From AI tools to connected workflows: The next challenge for enterprise automation

Artificial intelligence has moved surprisingly quickly from experimentation to routine business use. Employees are using AI to draft documents, analyse information and accelerate administrative work, while organisations are applying automation to areas ranging from financial processing to employee support.

Recent research from CPA Canada found organisational use of AI among the Canadian accounting profession had risen from 5 per cent in 2023 to 43 per cent in 2026, while 74 per cent of respondents said they now use AI at least weekly in their professional work. Yet widespread adoption does not necessarily amount to enterprise-wide transformation. The more difficult step is connecting individual AI capabilities so that work can move between departments, systems and people without repeatedly encountering the organisational boundaries that technology was meant to overcome.

This is an important distinction. An AI system that accelerates invoice handling can make one finance process more efficient. An employee-facing assistant can reduce the burden of responding to routine HR questions. An AI procurement tool can help staff analyse information. But business processes rarely begin and end within a single application or department.

The limits of task-by-task AI

The emerging problem is fragmentation. Businesses have generally adopted software around functions: finance has its applications, HR its platforms, procurement its systems and customer service its tools. Adding AI to each can improve specific tasks without addressing the transitions between them. This helps explain why productivity improvements do not necessarily translate into equivalent gains at company level. Recent McKinsey findings reported by Digital Journal indicate that 80 per cent of respondents said AI had improved productivity, while just 37 per cent of organisations reported a direct impact on earnings.

The challenge, therefore, is moving from task automation to workflow transformation. This also provides context for a Gartner finding supplied with the research for this article that only 22 per cent of organisations have successfully scaled AI across multiple business units or adopted an AI-first approach.

Other evidence nevertheless points towards the same scaling challenge. Research involving 801 C-suite executives at large US companies found that respondents’ organisations used an average of 183 AI tools, while 85 per cent said they lacked systems capable of connecting their technology. Only 15 per cent of AI investments were reported as operating as a unified system with shared intelligence across the organisation. The study was conducted by WSJ Intelligence and Code and Theory. More AI, in other words, does not automatically produce a more intelligent enterprise.

Enter AI orchestration

One answer is AI orchestration, an approach intended to coordinate different AI capabilities, software systems and workflows rather than leaving each to function independently. The concept becomes especially important as AI develops from systems that generate an answer to what is often termed agentic AI: software designed to take or coordinate actions towards an objective.

Instead of asking an AI tool to summarise a supplier document, for example, the longer-term ambition is for AI systems to support processes extending through different stages and applications while retaining appropriate human control. This potentially changes how organisations think about automation. Rather than asking ‘Where can AI save somebody ten minutes?’, companies can start with a wider question: ‘What needs to happen for this business outcome to be completed?’

The distinction matters because the inefficiency in a process may be found between activities rather than within an individual activity. Making one five-minute task instantaneous delivers limited value if the work subsequently waits days for information, approval or transfer into another system. This is one reason AI agents are attracting enterprise attention. As Hexaware Technologies’ Siddharth Dhar recently told Digital Journal, embedding agents within enterprise applications could mean employees spend less time manually navigating workflows and more time supervising results, resolving exceptions and improving processes.

Connecting the enterprise creates new risks

There is, however, another side to orchestration. Giving AI access to a larger portion of a business process potentially increases the consequences when something goes wrong. Connecting finance, procurement and HR systems is fundamentally different from providing an employee with a standalone writing assistant. Businesses consequently need to think about permissions, information security, data quality, accountability and the point at which human authorisation remains necessary.

This problem becomes more significant as AI moves from recommending something to doing something. Research into AI-assisted workplace decision-making found that 63 per cent of surveyed professionals said AI played a role in workplace decisions even where its use was not formally tracked or documented. Meanwhile, 55 per cent said multiple AI systems were commonly involved within a single workflow, highlighting the difficulty of establishing how an outcome was reached when several systems contribute to it.

Governance is also struggling to keep pace with adoption. In the CPA Canada study, 67 per cent of respondents reported having access to approved AI tools, but only 45 per cent reported a formal AI policy and 34 per cent said their organisation had introduced either employee training or a formal AI strategy.   For regulated businesses in particular, efficiency cannot be separated from traceability and accountability.

From assistants to business infrastructure

This points towards a change in how enterprises may need to measure AI. Counting licences, chatbots or individual use cases provides an indication of deployment, but says little about whether the technology has altered the way the organisation operates. Measures such as process cycle times, error rates, exception rates and measurable business outcomes can provide a more useful picture.

There is also a human factor. Connected AI does not necessarily mean removing people altogether. As systems perform more routine work, employees may increasingly be required to supervise outputs, manage exceptions and exercise judgement where circumstances fall outside automated processes. The emerging model therefore looks less like unrestricted machine autonomy and more like a redistribution of work between software and people. The next phase of enterprise AI may consequently prove less spectacular, but more significant, than the first. Generating text and accelerating individual tasks demonstrated what the technology could do. Connecting those capabilities across real business processes will test whether organisations can turn that technical potential into durable operational improvement.

The competitive question is shifting accordingly. It is no longer simply how much AI a company has, but how effectively the organisation can make its AI, data, applications and people work together.