A practical, realistic guide to making AI actually work in your business
Blind AI adoption is real. Leaders are putting a finger in the air and guessing a direction. Without the groundwork, the reality gets messy fast: shadow experiments, inconsistent quality and nothing you can scale with confidence.
Leaders feel they “should be doing something with AI,” yet the data says many programmes are stuck: 63% still in an experimentation phase, 61% lacking a formal change communication strategy, and 40% of employees saying AI doesn’t save them any time. That’s not a tooling problem; it’s a readiness problem.
I’ve spent 25 years in software, and new waves of tech aren’t the problem; poor housekeeping is and leaving your people behind. The way through is unglamorous: set policy and guardrails, standardise tools to reduce context switching, train people into good habits, and fix the data foundations that make outputs trustworthy. Do that, and AI stops being a side experiment and starts becoming an operating model you can run.
Remember that AI readiness is not a technology shopping list; it’s a people-first transformation where you need careful planning, investment, and management of a clear AI adoption roadmap. For good marketing agencies, the opportunity from AI is far greater than the threat, but only if they use it to raise the standard of thinking and delivery rather than automate the obvious.
Define AI readiness properly (it’s not a licence purchase)
AI readiness is the ability to reliably, safely, and repeatedly turn AI into measurable outcomes. It’s a mix of strategy, governance, skills, data and adoption; not “we bought licences.” It means deciding what’s allowed, what’s off-limits, and what “good” looks like before scaling.
Be optimistic, but sceptical: most hype is hiding a delivery gap
I’m optimistic about AI, but deeply sceptical of the hype around it, because we’re seeing the same pattern as the dot-com boom: huge promise, massive over-investment, and a lot of noise. When vendors sell a future that doesn’t fully exist yet, the gap between demos and delivery becomes your problem.
That noise is costly because it pulls teams away from work that lands value. A recent snapshot found 95% of British businesses say AI hasn’t improved productivity—yet we’ve seen an early 3% uplift at IDHL, which tells you the gap is execution, not possibility.
The tech is often the easy part; governance, data and process decide whether AI succeeds or fails. Miss the guardrails, training and data foundations, and you’ll spend the next year arguing about risk and redoing work that should have been established in the first place.
Start with outcomes and use cases, not a blanket rollout
Pick three to five priority use cases tied to outcomes you can defend: revenue, cost, risk reduction, or customer experience. Define what “good” looks like, set success criteria (quality, time saved, adoption, risk) and scale from assisted use to controlled agents before automation.
The real advantage comes from changing how people work: who does what, when humans review, and where AI sits inside the workflow. Without that, you generate content faster inside the same broken process. For agencies and service firms, AI doesn’t make you obsolete; it exposes whether you ever had a point of view in the first place.
Guardrails first: move fast without breaking trust
Clear guardrails let teams experiment with confidence because they know what’s permitted, what’s sensitive, and where human review is mandatory. A simple approval path of commercial sign-off plus IT/security and legal/data protection beats a free-for-all every time.
Standardisation matters because tool chaos kills adoption through context switching. Anchor to a core ecosystem, publish an approved tools list, and stop treating every new release like an organisational priority. If you want AI to work, you have to build habits, not just roll out licences.
Data readiness is the biggest lever (and nobody wants to talk about it)
In my experience, data is the unsexy truth, and it’s non-negotiable if you want AI to work. You can’t prompt your way out of poor data foundations. The moment you move from experimentation into business-critical workflows is when this becomes painfully obvious.
When I say “get your data in order,” I mean clear ownership, quality standards, lineage, and access control. I also mean taxonomy and metadata, plus a retrieval structure people can actually use, especially if you expect knowledge assistants to be accurate. If teams can’t confidently find the single source of truth today, AI will amplify confusion tomorrow.
Train early, then democratise: capability should scale beyond specialists
The fastest way to stall adoption is to assume people will “figure it out.” Training belongs near the top of the checklist, alongside guardrails, because trust and clarity matter more than hype. Teach prompt craft, evaluation, workflow redesign, and data literacy, then make good practice easy to repeat.
Generic policy rarely survives contact with real teams, so build team-specific playbooks instead. They translate governance into “how we work here,” with examples, do’s and don’ts, and role-specific quality checks. Combine that with safe autonomy and letting people build and share agents within guardrails so you can stop innovation getting trapped in pockets.
AI should take tasks, not roles, and raise the standard of work
A better framing for leaders and teams is that AI isn’t coming to take your job, it’s coming to take your tasks. That points at the real prize by removing administrative friction so experts can spend more time on judgement, strategy and creative finesse. AI’s biggest impact isn’t client‑facing automation; it’s removing admin so teams can think.
There’s a wider implication for service organisations, as AI raises the level of “average,” differentiation becomes harder and therefore more valuable. When everyone can do the basics, originality and judgement become the real differentiators, and that’s where good agencies win. The standard rises and so does the value of genuine expertise.
Measure impact credibly: early signals beat perfect certainty
Use a mix of signals: adoption, cycle time, quality, and fewer risk incidents. In IDHL’s internal enterprise AI platform, around 75% of participants used Copilot actively in the last 30 days and 93% said AI tools were beneficial to their role. Around 90% reported being more productive, 84% said quality improved, and 52% said it improved communication.
Early indicators suggest up to a 3% productivity uplift; the capacity IDHL estimates to be worth £1.5m in a £50m business. The tools still pay for themselves, so the question becomes whether you can adopt them safely and consistently.
Move beyond content: use AI to change how workflows
Readiness shows up when AI leaves the “draft me a paragraph” phase and starts improving operations. IDHL has used an internal agent to analyse project schedules, score “schedule health” (including context switching), and flag risks before the week goes sideways. AI becomes part of how work is managed, not just how content is produced.
Plan for the economics: AI won’t stay “cheap” forever
Readiness includes commercial realism. AI is underpriced right now compared to the value it can create, and it would be naïve to assume today’s unit economics persist. Track cost per task, forecast what happens if prices rise, and build value-based ways of working rather than volume for its own sake.
For service businesses, there’s an additional tension that speed forces a conversation about fees. The response shouldn’t be panic discounting; it should be clearer value and better outcomes. Less time administering accounts means more time doing the strategic work clients actually value, which is where pricing models will ultimately land: packaging judgement and outcomes, not hours.
A practical AI readiness checklist for leaders
Define readiness. Decide what “safe, repeatable value” means and what risks you won’t accept.
Pick outcomes. Choose 3–5 use cases you can defend, with clear success criteria.
Fix data foundations. Ownership, quality, access, metadata and a single source of truth; without this, accuracy won’t scale.
Put guardrails in writing. Approved tools, banned data, review points and how quality is checked.
Standardise. Reduce context switching by anchoring to a core ecosystem and an approved list.
Train early. Onboard new starters and use team playbooks as one size won’t fit all.
Copilots to agents. Increase autonomy in steps: assist first, then controlled action, always with monitoring.
Measure. Track adoption and outcomes, then sanity-check it with user reality: does it actually save time?
Build habits. Create a community and momentum: champions and a shared tips network.
Plan economics. Forecast cost, test ROI, and revisit pricing and value as expectations shift.
Most businesses are chasing shiny AI tools, but the advantage comes from changing how people work. AI will change organisations, but the messy middle is where leaders earn their keep. Playing it safe is no longer safe: treat readiness as an operating model shift where it is designed, governed, and embedded rather than a race to buy the next tool.
