Perspective · AI Readiness
June 2026 · 4-minute read · Asseblia Insights
Most AI consulting conversations start with the technology. Which model? Which vendor? What’s the architecture? These are reasonable questions, but they’re the wrong starting point. The question that actually determines whether an AI project succeeds is simpler and less exciting: is your organisation ready to change the process the AI is supposed to automate?
42% of AI initiatives launched in 2025 were discontinued before reaching production. The technology worked in most of those cases. The project failed because of organisational factors that had nothing to do with the LLM.
What ‘readiness’ actually means
AI readiness isn’t a binary state. It’s a profile across four dimensions:
Data readiness
Do you have the data the model needs, at the quality level it requires, accessible in a form that can be used?
Process clarity
Is the process you want to automate well-defined, consistently executed, and documented well enough to specify what ‘correct’ looks like?
Ownership clarity
Is there a named person who is accountable for the outcomes of this system after deployment, including when it gets something wrong?
Change readiness
Are the people whose workflows will change genuinely willing to change them, or will they route around the AI system the moment it produces one wrong answer?
Most organisations score well on data readiness (they overestimate it, but at least they’ve thought about it) and poorly on ownership and change readiness. Those are the failure modes that don’t show up in the pilot.
The data readiness illusion
‘We have all the data’ is the sentence that precedes most AI project failures. Companies have data. What they typically don’t have is data at the quality, granularity, consistency, and accessibility level required for a production AI system.
Common discovery surprises: the ‘CRM data’ is 40% incomplete because sales reps don’t fill in fields they don’t use; the ‘transaction records’ are in three different systems with different schemas that have never been reconciled; the ‘historical reports’ exist as PDFs rather than structured data; the API that should expose the data requires a security review that will take three months.
The fix isn’t to pretend the data is better than it is — it’s to scope the initial use case to the data that actually exists and is actually accessible. A smaller use case that works beats an ambitious use case that doesn’t.
The process clarity gap
You cannot automate a process that isn’t defined. This sounds obvious, but it’s consistently underestimated. Most knowledge work processes are a mix of documented procedure and tribal knowledge — and the tribal knowledge is often doing more of the work than the documentation suggests.
Before starting any AI implementation, we run a process mapping exercise that asks: what does the ‘correct’ output look like for this task? Who decides whether an output is correct? What are the edge cases, and how are they currently handled? If you can’t answer these questions, you can’t write a system prompt that reliably produces what you want.
The ownership problem
Pilots have sponsors. Production systems need owners. The sponsor who approved the budget for your AI project is probably not the person who will be accountable for maintaining prompt versions, reviewing flagged outputs, and deciding when to escalate to a human. That person needs to be named before deployment, not after.
Ownership means: someone who understands what the system is doing, monitors its performance, manages the escalation queue, and has the authority to pause the system if it starts behaving incorrectly. Without a named owner, the system degrades silently until someone notices it’s been wrong for three months.
How to assess your own readiness
Before starting an AI project, answer these questions honestly:
→ Can you produce a written specification of the process you want to automate, including edge cases?
→ Is the data you need available, structured, and accessible via API today — not ‘after we clean it up’?
→ Is there a named person who will own this system in production, including when it fails?
→ Have you shown the system’s expected outputs to the people who will use them, and confirmed they will actually use them?
If you can’t answer yes to all four, the project isn’t ready to start.
“That’s not a bad outcome — it’s better to know before you spend six months building something nobody adopts.”
Next step
Not sure if your organisation is AI-ready?
Book a 45-minute discovery call — we’ll map one high-impact automation opportunity. No commitment required.