Most AI transformations fail long before the first model is ever deployed.
Not because the algorithms are weak. Not because the organisation lacks data.
They fail because the discovery phase was never properly conducted.
Many years ago, while studying Earned Value Management, I encountered an insight that has stayed with me throughout my career: the probability of project success is largely determined within the first third of a project’s lifecycle. Once a project moves beyond that point, its trajectory is often already set. If the project is heading for the buffers, there is usually very little that can realistically be done to change the outcome.
This observation is reinforced by research conducted by Christensen and Heise in the early 1990s, which demonstrated that project cost and performance outcomes become highly predictable relatively early in the project lifecycle.
In simple terms, the conditions for success are established early.
AI projects are no different.
The success or failure of an AI initiative is largely determined by what happens during the discovery phase, which typically occupies the first third of the transformation effort.
For this reason, a well-structured discovery phase — ideally guided by a disciplined framework such as QMD — is critical.
Yet many organisations approach AI very differently. AI value is often assumed to emerge through experimentation, iteration, and pilot projects. Teams build proofs of concept, test models, and explore new tools in the hope that value will eventually reveal itself.
In reality, sustainable AI capability is not built through experimentation alone.
It is built through disciplined discovery.
Before an organisation can deploy AI responsibly and at scale, it must answer three fundamental questions.
1. What Is AI Expected to Achieve?
The organisation must clearly define the outcomes, benefits, and problems the AI capability is intended to address.
AI should never be pursued simply because the technology exists. It must be directly linked to strategic objectives and measurable business value.
Without a clear articulation of the business problem being solved, AI initiatives quickly become technology experiments rather than transformation initiatives.
2. Can the Organisation’s Data Support It?
AI systems require operationalised, governed, and reliable data.
To determine whether this capability exists, I typically conduct a QMD AI Data Readiness Maturity Assessment. This assessment identifies the gap between the organisation’s current data capabilities and those required to support the proposed AI initiatives.
Without this analysis, organisations frequently attempt to deploy AI on top of fragmented, poorly governed, or poorly understood data environments.
In such conditions, AI systems rarely produce reliable or trusted outcomes.
3. Does the Enterprise Operating Model Support Sustained AI Delivery?
AI systems do not end at deployment.
They require ongoing operational management through practices such as:
- MLOps
- model monitoring
- model retraining
- continuous enhancement
These activities are necessary to prevent model drift and ensure that AI continues to deliver value over time.
Without an operating model capable of sustaining these activities, AI initiatives often deteriorate shortly after deployment.
Without these foundations, AI initiatives quickly become disconnected experiments rather than components of a coherent business transformation.
The discovery phase exists to prevent exactly this outcome.
The Purpose of the Discovery Phase
The discovery phase of an AI business transformation exists to determine whether the organisation is genuinely ready to pursue AI-driven outcomes.
It answers a critical strategic question:
Are we ready to do this responsibly, reliably, and at scale?
Importantly, the objective of discovery is not to design algorithms.
The objective is to assess organisational readiness.
A properly executed discovery phase should produce three outcomes:
- A clearly defined portfolio of feasible and costed AI initiatives
- An evidence-based assessment of the organisation’s AI data readiness
- A structured transformation roadmap to close the capability gaps that would otherwise prevent successful AI delivery
Only once these conditions are understood should large-scale AI implementation begin.
Step 1 — Identify and Assess Potential AI Initiatives
Discovery begins by identifying Potential AI Initiatives (PAIs).
These represent candidate opportunities where AI could materially improve:
- decision-making
- automation
- forecasting
- customer experience
Typical examples include:
- fraud detection models
- customer churn prediction
- automated underwriting support
- intelligent document processing
- supply chain demand forecasting
However, not every idea should progress.
Each initiative must be assessed against clear criteria:
- alignment with organisational strategy
- technical feasibility
- availability and suitability of data
- expected business value
- ethical, regulatory, and operational risk
These assessments should be conducted under formal transformation governance.
At a minimum, three governance roles should exist.
Senior Responsible Owner (SRO) The executive accountable for delivering the outcomes and benefits of the AI transformation.
AI Transformation Board Responsible for prioritising initiatives, allocating investment, and managing cross-organisational dependencies.
AI Ethics and Compliance Oversight Ensures AI initiatives are lawful, ethical, explainable, and accountable.
Without these governance mechanisms, AI initiatives often emerge as isolated departmental experiments rather than part of a coordinated enterprise capability.
Discovery must therefore establish clear authority, accountability, and decision rights before transformation proceeds.
The output of this stage is the AI Initiative Dossier — an enterprise register of approved AI opportunities.
Each initiative within the dossier should include:
- clearly defined business problem
- defined use cases
- proposed technical solution
- data requirements
- estimated costs
- project brief
- prioritisation score reflecting strategic importance and alignment with organisational objectives
This step ensures the organisation is not pursuing AI for its own sake, but in support of clearly defined and measurable business outcomes.
Step 2 — Assess Enterprise AI Data Readiness
Once viable AI initiatives have been identified, the next step is to determine whether the organisation’s data environment can realistically support them.
This is where many organisations encounter their first major reality check.
AI does not fail because of algorithms.
It fails because the underlying data environment cannot sustain the AI systems being built.
The discovery phase must therefore conduct a structured AI Data Readiness Assessment.
This assessment evaluates capabilities such as:
- data governance and stewardship
- data ownership and accountability
- data quality management
- metadata and lineage transparency
- data architecture and integration
- data security, privacy, and regulatory compliance
- data operating model maturity
The objective is to establish an evidence-based baseline of AI data readiness, identifying structural weaknesses that would otherwise undermine AI outcomes.
Critically, this assessment must be evidence-based rather than perception-based.
Many organisations believe they are “data-driven” until discovery reveals the operational reality.
Step 3 — Identify the Data Readiness Uplift Gap
Once current data readiness has been assessed, the organisation must identify the data uplift required to support sustainable AI capability.
This involves understanding the gap between:
- the organisation’s current data maturity, and
- the level of capability required to support the approved AI initiatives.
Closing this gap typically requires targeted investment across areas such as:
- data governance and stewardship
- operating model design
- technology platforms
- organisational skills and capability
- delivery frameworks
- assurance and regulatory compliance
The discovery phase therefore produces a structured AI data transformation roadmap, including the estimated effort and cost required to achieve the required level of readiness.
Step 4 — Develop the AI Transformation Business Case
The final output of the discovery phase is the AI Transformation Business Case.
This document provides the evidence required for executive and board-level decision making and typically includes:
- strategic objectives the transformation supports
- market pressures and competitive threats
- operational inefficiencies or performance gaps
- regulatory and compliance drivers
- opportunities to improve customer experience
- industry trends and technology disruption
- full details of the AI Initiative Dossier and prioritised initiatives
- data readiness uplift requirements
- target AI and data operating model
- technology architecture and platform strategy
- investment and cost model
- benefits and value realisation plan
- risk assessment and assurance approach
- the overall transformation roadmap
Nothing should proceed beyond discovery without formal board approval of this business case.
Once approved, the AI transformation can move into the execution phase.
Why the Discovery Phase Matters
The discovery phase is often treated as a formality.
In reality, it is the most important stage of the entire AI transformation lifecycle.
It determines whether an organisation will:
- deploy AI responsibly
- deliver measurable business value
- scale AI beyond isolated experiments
Or whether it will repeat the pattern seen across many industries:
pilot projects, impressive demonstrations, and ultimately very little operational impact.
Successful AI transformations do not begin with models.
They begin with discovery.