The strongest AI opportunity in your organisation may be hiding inside a process people already know is inefficient, tedious and expensive. These processes often have a clear baseline, repeated manual work and experienced employees spending time on tasks that do not require their expertise. McKinsey’s 2025 research found that workflow redesign had the strongest relationship with EBIT impact from generative AI, yet only 21% of organisations using generative AI had fundamentally redesigned at least some workflows. Before choosing your next AI project, look at the work your best people would most like to stop doing.
When you decide to invest seriously in AI, where do you look first?
Often, you look for something visible. Perhaps a customer-facing chatbot, an AI assistant for sales, or an internal copilot that can make for an impressive demonstration. These projects can have genuine value, particularly when they address a clear business need. But visibility is a poor way to identify where AI can create the greatest business impact.
Your best opportunity may be sitting somewhere far less exciting.
It could be the spreadsheet that gets rebuilt every month. The reconciliation exercise that requires data to be copied between systems. The approval process that involves six emails and three reminders. Or the report that takes a senior analyst an entire afternoon to prepare before they can spend five minutes interpreting it.
These are the processes people complain about repeatedly, yet somehow continue to run.
That makes them worth examining closely. Your worst process may contain the clearest business case for AI because its inefficiency is already visible, measurable and understood by the people doing the work.
Why The Most Visible AI Projects Can Miss The Mark
There is a natural temptation to begin with an AI project that demonstrates what the technology can do.
A customer-facing application feels strategic. An AI sales assistant feels modern. A sophisticated internal chatbot can generate plenty of interest. These projects can also attract executive attention because their value is easy to see.
The problem is that visibility does not necessarily translate into measurable value.
Customer-facing AI applications operate in environments where accuracy, brand reputation, security and customer experience matter immediately. That means extensive testing and governance are often required before the solution can be used at scale.
There is another issue. When you begin with the technology rather than the process, you can end up asking, “Where can we use AI?” instead of asking, “Where is the business losing time, money or expertise?” That distinction matters.
McKinsey’s 2025 global AI research found that, among 25 organisational factors studied, redesigning workflows had the strongest relationship with EBIT impact from generative AI. Yet only 21% of organisations using generative AI said they had fundamentally redesigned at least some workflows.
The implication is important. You can introduce AI into an existing workflow and still leave the underlying process largely unchanged. The greater opportunity often comes from reconsidering how the work should happen once AI becomes part of it.
What Makes A Bad Process A Good AI Candidate?
A process does not become a good AI opportunity simply because employees dislike it. You need to look for a particular combination of characteristics.
The first is repetition. If the same activity happens every week or month, even a modest reduction in effort can accumulate into significant value.
The second is variability. Traditional automation works particularly well when every input follows the same rules. Many business processes sit somewhere between rigid and completely unstructured. They involve documents, emails, exceptions, interpretation and decisions based on context. That is where AI can add a different layer of capability.
Consider a finance team that receives invoices from hundreds of suppliers. The information may appear in different formats, descriptions may vary, and exceptions may require human review. A conventional rules-based system may struggle with the variation. An AI-enabled workflow can extract information, classify documents and route exceptions to the appropriate person, while leaving higher-risk decisions with the finance team.
This is where GenAI for business processes becomes particularly relevant. The value does not come from putting a chatbot on top of an existing process. It comes from changing how information moves through that process and deciding where AI should perform a task, where software should handle it, and where a person still needs to make the decision.
Start With The Work Your Best People Hate
One of the simplest ways to find valuable AI use cases is to ask where your most capable people are spending time on work that does not require their expertise.
This is a more useful question than asking employees where they would like to use AI.
Imagine a senior analyst spending six hours every week cleaning and reformatting data before they can begin analysing it. The problem is not that the analyst needs to work faster. The problem is that the organisation is paying for analytical expertise and consuming it on data preparation.
It also gives you a useful starting point for AI opportunity identification because the cost of the problem is already visible. You can establish a baseline using the hours spent, frequency of the activity, number of people involved and downstream delays created by the process.
You can then ask a small group of experienced employees a straightforward question:
“What is the most repetitive part of your week that has little to do with why you were hired?”
Ask people across different functions and listen for recurring patterns. You will often find that the same frustrations appear in different forms.
The sales operations team may be manually updating CRM records. Marketing may be consolidating campaign data. Finance may be reconciling information across systems. HR may be checking and categorising employee requests.
The details differ. The underlying signal is the same: skilled people are spending valuable time moving, checking, formatting or interpreting information that could potentially be handled differently.
Measure The Pain Before Choosing The Technology
Once you identify a problematic process, resist the urge to select a tool immediately.
First, understand the process.
How many people touch it? How often does it happen? Where does information enter the workflow? Where is it copied, checked or re-entered? Which steps depend on individual knowledge? Where do errors occur? How long does the process take from start to finish?
Most importantly, what happens when the process fails or gets delayed?
This gives you something much more useful than an interesting AI demonstration. It gives you a baseline against which you can measure value.
Suppose a monthly reporting process takes 40 hours across a team. Perhaps the work involves collecting information from several systems, checking inconsistencies, formatting the output and preparing a narrative for leadership.
You might discover that AI can handle the first stages, while a person reviews exceptions and validates the final interpretation. The opportunity is then measurable. You can compare the redesigned workflow against the original process and track time saved, error rates, turnaround time and quality.
That is a much stronger starting point for AI process automation than simply asking which AI tool your organisation should buy.
Redesign The Process Before You Automate It
There is a trap here. If you automate a bad process without questioning its design, you may simply make a bad process run faster.
This is why workflow redesign matters.
Break the process into its individual activities and examine what actually needs to happen. Some steps may disappear altogether. Others may be handled by software. Some may become AI-assisted. A small number may still require human judgement.
This creates a different conversation around AI.
Instead of asking, “How can AI perform this task?”, you ask, “What should this workflow look like if AI can reliably perform part of it?”
That question can lead to a very different operating model.
For example, imagine a customer support process where every request is manually read, categorised and assigned to an employee. An AI-enabled redesign could classify incoming requests, identify priority, retrieve relevant customer information and recommend a response. A human then handles exceptions, sensitive cases and decisions requiring judgement.
The result is not simply an automated task. It is a redesigned flow of work.
Your AI implementation strategy should therefore begin with the process and the desired business outcome, with technology selection following from that assessment.
Choose AI Use Cases That Can Prove Their Value
A strong first project does not need to be the most sophisticated one, but it needs to produce evidence.
That is why the best AI use cases often have a clear before-and-after comparison. You know how long the process currently takes. You know how many people are involved. You can identify error rates, bottlenecks and delays. After implementation, you have something against which to measure the result.
If your first AI initiative reduces a painful process from three days to several hours, the people who experienced that improvement become credible advocates. They can explain exactly what changed because they lived with the problem before the solution existed.
That kind of evidence travels further internally than an impressive demonstration.
It also gives leadership a better basis for deciding where to invest next. One successful workflow can reveal another adjacent process that suffers from the same data, approval or information bottleneck.
The Process People Complain About May Be Your Best Starting Point
The next time someone describes a process as “the way we’ve always done it”, pay attention.
The process everyone avoids discussing may contain more AI potential than the project currently receiving the most attention. Look at where experienced people spend time copying information, checking documents, reconciling systems, preparing repetitive reports or searching for information they should already have.
Then ask a more useful question: what would this process look like if you redesigned it around the capabilities you now have?
AI adoption does not have to begin with the most impressive application. It can begin with a process that everyone knows is inefficient, where the cost is understood, and the people doing the work are ready for a better way.
Your worst process may already be telling you where to start. You just need to listen to the people who deal with it every week.
How XITE Create Can Help
At XITE Create, we help organisations move from isolated AI ideas to practical business applications grounded in real processes. Our approach brings together business understanding, AI capabilities, technology architecture and workflow thinking to identify where AI can create measurable value.
Whether you are assessing your first AI opportunity or looking to redesign an existing process around AI, our team can help you evaluate the workflow, identify the right intervention and build a practical path from opportunity to implementation.
