From Chatbot to Operating System: How Open Protocols Are Rewiring Enterprise AI

Finance leaders are no strangers to AI experimentation. A survey from KPMG found that three-quarters of organizations actively use AI within the finance function, up from 30% just two years ago. But while adoption is high, the first generation of AI hasn’t delivered the transformation most finance leaders hoped for. Instead, it’s left three in 10 finance leaders struggling to prove a return on investment.  

Today, the most widely adopted AI tools take the form of chatbots. They can summarize information, answer questions, and provide suggestions in response to user queries. But because these tools operate in silos, they can’t meaningfully change the way most finance teams work.  

To see real transformation, finance teams need AI with access to live systems, proper permissions, and the right business context. Open protocols like Anthropic’s Model Context Protocol (MCP) can provide the infrastructure to transform AI from a generic chatbot into something much closer to an operating system for finance.  

AI needs a common language  

When finance leaders have budget to invest in AI, they have no shortage of options. There are countless standalone tools on the market, as well as established vendors bolting new AI features onto their existing platforms.  

But traditionally, each AI application functions independently, with its own way of accessing the software, databases, and internal tools needed to complete its tasks. This inconsistency creates unnecessary complexity for organizations and makes it difficult to scale AI across the business. Open protocols like Anthropic’s Model Context Protocol (MCP) help solve this problem.  

MCP is an open-source standard for how AI models connect to external systems. AI applications that use MCP can securely access the data, systems, and workflows needed to perform their intended function without exposing unnecessary risk.  

Think of MCPs as the AI equivalent of USB-C. Instead of every device using its own proprietary connector, USB-C allows compatible devices to connect seamlessly. Similarly, MCPs provide a standardized way for AI agents to connect to enterprise systems and external tools.  

From chatbot to finance operating system  

In its first wave, enterprise AI largely functioned as a chatbot. Teams adopted tools like ChatGPT and Claude en masse to answer questions, summarize information, or perform simple tasks such as drafting an email. While these capabilities can increase productivity, they don’t fundamentally change how finance operates.   

In finance, data and workflows span systems including ERPs, accounts payable platforms, and procurement systems, among others. Finance needs AI tools that can work across these existing systems. Otherwise, AI assistants can make suggestions, but humans must do the heavy lifting in terms of execution. For example, an AI tool might flag an invoice discrepancy, but a human still needs to investigate what’s actually going on and address it accordingly.  

Open protocols introduce infrastructure that shifts the dynamic. Rather than treating AI as a standalone tool, open protocols allow organizations to connect AI agents to the systems where finance work happens. An AI agent can retrieve relevant data, maintain context across multiple steps in a workflow, and execute approved tasks within defined guardrails. With this infrastructure in place, AI starts to feel less like a generic chatbot and more like a digital teammate that plays an active role in finance workflows.  

Standardization supports security  

According to the Association of Financial Professionals, three-quarters of organizations experienced attempted or actual fraud in 2025. There’s no doubt that the rise in AI contributes to the uptick in fraud, as my colleague, Ottimate CFO Dan Kim, wrote in a recent article.   

In the chatbot era, security was often treated as an afterthought. But as AI agents gain access to more systems and take a more active role in workflows, security must be built into the architecture.   

Without a standardized protocol, every connection between an AI application and an enterprise system is a unique security surface with its own authentication methods, permissions, and logging requirements. When organizations are using several standalone tools, it’s hard to govern integrations consistently and easy to misconfigure them.  

Open protocols can help by creating a more standardized framework for how AI agents request and use system access. Finance leaders can define what an AI agent is allowed to see, which actions it can take, and under what circumstances it’s allowed to take them. Read-only access, time-limited credentials, and detailed activity logs can all help keep AI activity within clearly defined guardrails.  

Standardization becomes especially important as AI agents evolve from answering questions to executing tasks within finance workflows. In the past, finance leaders focused on whether a bad actor could trick a human employee into approving a phony invoice or payment. But in the new era of AI, it’s equally important to consider whether an attacker could manipulate the data, instructions, or systems an AI agent relies on.   

This doesn’t mean organizations should avoid using AI for higher value work. It simply underscores the importance of building governance into how agents connect, authenticate, execute, and leave an audit trail. Open protocols provide standardization, giving finance teams a stronger foundation for enhancing security, even as AI capabilities continue to expand and become more deeply ingrained in key business processes. 

Context changes everything  

Two finance teams can use the same AI model and have very different experiences. The difference lies in the context that’s available to the model.  

A company using a standalone, disconnected AI agent will get generic recommendations. For example, an agent can recommend an approval on an invoice. But if an AI agent understands a company’s approval policies, vendor agreements, payment terms, and historical transaction patterns, it can provide more intelligent recommendations and complete tasks that align with how the business actually operates. For example, it can automatically route the invoice in the previous example to the right approvers at the right time.  

Achieving the latter result doesn’t require swapping out tools, retraining models, or asking staff to copy and paste sensitive financial data into standalone AI tools. Instead, it requires open protocols that enable AI agents to securely retrieve the information they need from the systems where the data lives. Because change is constant, agents can always work from the most current information, rather than relying on outdated knowledge. 

This approach enables a hyper-personalized AI experience that scales across the finance industry. For example, one agent can review invoices to identify recurring exceptions based on the company’s own purchasing history. Another can provide payment recommendations in accordance with established approval rules and current vendor terms. A third can flag unusual spending patterns by comparing transactions against historical activity. Each of these agents works within the same governance framework, but the recommendations and actions it provides are tailored to the organization’s own data, policies, and workflows. 

AI is becoming increasingly embedded in finance operations, but not all organizations will have the same results. The organizations that see the greatest impact will be those that use AI that can understand the unique context of their business. Open protocols provide the infrastructure that makes that possible.  

Preparing for the next generation of AI in finance  

Getting ready for the next generation of AI doesn’t mean completely overhauling an existing finance tech stack. But it does require finance leaders to start thinking differently about how AI fits into their broader technology strategy. Before making investments, it’s important to ask the right questions to understand interoperability, governance, and long-term flexibility.

One key question to ask during vendor evaluations is how their AI connects to the rest of the business. Does it require proprietary integrations, or does it support open standards that make it easier to work across existing systems? The latter can simplify integration complexity today and make it easier to roll out new AI capabilities down the road.  

In addition, governance should always be a part of AI discussion rather than an afterthought.  The same rigor applied to user permissions, financial controls, and audit trails should extend to AI agents that can access sensitive data or actively participate in financial workflows. Starting small with clearly defined, low-risk use cases allows organizations to establish and optimize processes and controls and build confidence before expanding AI into other finance workflows. 

Finance leaders often feel the pressure to move quickly. But in the case of AI, speed doesn’t equate to success. A more effective path forward is to take the time to build the foundation that can support AI securely, responsibly, and at scale.  

Looking ahead 

Artificial intelligence is moving beyond standalone chatbots that provide generic recommendations but can’t act on them. In this new era, success depends just as much on connectivity and governance as it does on model performance. 

Open protocols lay the foundation for this evolution, enabling AI to work securely across the systems where finance work already happens. As AI becomes more embedded in finance operations, organizations that prioritize interoperability will be best positioned to unlock its full potential.