Avoiding common pitfalls when deploying AI
Most AI projects stall for fixable reasons. Three operational problems explain the gap between adoption and value.
The gap between adoption and value
Most organisations have investigated generative AI. Fewer than one in 20 have it meaningfully integrated into their workflows.
That is an operations problem.
MIT researchers call it the Learning Gap. When AI tools forget corrections and keep producing first-draft output, teams spend more time fixing than creating. The result is visible adoption with invisible value. People use the tools. The metrics don't move.
The number that matters is how many workflows have been redesigned around AI. McKinsey found that the top-performing 6% of organisations achieving real margin impact from AI share one trait: they redesigned their workflows. 55% of high performers did this, compared with 20% across the broader market.
The gap is not technology. It is how the work is organised around the technology.
Three failure modes
When AI projects stall, the cause is almost always one of three things. Recognising which one you are dealing with helps you determine the right response.
1. You chose the wrong starting point
The best starting points have clear inputs, repeatable steps, and measurable outputs. If you automated a process that relies on tacit knowledge, undocumented exceptions, or cross-team negotiations, the tool was never going to work. The workflow was not the right place to start with applying AI.
2. You automated a broken workflow
Your process documentation says: Brief, Work, Review, Deliver. The actual process is: Brief, Clarification, Re-brief, Partial work, Scope creep, Fire drills, Rework, Stakeholder loops, Version chaos, Delivery.
If you automate the workaround, you scale dysfunction.
Your real process lives in someone's browser history, in Slack DMs, in the spreadsheet called FINAL_v5. When the documented process and the actual process diverge, organisations run on goodwill. Handoffs break. Metrics lie. Rework spikes.
Before any automation, watch how work actually gets done. Count the browser tabs. Document the workarounds. That is the process you need to fix.
3. You didn't equip the team
AI tools arrive without operating instructions. Teams receive access but not context: which tasks to use them for, what good output looks like, how to review AI-generated work, or when to override it.
The result is a tool that sits unused after week two. BCG research suggests that while most organisations have adopted AI tools, fewer than one in 20 are creating real value from them. Without context for how to use the tools well, adoption stalls.
Between paralysis and chaos
Beyond individual workflows, organisations face two traps.
Paralysis
Waiting for the perfect strategy, the perfect tool, the perfect moment. Nothing ships. Months pass.
Chaos
Starting ten projects simultaneously with no coordination or measurement. Everything ships. Nothing lands.
The answer sits between them. Progress with intention.
Start where three conditions intersect:
High impact on a measurable outcome
Low complexity to implement
High learning value for the team
One focused project beats ten half-started ones.
The right first workflows share common traits
Clear inputs. Repeatable steps. Measurable output. Examples:
- Weekly reporting
- Pipeline updates
- Meeting preparation
- Proposal drafting
- Client onboarding documentation
Pick one workflow. Redesign it end-to-end. Measure the result. Learn from it. Then pick the next one.
What “fix work first” actually means
Fixing work means a focused diagnostic on how value actually flows through your organisation.
Start with one critical value flow. Pitch to retainer. Brief to delivered work. Intake to matter close. Map it as it actually happens.
Look for five things:
- Time spent on work about work. Coordination, chasing approvals, reformatting, copying data between systems. Capacity that creates no client value.
- Rework and revision rates. How many passes does a deliverable take before it is right? Where do errors enter?
- Handoff friction. Where does work stall between people or teams? What information gets lost in the transition?
- Scope creep patterns. Where does agreed work expand without corresponding adjustment to timelines or pricing?
- Decision bottlenecks. Where does one person's availability hold up an entire workflow?
Quantify these. Put hours and pounds against them. That creates the business case for change and tells you precisely where AI and automation will create the most value.
The operating architecture beneath your tools
AI delivers durable value only when your pricing model, your delivery workflows, and your technology work together. Most firms invest in the technology and hope it redesigns the rest. It does not.
Start with one workflow
You don't need a company-wide AI strategy to make progress. You need one workflow, diagnosed honestly, redesigned thoughtfully, and measured rigorously. We will identify which one to start with.