Automation, AI & Agents:
Why not every challenge needs AI
Introduction
AI is now part of most business transformation discussions, but introducing AI is not the same as improving a business process. Copying information between systems, waiting for approvals, and correcting incomplete requests are problems that clearer workflows or conventional automation can often solve without AI.
Always start with the business outcome: what needs to move faster, become more reliable, or cost less? Understanding how work currently happens makes it easier to distinguish between a process problem and one that genuinely requires AI.
Measuring Results from AI Investments
AI adoption is widespread, but measurable value remains uneven
AI adoption has accelerated, but using AI and generating measurable value from it are not the same thing. Organisations can deploy tools successfully at an individual use-case level without changing the processes that determine cost, speed or service quality. Looking at adoption alongside business impact helps understand why improving processes matter as much as the technology being introduced.
Regular AI use
Respondents reporting AI use in at least one business function
Enterprise-Level impact
Respondents reporting any enterprise-level EBIT impact from AI
McKinsey’s November 2025 State of AI survey found that 88% of organisations use AI regularly in at least one business function. Yet only about a third had begun scaling AI across the enterprise, and roughly two out of five reported any enterprise-level EBIT impact. McKinsey identified a small group of AI high performers, about 6% of respondents, combining significant reported value with at least 5% EBIT impact attributed to AI. Those organisations were nearly three times as likely as others to have fundamentally redesigned individual workflows. This is an association in survey results, not proof that redesign alone causes success.
BCG’s September 2025 research likewise separated widespread activity from realised results: it classified 5% of firms as future-built, 35% as scaling AI and beginning to generate value, 60% as reporting little material value. The categories and methodology differ from McKinsey’s, so the percentages should not be treated as interchangeable. Together they point to a question worth asking before another investment: what operational change will produce that benefit?
Where organisations stand on AI value
BCG research, September 2025
5%
35%
60%
Future-built
Scaling AI and generating value
Reporting little material value
Automation: Process vs AI vs Agentic
Choosing the right approach
The distinction is less about one being more advanced than the other and more about the type of work involved. Some activities rely on consistent rules and predictable steps; others involve information or decisions that require interpretation.
Process automation executes defined actions and decisions. A purchase request can be routed to a budget owner, checked against a spending threshold and recorded in an audit trail. Where inputs and decisions are sufficiently structured, this approach is usually easier to test, explain and control.
AI automation introduces capabilities such as extracting meaning from documents, classifying free-text enquiries or generating a suggested response. It is valuable where variation and ambiguity make fixed rules cumbersome. AI can interpret incoming supplier email and extract relevant details, for example, while a conventional workflow checks mandatory fields, routes approval, and updates the system of record. The two approaches complement one another rather than compete.
Agentic AI goes further: an AI-enabled system plans and executes multiple steps towards an objective using available tools. It may suit tasks involving changing information and decisions across several stages, but that autonomy raises the stakes. Poorly defined processes, permissions or controls have greater consequences, so testing, monitoring and escalation matter more. A sensible rule is to consider agentic approaches only once the underlying process is well understood and simpler options have been ruled out.
Gartner forecasted in June 2025 that over 40% of agentic AI projects will be cancelled by the end of 2027, citing costs, uncertain value, and insufficient risk controls as the main contributors. Gartner also cautioned that some products marketed as agentic simply repackage existing assistants, chatbots, or robotic process automation.
Prepare Before Automating
Start with the workflow, not the technology
A slow process does not necessarily need faster technology. Before choosing a form of automation, find out where time, effort, and errors enter the workflow.
Consider a customer onboarding process that regularly takes two weeks. Following the workflow from submission to completion can reveal whether the delay comes from duplicated data entry, unclear ownership, unnecessary approvals, disconnected systems, or decisions that genuinely require human judgement. Each points to a different solution, and not all of them require new technology.
Mapping the current process makes those distinctions visible and helps establish where intervention will create value. Removing duplicate checks or establishing a single owner may create immediate improvements. A workflow engine can then automate predictable routing and notifications. If staff still spend substantial time reading inconsistent documents, AI-assisted extraction may be a justified additional step. Measuring the time, error rate and rework before and after each change shows which intervention helped.
Read more on how process mapping can help you find automation opportunities here.
This emphasis on workflow design is consistent with McKinsey’s finding that AI high performers are much more likely to redesign workflows. It also reinforces a broader point: AI is more likely to create value when it is applied to a clearly understood process rather than used as the starting point for transformation.
When to Introduce AI
AI is best for unstructured inputs, complex language, or context-based decisions
Once a process has been understood and unnecessary complexity addressed, the remaining work can be assessed for areas where conventional rules are no longer enough. AI becomes more relevant when the challenge is not simply moving information or following predefined steps but interpreting information and responding to variation.
It has a clearer role where work involves genuinely unstructured inputs, complex language, or decisions that benefit from context. A service desk might use AI to suggest categories for incoming requests, identify relevant knowledge articles or prepare a response for an analyst to review, while conventional automation still manages ticket ownership, priority rules, service-level deadlines and approvals. In higher-risk workflows, AI output should be treated as a recommendation until it has passed appropriate checks.
Read more about developing a successful AI strategy here.
AI also introduces additional considerations around testing, governance, access and human oversight, and these belong in the business case, particularly in regulated or higher-risk processes.
AI governance remains a challenge
Findings from Deloitte’s 2024 generative AI survey
Cited regulatory compliance as a leading barrier to GenAI adoption
Expected full implementation of AI governance to take over a year
Deloitte’s generative AI survey found that regulatory compliance was a leading barrier, cited by 38% of respondents, while 69% expected full implementation of a governance strategy to take more than a year. These figures concern the surveyed AI-involved organisations, not every business.
Deciding Where to Invest
AI is best for unstructured inputs, complex language, or context-based decisions
Identifying a technically suitable use case does not necessarily make it a worthwhile investment. The next step is establishing what will be measured.
Once you understand the workflow, find where investment will make the greatest difference. Establish a baseline for measures such as cycle time, error rates, rework, and the cost of exceptions, then compare potential interventions against it. Some problems may be solved by removing or redesigning a step; others may justify conventional automation, AI assistance or a combination of the two. This makes the technology choice a consequence of the business case rather than the starting point.
For companies that already own workflow, integration, or process-mapping technology, the first investment may not need to be another platform. You might find more value in extending existing capabilities to manual handoffs, workarounds, and disconnected activities that still sit around them.
This is also an opportunity to assess whether your existing technology is being used to its full capability. You might already have a platform being used for a small number of processes, while other teams continue to rely on spreadsheets, email, or manual data transfer. Extending automation into those areas may offer a clearer and more measurable return than introducing another platform.
New AI tools should be assessed where interpretation, variation or unstructured information remains a genuine constraint, ideally through a focused use case with measurable success criteria.
Find out how Responsiv uses process mapping to identify inefficiencies and opportunities for further automation here.
Conclusion
Choose the simplest approach that solves the problem
There is no universal point at which an organisation should move from process automation to AI automation. The appropriate level of technology should follow the complexity of the problem. Starting with the simplest intervention that can achieve the required outcome can reduce unnecessary cost and complexity while leaving room to introduce more advanced capabilities where they have a clear purpose.
Process automation and AI automation are not competing destinations. The right approach depends on the work itself. Repeatable, rules-based activity is usually better suited to conventional automation, while AI becomes more useful where information is unstructured or interpretation is required. In many cases, the strongest design will combine the two: AI handles the ambiguity and established automation controls what happens next.
The important decision is therefore not where AI can be inserted, but where it improves an outcome that has already been clearly defined.
Join the discussion with Responsiv and IBM
Join us and discover how to identify process friction, choose between process automation and AI, and focus technology investment on measurable business outcomes.
Beyond the AI Hype: Are You Solving Problems or Buying Technology?
4 Nov 2026 | 9:00 – 14:30 | London
