The Vendor Is Not the Problem, the Process Is, Why Seventy Percent of Enterprise AI Briefs Change in a Two-Hour Discovery Session

The headline finding from TechRadar

TechRadar published a sharp piece on 6 June 2026. The headline finding, the single most common enterprise AI mistake is starting with the tool rather than the outcome. Organisations evaluate solutions based on vendor reputation and feature sophistication, without anchoring the decision in a clearly defined, prioritised business problem. The Addepto research from April 2026 sharpened the same point, misaligned AI can produce plausible outputs for months before business outcomes fail to appear.

The two findings together describe a pattern that is happening at scale across the UAE, Nigeria, Kenya, Tanzania and Ethiopia right now. Procurement runs a thorough RFP. The vendor with the best demo and the strongest documentation wins. The pilot is designed to validate the decision already made. The pilot passes. Production runs for three quarters. The business outcomes that justified the budget never appear. Nobody calls the pilot what it was, a demonstration rather than a test.

The three failure points

Across more than fifty enterprise AI vendor selections we have reviewed in the GCC and Africa, the failure is concentrated in three places.

• Failure point one, the brief is written before the problem is mapped. Enterprises issue RFPs based on an internally drafted brief that has never been stress-tested against actual operational data. Sixty to seventy percent of those briefs change materially in a two-hour facilitated discovery session. The vendor selected against the original brief is often the wrong fit for the real problem.
• Failure point two, compliance and procurement are confused with evaluation. Procurement is designed to reduce risk and ensure fairness. It is not designed to evaluate whether a specific AI system will solve a specific business problem in a specific operational context. A vendor who wins the RFP has demonstrated documentation, references and pricing structure. None of those predict the business outcome.
• Failure point three, the pilot is designed to succeed, not to learn. Once a vendor is selected, internal champions design the pilot to validate the decision already made. Metrics are chosen that the tool will perform well on. Edge cases are excluded. The pilot passes. Production deployment begins. Three months later, the business outcomes that justified the budget are still absent.

The cost of buying in the wrong sequence

The cost of the wrong sequence is not a single line item. It is five layers of compounding loss. Vendor contracts signed before the business problem is validated. Pilots that produce reassuring metrics and no business impact. Production deployments that run for quarters without P and L movement. Leadership credibility eroded when boards ask for results that were never defined. Switching costs averaging USD 315,000 per project when the wrong vendor is replaced, before counting the opportunity cost of the eighteen months lost.

Multiply that across the four or five enterprise AI vendor selections an organisation typically makes in a year, and the total cost of the wrong sequence comfortably exceeds USD 1.5 million per year for a mid-sized enterprise. That number is also conservative because it does not include the regulatory and reputational consequences of an agent acting on the wrong problem.

The fix, a better sequence

The fix is not a better vendor. The fix is a better sequence. Define the business problem first. Validate it with a structured discovery session. Sketch a vendor-agnostic solution architecture. Then, and only then, invite vendors to respond to the problem and the architecture, not to a feature list. That reversal alone recovers the majority of enterprise AI spend that is currently producing nothing.

• Step one, two-hour facilitated discovery session with the business owner, the executive sponsor and a senior AI practitioner. Pressure-test the brief against operational data, customer impact and reversibility.
• Step two, write a one-page problem statement with measurable outcome targets, named human owner, baseline metrics and target metrics.
• Step three, sketch a vendor-agnostic solution architecture that names the data flows, the system boundaries and the governance controls. This is a sketch, not an engineering specification.
• Step four, write the RFP against the problem statement and the architecture. Ask vendors to respond to the problem, not to a feature list. Score responses on business outcome credibility, not on documentation polish.
• Step five, design the pilot to learn, not to succeed. Include edge cases. Define the criteria for a failed pilot as carefully as the criteria for a successful one. Reserve the right to walk away without penalty.

What the boards in the GCC and Africa are now asking

Across the UAE, Nigeria, Kenya, Tanzania and Ethiopia, the boards that lead on enterprise AI are now asking a different set of questions before any contract is signed. Has the problem been mapped by a structured discovery session, or by an internally drafted brief. Is the pilot designed to learn or to validate. What is the rollback path if the vendor is the wrong fit. What is the documented switching cost if we need to replace this vendor in eighteen months. Who in the leadership team has the literacy to interpret the pilot results.

Boards that ask those five questions consistently land on better vendor decisions. Boards that do not are paying the USD 315,000 switching cost on roughly half of their enterprise AI selections.

How the Applied AI MasterClasses build the discovery capability

The Generative AI for CXOs and Business Leaders MasterClass teaches the four-step sequence as a working framework, with worked examples from BFSI, telecom, retail and the public sector across the GCC and Africa. The Applied AI and Predictive Analytics MasterClass equips business leaders to interpret vendor pilot results critically, including statistical significance, edge-case coverage and operational realism. The Adaptive Leadership in an AI-Accelerated Business Environment MasterClass prepares the executive committee to lead a discovery culture rather than a feature-procurement culture. The Enterprise AI Readiness Assessment Audit gives the board the baseline score that determines where the discovery capability gap sits today.

Cohorts run virtual on July 16 to 18 and August 13 to 15 2026, and onsite on July 23 to 25 and August 19 to 21 2026. Early Bird pricing of USD 650 is open until 30 June 2026.

Five actions before your next vendor selection

First, take the Enterprise AI Readiness Assessment Audit and capture the Strategy and Discovery pillar score. Second, run a two-hour facilitated discovery session on the next vendor selection before the brief is written. Third, write a one-page problem statement with measurable outcome targets. Fourth, design the pilot to learn rather than to validate, including edge cases. Fifth, reserve seats in the July or August 2026 Applied AI MasterClass cohort before Early Bird closes on 30 June 2026.

Recommended further reading

1. Enterprise AI Readiness Assessment, CXO and board audit playbook.

2. Microsoft Build 2026, agent-to-agent goes live and the trust question.

3. Generative AI for CXOs and Business Leaders MasterClass.