By Lee Wearne, Operations Director at Insight Consulting.
In nearly every client engagement we’re in right now, some version of the same story surfaces: AI tools have been adopted from the bottom up, faster than governance has come from the top down. The productivity gains are real and visible, but what’s less visible is the gap between what’s being generated and what the business can actually stand behind.
Teams are generating content, writing code, building dashboards, and automating workflows faster than ever before. The key question that’s going unanswered is – “Is any of this actually ready for the business to rely on?”
This is the conversation we’re having with leadership teams across industries right now. The question is “How do we use AI well, safely, and in a way that actually holds up when and where it matters?” The gap between AI-generated and enterprise-ready is wider than most organisations realise, which is why AI strategy is a key focus area for most of our customers today.
Most companies don’t yet have an AI strategy. They have AI experiments. A bit like data in the old days, where everyone was downloading spreadsheets, doing their own filtering and analysis, and sharing numbers in meetings which didn’t always match the numbers that others were sharing. Data strategy helped to improve this situation, by focusing organisations on governed data platforms, data stewardship, data literacy, appropriate self-service analytics tool choices and data councils to lead the way to a single source of truth available for everyone to use when and where they needed it.
Whilst many organisations are still bedding down their data strategies, they’ve now had to look towards AI strategies in parallel. Fortunately, the two are complementary. For AI to work well, solid data foundations are a must.
In our work with organisations across many sectors, we’ve consistently found that AI adoption stalls or creates risk when three things are absent: strategic intent, digital foundations, and maturity awareness. The fourth thing that’s increasingly urgent, and in our view, getting less attention than it deserves is governance. Specifically: what your organisation’s AI policy actually says, what tools are sanctioned and why, and who is accountable for the outputs AI produces.
The AI policy conversation organisations are avoiding
Most organisations have defaulted to one of two positions on AI tools: blanket permissiveness (“use whatever helps you”) or blanket restriction (“don’t use anything until we figure this out”). Both are strategic failures.
Blanket permissiveness means staff are feeding sensitive client data, financial information, and proprietary processes into consumer-grade AI tools – tools that, in their free or standard tiers, may use that data for model training, offer no data residency guarantees, and provide zero contractual accountability. This isn’t hypothetical – it’s happening now, in most organisations, with no visibility and no audit trail.
Blanket restriction, on the other hand, doesn’t stop AI adoption – it just drives it underground – shadow AI. Staff find workarounds. The tools get used anyway, just without IT’s knowledge and without any of the safeguards that a structured approach would provide.
A mature AI policy sits in neither camp. It draws a deliberate distinction between tools that are appropriate for different contexts: consumer tools for low-stakes, non-sensitive tasks; enterprise-grade platforms with proper data agreements, audit logs, role-based access, and compliance certifications for anything that touches business-critical or sensitive information.
Consumer AI tools
- Appropriate for drafting personal communications, brainstorming, public research, personal productivity.
- Not appropriate for client data, financial analysis, proprietary processes, regulatory-sensitive content.
Enterprise AI platforms
- Data isolation, contractual privacy guarantees, audit logging, role-based access, compliance certifications (ISO 27001, SOC 2, GDPR).
- Required for any AI-assisted work involving client, financial, or operational data.
More than just a security decision, the investment in enterprise AI is a trust decision. It’s what allows you to deploy AI-assisted capabilities to clients and stakeholders without the caveat that it was “just AI-generated” rather than a deliverable you stand behind.
AI-generated vs. enterprise-ready
There is a version of AI productivity that looks impressive and moves fast, and there is a version that is actually reliable. They are not the same thing. The difference is human judgment applied at the right points in the process. Here are some real-world examples:
REAL-WORLD EXAMPLES
Vibe coding a platform vs. shipping production software
AI coding assistants can generate a functional-looking web application in hours. The code often works – in a demo environment, on clean data, under no load. What it typically lacks is proper error handling, security hardening, scalability testing, accessibility compliance, and the kind of architectural decisions that prevent systems from collapsing under real-world conditions.
The AI accelerated the build phase dramatically. But the engineers who reviewed the output, refactored the architecture, wrote the test coverage, and signed off on the security audit are the reason it became enterprise-ready. Removing them from the process doesn’t make development faster. It makes delivery impossible.
AI-generated dashboards vs. analytics you can trust
AI can produce a beautiful analytics dashboard from a dataset in minutes. The charts will be well-designed. The numbers will add up. The visualisation will tell a compelling story. The problem is that AI has no inherent understanding of whether the underlying data is clean, whether the metric definitions are consistent with business logic, or whether the trend it’s highlighting is meaningful or a function of a forgotten filter.
A board presentation built on AI-generated analytics, without a data analyst validating the source, the logic, and the interpretation can become a liability rather than a strategic asset. The AI built the lens, but a human being has to confirm what it’s actually looking at.
AI-drafted proposals vs. client-ready deliverables
AI can produce a first draft of a client proposal that covers the structure, the boilerplate and the industry context. What it cannot do is understand the specific dynamic of your client relationship, the nuances of what was said in the last three meetings, the political sensitivities in the room, or the subtle positioning choices that make a proposal land smoothly rather than awkwardly.
The value lives in the person who reviews the proposal, rewrites the critical sections, adjusts the tone, and applies their judgment about what to emphasise. AI compressed a three-hour drafting task into twenty minutes. The human’s thirty-minute review made it deliverable.
Human in the loop is not a safety feature. It’s the strategy.
The phrase “human in the loop” is increasingly used as a governance checkbox – something organisations say to demonstrate AI responsibility. But it should be understood as something far more fundamental. We see it as the operating model that determines whether AI actually creates value or just creates volume.
Staff need to understand that their role in an AI-augmented workflow is not to rubber-stamp outputs but to bring the contextual intelligence, ethical judgment, and domain expertise that AI genuinely does not have. This requires training – not only in how to use AI tools, but in how to evaluate AI outputs critically. What does a hallucination look like in a legal summary? How do you spot a data anomaly in an AI-built report? When is an AI-generated code block technically functional but architecturally wrong?
These are the questions that a strong AI literacy programme answers. And they’re the questions that separate organisations who use AI well from those who use it enthusiastically.
AI should make your best people more powerful, not make your organisation less dependent on having good people. AI strategy shouldn’t be oriented around headcount reduction – it should rather focus on capability amplification.
AI as an accelerant – not a replacement
The most productive framing we’ve found for AI strategy is that AI is an accelerant. It amplifies what is already there. If what is already there is strong – experienced people, sound processes, clean data, clear strategic intent – AI makes it dramatically more powerful. If what is already there is weak – fragmented systems, unclear accountability, low data quality, inconsistent processes – AI makes those problems bigger, faster, and harder to see.
This is why AI strategy cannot be separated from digital strategy. The organisations that are getting the most from AI are invariably those that got their digital foundations in order first. Not because they planned for AI years in advance, but because the discipline of building well – investing in data quality, integration, governance, and digital capability – turns out to be exactly what AI adoption requires.
The question is not “are we using AI?” Almost everyone is, in some form. The question is: “have we built the strategy, the policy, the literacy, and the culture to make it count?”
Where to start: know where you actually are
In our experience, the single most useful first step for any organisation trying to move from AI experiments to AI strategy is an honest assessment of current maturity. The dimensions that actually determine whether AI will create value in your specific context are strategic alignment, data readiness, tool governance, staff capability, and process integration.
Most organisations, when they do this assessment seriously, find they are stronger in some areas and more exposed in others than they assumed. That asymmetry is the starting point for a real strategy. It tells you what to invest in first, what risks to mitigate immediately, and what ambitions are realistic in the next twelve months versus the next three years.
We’ve built maturity frameworks specifically for this – designed to give leadership teams a grounded, honest view of their AI readiness and a clear-eyed path forward.
Discover Your AI Maturity with Our Free Assessment at discover.insightconsulting.co.za
This article first appeared in Insight Edge, our newsletter on LinkedIn.