Building AI literacy across your organisation is one thing. Getting your team to actually apply it is quite another.
Walk into most offices today and you’ll find a curious pattern. Staff have access to Claude, ChatGPT, Copilot, or whatever AI tools the company has approved. They’ve sat through the training sessions. They’ve nodded along to presentations about AI’s transformative potential. Yet when you ask them what they’re actually using AI for, you get kind of vague answers. They mention “writing emails” or “brainstorming ideas.” These are argubaly really useful and I’m not knocking this usage of them
The application gap isn’t a knowledge problem. It’s a recognition problem.
The invisible opportunities
Your finance team is still manually reconciling invoices because they don’t realise AI could flag anomalies in seconds. Your HR department is crafting job descriptions from scratch when AI could analyse successful past postings and current market language. Your operations manager is building project timelines in exactly the same way they did five years ago.
These aren’t failures of capability. They’re failures of imagination constrained by habit.
Research from MIT’s Centre for Collective Intelligence found that employees typically use AI for less than 20% of the tasks where it could meaningfully contribute. The bottleneck isn’t tool access or technical skill. It’s the ability to see their own work differently.
Why training doesn’t translate to action
Traditional AI training focuses on what the technology can do (there are some great companies out there – Section AI, Pedagogue even some of the own brand training from Anthropic or Sanity is great). But these are more often than not – Here’s how to write a prompt. Here’s how to refine outputs. Here’s a list of use cases from other companies. Employees leave these sessions intellectually informed but practically uncertain.
The missing piece is task decomposition – the ability to look at your actual work, break it into constituent parts, and identify which elements AI could handle. This skill isn’t taught in conventional training because it requires deep familiarity with specific roles and workflows.
Consider a procurement officer who spends hours researching potential suppliers. Standard AI training might mention “research assistance” as a use case. But that doesn’t help them recognise that AI could specifically:
Extract key requirements from internal stakeholder emails. Generate search criteria based on previous successful supplier relationships. Summarise vendor websites against specific compliance requirements. Draft comparison matrices from gathered data.
The procurement officer doesn’t need to know AI can help with research. They need to recognise these discrete, AI-suitable tasks within their existing workflow.
Nearly eight in ten companies have deployed gen AI in some form, but roughly the same percentage report no material impact on earnings.1 We call this the “gen AI paradox.”
McKinsey, Seizing the agentic AI advantage, June 2025
The pattern recognition deficit
Humans are exceptionally good at executing familiar processes. We’re considerably worse at stepping outside those processes to analyse them objectively. Your team runs on autopilot through established workflows, rarely pausing to question whether each step still requires human intervention.
This automation blindness affects everyone, from junior staff to senior leadership. A managing director might spend an hour crafting a nuanced client proposal – appropriate for AI assistance – but never consider it because “client communication is too important to delegate.”
The irony is that the most AI-resistant tasks are often the most suitable for AI augmentation. People protect high-stakes activities from AI precisely because they matter, not recognising that AI could elevate the quality whilst freeing human attention for genuine strategic thinking.
Building recognition capability
Several organisations have cracked this challenge by shifting their approach from education to observation. Rather than teaching what AI can do, they’re helping employees see what they actually do.
Unilever introduced “work shadowing” exercises where employees spend one day documenting their tasks in 15-minute increments. They then review this log with a facilitator trained to spot AI opportunities. The revelation for most participants isn’t what AI can do – it’s recognising how much of their day involves pattern-based work suitable for AI assistance.
The accounting firm PwC takes a different approach. They’ve created “AI spotters” – employees who’ve developed strong pattern recognition for AI-suitable tasks. These spotters don’t implement AI solutions. They simply help colleagues see their work through an AI-opportunity lens, asking questions like “What parts of this task feel repetitive?” and “Where do you find yourself looking up similar information?”
The consultancy McKinsey embeds brief “AI opportunity moments” into existing team meetings. Before discussing solutions to problems, teams spend five minutes considering: “If we had perfect AI assistance, which parts of this challenge would we hand off?” This reframes problem-solving to explicitly include AI capabilities.
The task inventory exercise
The most effective method for building recognition capability involves systematic work analysis. Have teams create comprehensive inventories of their regular tasks, then categorise each according to three criteria:
Information processing intensity – does this task primarily involve reading, analysing, or synthesising information? These tasks typically suit AI assistance well.
Pattern recognition requirements – does this task involve identifying similarities, differences, or trends across data sets? AI excels here.
Creative constraint balance – does this task require creativity within defined parameters? AI can handle constrained creativity (like writing to specific formats or styles) whilst humans tackle unconstrained creative challenges.
Tasks scoring highly on one or more criteria become immediate candidates for AI experimentation. This exercise doesn’t require AI expertise. It requires honest task analysis.
From recognition to experimentation
Identifying opportunities is only valuable if it leads to action. The gap between recognition and implementation is where many AI initiatives stall. Employees see possibilities but lack confidence to experiment.
Successful organisations create explicit permission structures for AI experimentation. Atlassian introduced “AI prototype Fridays” where teams spend one hour testing AI applications for tasks they’ve identified as suitable. No business case required. No approval process. Just structured time for trying things. Atlassian are doing some really interesting things in terms of applying AI – for both internal and external stakeholders.
The results of these experiments aren’t always successful. Many experiments fail. But the learning compounds. Teams develop intuition about which tasks genuinely benefit from AI assistance and which remain better suited to traditional approaches.
The managerial blindspot
Middle managers face a particular challenge with AI application. Their work involves complex judgement calls, relationship management, and strategic thinking – areas where AI’s value is less obvious than in administrative tasks.
Yet managers who don’t develop their own AI recognition capability can’t effectively guide their teams. They become bottlenecks, unable to evaluate when team members should use AI and when they shouldn’t.
The solution isn’t sending managers to more training. It’s encouraging them to document their own decision-making processes. What information do they consider? What patterns do they look for? What factors weigh into their judgements? This documentation reveals opportunities for AI to handle information gathering and initial pattern analysis, elevating managers to focus on genuine judgement.
Last week, my team had confusion about how to use AI for meetings. Everyone had different ideas and didn’t understand the different packages in Microsoft Teams. Some team members thought meeting recording was only available with Teams Premium, but it’s included in our M365 license.
Measuring application, not just literacy
Organisations tracking AI adoption typically measure the wrong things. They count training completions, tool access rates, or employee confidence surveys. These metrics indicate literacy but not application.
Better metrics focus on recognition and implementation:
How many discrete AI use cases has each team identified in their workflows? How many have they tested? What percentage of identified opportunities are now part of regular practice?
These questions reveal whether AI literacy is translating into actual behaviour change. They also highlight which teams are developing strong recognition capability and which need additional support.
The compounding effect of application
When employees begin recognising AI opportunities in their own work, something interesting happens. They start seeing opportunities in adjacent areas. The marketing coordinator who uses AI to analyse campaign performance begins suggesting ways the sales team could use similar analysis. The operations manager who automates schedule creation shares insights with the project management office.
Recognition capability spreads through informal networks more effectively than through formal training. Employees who’ve successfully applied AI become credible advocates, helping colleagues see possibilities in their own workflows.
This organic spread creates momentum that top-down mandates never achieve. Within months, organisations shift from asking “How do we get people to use AI?” to managing an abundance of implementation ideas.
Beyond the obvious applications
The recognition problem is particularly acute for non-obvious AI applications. Everyone understands AI can write emails or summarise documents. Fewer recognise that AI can:
- Identify gaps in project plans by comparing current timelines against historical data.
- Detect subtle shifts in customer feedback sentiment over time.
- Generate test scenarios for new processes based on edge cases in existing workflows.
- Suggest meeting agenda items based on email patterns and calendar context.
These applications require seeing your work as data and processes rather than as indivisible activities. They require stepping back from doing work to analysing how work gets done.
The recognition imperative
Your organisation’s AI literacy programme might be excellent. Your tool access might be comprehensive. Your leadership might be supportive. But without recognition capability, your team will continue working exactly as they did before, using powerful AI tools for superficial tasks whilst missing transformative opportunities.
The good news is that recognition is a learnable skill. It doesn’t require technical expertise or deep AI knowledge. It requires curiosity about your own work, permission to experiment, and structured opportunities to see familiar tasks through fresh eyes.
Ten minutes of AI learning each day builds literacy. Ten minutes of work analysis each week builds recognition. Together, they transform AI from an interesting capability into a practical advantage woven throughout how your organisation operates.
The question isn’t whether your team knows about AI. It’s whether they can see their own work clearly enough to recognise where AI belongs.
