Loading...
AC
Loading...
Loading...
AI Strategy
Portable agent skills and connected consumer AI are moving customer decisions and actions beyond the traditional CRM interface. Businesses need an expanded control model covering identity, permissions, authoritative data, auditability and human approval across advertising, commerce, CRM and accounting systems.
Loading...
If your customer journey is already crossing CRM, advertising, commerce and accounting systems, talk to Agent Crew about mapping a governed AI workflow before adding another disconnected assistant.
Book a demoCustomer relationship management platforms have traditionally controlled the operational centre of the customer journey. Leads entered the CRM, sales teams updated opportunities, marketing tools synchronised audiences, and finance received the final handover after a deal closed.
That boundary is becoming less distinct.
Consumer AI assistants are developing connections to advertising, product discovery and commerce. Business assistants can work with customer, campaign and accounting data. At the same time, portable agent skills allow organisations to package procedures, policies and domain knowledge so they can be used across compatible AI environments.
The result is not the death of the CRM. It is a change in its role. The CRM remains an important system of record, but it may no longer be the only place where customer intent is interpreted, decisions are made or commercial actions begin.
The strategic question is shifting from which platform owns the customer to which control model governs customer actions across platforms.
Three developments are converging.
An agent skill is a reusable package of instructions, supporting resources and, where required, scripts that teach an AI agent how to perform a particular type of work. A skill might describe how to qualify an enquiry, prepare a compliant proposal, classify a support request or review an overdue account.
Unlike a prompt saved inside one employee's account, a properly managed skill can become a versioned organisational asset. It can encode approved terminology, escalation rules, templates and checks.
This concept is also becoming more portable. In October 2026, Anthropic transferred its Agent Skills open standard to the Agentic AI Foundation, creating a vendor-neutral home for the specification. (anthropic.com)
Portability does not mean every skill will work identically in every environment. Models, permissions, available tools and execution methods still vary. It does mean businesses can increasingly treat procedural knowledge as a layer that is distinct from a particular CRM screen or AI vendor.
AI assistants are moving beyond answering questions. They are becoming places where consumers discover products, compare options, interact with brands and, in some cases, initiate transactions.
OpenAI's Agentic Commerce Protocol, for example, provides a structured connection between merchant systems and product experiences in ChatGPT. OpenAI has also announced advertising integrations involving HubSpot and Shopify, alongside sponsored conversational agents. These developments remain relatively new, and availability can differ by market, account and programme. (developers.openai.com)
For a business, this creates customer interactions that may begin outside its website and CRM interface. A prospect could express intent, receive product information and engage with a brand-controlled experience before completing a conventional lead form.
Accounting platforms contain signals that are often absent or delayed in the CRM. These include invoice status, payment history, recurring revenue, credit notes, refunds and the financial position of an account.
Connections between AI assistants and accounting systems are making this context more accessible. Xero, for example, provides a hosted Model Context Protocol connection, which allows authorised AI clients to access supported accounting data and actions. OpenAI and Intuit also describe QuickBooks workflows that can be accessed through ChatGPT with existing permissions and user confirmation.
This does not turn an accounting platform into a CRM. It means that decisions about customers can be informed by operational and financial reality rather than sales data alone.
The control boundary is the set of systems, rules and people that determine what may happen during a customer interaction.
It includes more than contact records. It covers:
In a traditional implementation, many of these controls sit inside the CRM or closely connected applications. In an agent-enabled environment, the interaction may be initiated in a consumer assistant, interpreted using a portable skill, fulfilled through a commerce system and reflected in accounting software.
The CRM may receive the result without having controlled every preceding step.
Consider a professional services firm receiving an enquiry through an AI assistant.
The assistant may collect the prospect's requirements and with appropriate disclosure and permission, pass structured information into the firm's environment. An agent applies the firm's qualification criteria, checks service availability and drafts a recommended response. A staff member approves the scope and fee before the proposal is issued.
Once accepted, a commerce or document system records the agreement. The accounting platform creates the customer and invoice, while the CRM receives the opportunity, correspondence summary and next action.
No single application owns the complete journey. The business process spans several systems, with AI helping to translate intent and coordinate work between them.
This can reduce repetitive handovers, but only if the systems agree on identity, status and authority. Otherwise, the same architecture can produce duplicate contacts, inconsistent offers, premature invoices or customer communications based on incomplete data.
A CRM should still provide a dependable commercial record. Sales history, relationship ownership, pipeline state, customer communication and service context remain valuable.
What changes is the assumption that every decision must originate inside the CRM.
In a distributed customer environment, the CRM becomes one part of a wider control plane. That control plane may include:
The practical architecture will vary. A small hospitality group will not need the same controls as a financial services organisation. The design should reflect the consequences of each action rather than applying one autonomy setting to the entire workflow.
AI reasoning is useful where customer requests are ambiguous. It can classify intent, summarise conversations, extract requirements and draft tailored material.
It is less suitable as the only control for actions that require exact, repeatable outcomes. Consent status, tax treatment, ledger posting, discount limits and approval thresholds should generally be enforced through deterministic rules and system permissions.
A robust workflow might allow an agent to recommend that an overdue customer receive a revised payment plan. A deterministic check would confirm eligibility, account status and approval authority before anything is sent or changed.
This combination is central to dependable AI agent and workflow automation. The model handles interpretation where flexibility is valuable. The surrounding workflow constrains actions where consistency matters.
Each important data object should have an authoritative source. The CRM might own relationship status, the commerce platform might own orders, and the accounting system should own posted financial transactions.
Read permission is different from permission to send, publish, discount, refund or post a journal entry. Access policies should distinguish between analysis, drafting and execution.
Portable skills should have owners, version histories, test cases and review dates. Installing an external skill without inspection can introduce unsuitable instructions or executable components into a business workflow.
Logs should connect the original customer request, the information retrieved, the agent's recommendation, any human approval and the resulting system updates. Separate application logs are not enough if nobody can reconstruct the complete event.
A customer process should not depend entirely on one assistant or advertising platform. Core policies, records and controls should remain within infrastructure the organisation can govern, even when external channels initiate the interaction.
CRM strategy can no longer be assessed solely through feature comparisons between platforms. Leaders should examine how customer intent, organisational knowledge and commercial authority move across the wider technology environment.
Useful questions include:
Technology controls alone will not solve the operating model. Marketing, sales, service, finance and IT need a shared understanding of ownership and escalation. Targeted AI training for business teams can help staff use connected assistants without confusing convenience with authority.
The expanding boundary creates genuine opportunities. Businesses can respond to intent earlier, reduce manual re-entry and connect customer engagement with fulfilment and financial context.
It also creates a new governance requirement. Customer processes must remain coherent when they cross systems that were purchased by different departments and designed for different records.
The strongest approach is not to force every interaction back through one platform. It is to establish clear authority, permissions and auditability across the full journey, while preserving human judgement where consequences are material.
If your customer journey is already crossing CRM, advertising, commerce and accounting systems, talk to Agent Crew about mapping a governed AI workflow before adding another disconnected assistant.