature together, more of the journey shifts from clicking through pages to delegating a task.

Here is what agents already do. They take a plain instruction, interpret it, and act. Using Model Context Protocol (MCP) that communicates with back-end applications, operators publish machine readable offers, terms, and SLAs that agents can parse at source. Agents coordinate with operator and aggregator services using a protocol that enables AI intercommunication (e.g. A2A), then prefill forms, request proofs from an identity wallet, initiate payment, and complete the transaction on Agent Communication Protocol (ACP) rails. After purchase they monitor status, trigger remediation rules, and escalate to a human with full context when needed.

The result is less friction and better economics for everyone using them. Search and comparison time falls. Abandonment drops because there is no rekeying or confusing hand-off. Conversion rises because the agent matches intent to inventory and executes cleanly. Operators are accelerating the shift by launching MCP endpoints, adopting ACP, and releasing their own brand agents so customers can transact directly through a trusted, automated flow.

For the last twenty years the roles were clear. Aggregators, brokers, marketplaces and metasearch engines were independent intermediaries or co-mediaries. They took commission from operators and felt free to the consumer. Their job was to attract demand, compare options, and pass you on. They acted for their own platforms and shareholders, not as a fiduciary for the buyer.

Adding a personal agent changes the cast. A buyer can now bring an agent [PERSONAL AI]  that sits with them, talks directly to operators, or works through an aggregator. The legacy routes remain in play, where a customer goes direct to an operator, or uses an aggregator that hands off to an operator. Two more are common when operator platforms deploy their own assistants [VENDOR AI] , where a customer engages an aggregator’s in-house agent [BROKER AI] , or an operator’s brand agent, and the platform’s interests lead the way. Finally, the hybrids appear, where a buyer’s personal agent coordinates with an aggregator’s in-house agent, or with an operator’s brand agent, to complete the job. Taken together there are eight active paths through, and they will coexist by sector and regulation.

Ownership governs loyalty. If the agent belongs to a party, assume it serves that party. An operator’s brand agent optimizes for the operator’s goals within disclosed service levels. An aggregator’s in-house agent routes to outcomes that suit the aggregator’s business model unless conflicts are disclosed and bounded. A buyer-owned personal agent should act only for the buyer, under a clear mandate the buyer controls. This is the core argument in Who Owns the Agent.

Fiduciary duty is how we keep this honest. Richard Whitt’s human-centered paradigm describes a digital fiduciary as a filtering conduit for a client’s online life, bound by duties of care and a thin duty of loyalty that avoids conflicts of interest, with practical tools like alt-consent, local data access, and intent casting. In business terms, a buyer-owned personal agent carries a signed mandate, discloses or rejects conflicted compensation, and leaves an audit trail that the customer and a regulator can inspect. Although as my friend

Tony Fish

correctly points out “There is no fiduciary duty other than to act in the best interests of the shareholders; all other duties are nice to have, but you are not held to account, with the exception in the UK of the HSE – but this is a criminal offence for directors if they fail.”

For your industry, your business, your customer journey you should document all eight paths, show who owns which agent, who pays whom, and where liability sits. It should mark where the buyer’s mandate lives and how it travels. Once those roles and money flows are explicit, the next sections can describe the technical foundations that enforce fiduciary behavior and the financial model that pays for it.

Agentic AI could give customers real power, but only if the provider allows the agent can act on the customer’s behalf safely. That is what empowerment tech delivers. It lets a personal AI prove who it represents, share only the proofs a counterparty needs, and complete the job with an auditable record. Without it, an “agent” is just a chat bot. With it, the agent can participate in the transaction.

In practical terms, empowerment tech ties three pieces together. First, a decentralized identifier (DID) that binds the agent to a real person. Second, verifiable credentials pulled from a personal data store the customer controls. Third, selective disclosure and consent so the agent presents only what is necessary to move forward. If you are shopping for a mortgage, your agent asserts you are a legal person via your DID and presents a credential that you meet the lender’s income threshold without revealing your actual salary. The lender gets what it needs, you keep your privacy, and the process moves without screenshots or rekeying.

Banking shows why this matters. Incumbents built economics on inertia. Open banking and agents will erode that. McKinsey suggests that Agents can monitor balances in real time, compare returns across institutions, sweep idle cash into higher yield accounts, and sweep it back in time for bills. The spread that sat with banks starts to flow to account holders. New AI-first banks will lean into this and launch services around it. If incumbents unbundle into collections of services that can be mixed and matched, aggregators are well placed to coordinate those services and guarantee outcomes.

Either way, adoption depends on trust. This is the ground Jamie Smith at Customer Futures calls empowerment tech: the convergence of personal data, identity, and AI in a form people control. In his words, “the next-generation digital economy is going to be powered by two things: personal data, and AI.” That power comes with risks “of discrimination… bias… security… privacy… manipulation,” which is why humans “are going to need control over their (personal) data, and the rich and deep (personal) insights that come from it”. Smith’s checklist is the right discipline for any agentic experience: Why are we using these tools. What data feeds them. Who decides and who sees the results. Where is the data processed. How are insights revealed, especially for money or health. These questions are not pedantry. They are the guardrails that make some delegation safer.

Designing new aggregation experiences starts with what agents have commoditized and where differentiation now lives. Two prompts help. First, what additional services can you offer that agents cannot or should not do alone. Second, how do you make those services personal using proofs rather than more data grab.

Here is how that plays out beyond search, compare, and complete.

Position: the marketplace sits with operators and exposes an endpoint your personal agent can call. Value beyond comparison: authenticity checks, price plus fees transparency, inventory and delivery guarantees, and unified returns and warranty execution. Empowerment tech lets the marketplace accept agent-originated orders with proofs, hold funds until conditions are met, and automate redress if a promise slips.

Position: the OTA coordinates airlines, hotels, rail, and ground while speaking A2A with supplier agents and your personal agent. Value beyond comparison: disruption recovery with authority to reissue, cross-supplier orchestration on one record, chargeback defense, and duty of care at scale. Your agent shares identity, preferences, and payment scopes once. The OTA enforces machine-readable SLAs and remediation rules across suppliers.

Position: the aggregator sequences lenders and bureaus while your personal agent carries your mandate and proofs. Value beyond comparison: safe eligibility and affordability checks with data minimization, packaging of disclosures, lender failover without restarting, and a clear time to offer with remedies if missed. The promise shifts from “we found a low rate” to “we will get you a decision you can rely on, quickly, with your data under your control.”

As

Jamie Smith💡

argues, we should treat these systems like advisors and demand answers to the WHY, WHAT, WHO, WHERE, and HOW before letting them near money, health, or relationships. If we do, agents will free up time and working capital, while aggregators and marketplaces can rebuild customer value on top of empowerment tech: identity that travels, credentials that prove, consent that is revocable, and service that owns the outcome.

If you run an aggregator, you need a clear map of where agents will take over and where you still create outsized value. This section gives you that map. It separates “agent-native” tasks that will be automated from “aggregator-advantaged” tasks where you can win, then defines three moats you can build and shows quick sector notes so you can prioritise investment.

Use this table to decide what to stop, start, and scale.

1. Resolution and liability at scale Own the outcome, not the click. Publish specific guarantees, set reserves, and give human support the authority to fix problems on the spot. Price access to your rails on fulfilled outcomes and recovery, not impressions.
2. Bundling and inventory orchestration Coordinate multiple operators behind one ActionID. Standardise how identity, consent, payments, schedules, and exceptions move through the journey. Make it trivial for personal agents and operator agents to trigger your remediation playbooks.
3. Performance data that trains agents to trust you first Build a reliability graph that scores partners on fulfilment, refund time, dispute rates, and time to resolve. Expose these signals in machine-readable form so personal agents route to you by default when stakes are high.

This is how an aggregator stays in the value chain when agents handle discovery. You become the service and trust layer that turns intent into clean, reliable outcomes.

There is no neat answer yet. What we do have are clear signals about the technology, how fast it is being adopted, and where it presses on the aggregator model. The next phase is to stay open to ideas built for a new kind of teammate, the personal agent, and to redesign journeys so humans and agents complement each other.

I’m continuing the research so you don’t have to, and I can help you define an operating model where agentic AI creates the space to differentiate and compete. If you want a sounding board, I’m happy to run a short working session with your team.