AI spending is growing 500 per cent in four years¹. Seventy per cent of organisations are using generative AI². Yet only 16 per cent have scaled enterprise-wide³, and CEOs report that just 25 per cent of AI initiatives have delivered expected ROI³. The gap between massive investment and minimal integration is the defining challenge for marketers and the agencies and technology partners that serve them. After 25 years inside enterprises — and having reviewed every major agency and technology company’s AI pitch from the client side — these are the ten things that separate the organisations capturing real value from everyone else.
1. Start with value out, not value in
Too many AI purchases are evaluated on Value In — did we get a good price, are the terms favourable? The question that actually determines success is Value Out: can the business use this technology to drive growth? Have we accounted for the people, processes, and time required to make it work? Before evaluating any platform, tool, or partner, define the specific commercial outcome you’re pursuing — revenue growth, cost reduction, speed to market, or competitive differentiation. If you can’t tie the AI initiative to a measurable business outcome within 12 months, you’re buying a demo, not making an investment.
2. Know the difference between efficiency AI and opportunity AI
Not all AI delivers the same type of value, and confusing the two is how budgets get wasted. Efficiency AI does the same things faster and cheaper — automating ad versioning, summarising reports, compressing production timelines. Opportunity AI does new things that weren’t previously possible — predicting customer churn, dynamically generating campaign strategies, entering new markets through personalisation at scale. Most organisations need both, but they require different business cases, different measurement frameworks, and different vendor relationships. Know which you’re buying before you sign.
3. Audit your readiness before you start buying
Most organisations overestimate their readiness and underestimate the gaps. A proper assessment looks across three dimensions simultaneously: business strategy and growth priorities, technology and data infrastructure, and people and culture. Assessing only one dimension gives you a dangerously incomplete picture. Fifty per cent of surveyed CEOs report that rapid investment has resulted in disconnected technology within their organisation³. The companies that move fastest are the ones that know exactly where their bottlenecks sit before they start selecting vendors.
4. Don’t let pilots become permanent experiments
The pilot trap is the single biggest waste of AI budget in marketing today. Isolated pilots deliver isolated results — time savings here, cost reductions there — but never compound into organisational capability. Before launching any pilot, define the pathway to scale: what governance approvals are needed, what data access is required, and how the learnings feed into the next initiative. Pilots without a scaling plan are just expensive demonstrations. Seventy per cent of organisations now use generative AI in at least one business function², yet only 16 per cent have scaled enterprise-wide³. That gap is the pilot trap in action.
5. Put governance first, not last
AI pilots don’t fail because the technology doesn’t work. They fail because nobody mapped the approval architecture. Governance — data handling, IP ownership, compliance controls, single sign-on requirements — should shape your pilot design from day one, not appear as a checklist at the end. The organisations getting AI into production fastest are the ones that designed for their risk committee before they designed for their marketing director. If your vendor can’t articulate how they handle governance from the outset, that tells you everything about their production readiness.
6. Deconstruct the vendor hype
Every vendor now leads with AI in their pitch. When you hear ‘our AI delivers 50 per cent efficiency gains’, ask: show me the before-and-after process map, what new human oversight is required, and what is the net time saving? When you hear ‘we unlock 1:1 personalisation’, ask: what specific data does it need, in what format, and what is the true integration timeline — not just the licence activation? When you hear ‘we provide unparalleled strategic insights’, ask: is our data used to train your central model, and what prevents you from selling that trained model to our competitors? The procurement leader’s job is to prove every claim, not accept it.
7. Demand contractual guardrails
The AI vendor contract your legal team signed three years ago is not fit for purpose. Six areas now require explicit protection. Model transparency and change control: no black boxes — you need to know what model is being used and be notified before it changes. IP and ownership of outputs: you own what you pay for. Cost controls and usage caps: no billing surprises as usage scales. Brand safety and content provenance: clear accountability for what the AI produces in your name. Exit and portability: no lock-in to proprietary formats or platforms. Data use limits: no surprise training or sharing of your data. If your vendor pushes back on any of these, that’s your answer.
8. Treat people and culture as a delivery requirement
Fifty-seven per cent of work hours are automatable⁴. That number terrifies your workforce if you don’t contextualise it. Skills evolve rather than vanish, but only if someone deliberately redesigns roles, expectations, and workflows. The biggest challenge isn’t buying or implementing the technology — it’s bridging the gap between the tool and the team to get improved outputs. AI adoption without a people strategy creates the worst of both worlds: expensive tools sitting unused while teams carry on working the old way. Change management isn’t a nice-to-have. It’s the difference between adoption and abandonment.
9. Evaluate partners on integration, not features
Technology has migrated from IT to marketing. Marketing now owns the budget and the business need, but decisions are decentralised and focused on features and speed. This creates a critical capability gap: marketing understands the ‘what’ but not always the ‘how’ of integration and scaling. The differentiator when choosing a partner isn’t whether they have AI capabilities — they all do. It’s whether they can integrate those capabilities across your business strategy, your technology stack, and your people. Ask how their solution connects to your existing martech investments. Ask what change management support they provide. The best partners think in systems, not features.
10. Think portfolio, not project
The CFO is no longer giving you three to five years. AI ROI expectations have compressed to one to three years, with CEOs reporting that only 25 per cent of AI initiatives have delivered expected ROI over the last few years³. The answer isn’t to bet everything on one initiative. It’s to run a portfolio: some Efficiency AI plays to stay competitive, some Opportunity AI bets for differentiation, and the discipline to measure each honestly against different success criteria. The organisations capturing transformational value are the ones deploying across multiple benefit categories simultaneously — not chasing a single use case.
References
1 IDC — Worldwide AI & Gen AI Spending Guide. Global AI spending forecast to grow from £40bn to £202bn between 2024 and 2028.
2 McKinsey, State of AI 2025. 71% of organisations regularly use generative AI in at least one business function.
3 IBM Institute for Business Value, 2025 CEO Study. 50% of CEOs report disconnected technology from rapid investment; 16% report enterprise-wide AI scaling; only 25% of AI initiatives have delivered expected ROI.
4 Accenture — A New Era of Generative AI for Everyone (2024). Analysis of automatable work hours across enterprise functions.
Signal42 is a practitioner-led consultancy helping enterprises navigate AI transformation across three integrated dimensions: business strategy, technology, and people. We’ve sat in your seat — and delivered results. If any of these points resonated, we should talk.
Signal42 | Beyond the hype. Into the value.