The 74% Problem: Why Most Companies Are Getting AI Wrong

At the Audit level, it delivers an AI readiness scorecard and a 30-day action plan with business-relevant use cases. Think of this as walking.

At the Adopt level, it provides structured discovery, a governance plan and a 6-12 month implementation roadmap with defined success criteria. Think of this as running.

At the Accelerate level, it provides true operational integration, embedding AI into day-to-day operations to drive efficiency and growth. Think of this as flying.

The key distinction? It’s built around the client’s specific business workflows, not around the technology. The technology comes third. People and strategy come first.

The role of Opal

Opal is the intelligence layer running through everything: CMS, CMP, Experimentation, Commerce, and your Data Platform.

It's an agent orchestration platform and that makes it a meaningful distinction. 

You can build agents, chain them together, and set them running autonomously on a schedule. 

One workflow can handle ideation, planning, variant creation and results summarisation, with or without anyone pressing a button in between.

And critically, it draws on your own data. Your brand guidelines. Your past campaign performance. Your experiment history. That's what separates it from a generic AI tool. It knows your brand deeply.

But before you switch it on, focus on these 4 things:

• First, define your outcomes. Not “we want to use AI more.” What does success look like in 90 days? How many experiments per week? What's your time to publish? If you can't measure it before Opal, you can't prove Opal moved it after.
• Second, brand infrastructure. Opal has an instructions layer and it genuinely transforms the quality of output. Brand voice, tone of voice, content guidelines, target personas. It's the difference between generic content and content that sounds like you. 
• Third, clean content architecture. One of the demos I've seen shows Opal working with a landing page, scanning it, checking content types, adding structured metadata for AI discoverability. That only works if what's underneath is solid. Opal can't structure what's already broken.
• Fourth, a specific use case. Not a general instruction to use AI more. One problem. One workflow. Something measurable. Prove value, build confidence, then expand. Perhaps start with experiment planning, perhaps with content adaptation. But start somewhere specific.

Clients could be forgiven for thinking that Opal is primarily a content tool. That's understandable, because a lot of marketing is content-led.

But nearly 60% of all Opal agent usage is in experimentation. That's where it's being used most heavily and where the results are strongest.

Teams using it across the full experimentation lifecycle are running 78% more experiments. Experiments are starting 19% faster. They're reaching statistical significance 25% faster. 

What good looks like in practice

Once the strategy is in place, the results can be remarkable.

But here’s the important caveat. None of these work without the strategic groundwork being done first.

I read a quote recently that sums up the danger perfectly: “Agents amplify strategy when it exists and amplify chaos when it doesn’t.”

Organisations that go straight from buying an AI platform to deploying agents won’t just fail to see value. Worse. They’ll encode their lack of strategy into AI systems that then execute bad processes really, really efficiently. Speed in the wrong direction is not progress.

At Niteco, we've built a really cool Automated Experimentation agentic workflow in Opal.

• We use Opal to analyse a webpage
• Then it combines real user behaviour data from GA4, and brings in proven UX research from Baymard.
• It then identifies opportunities and automatically suggests and prioritises experiments.
• It goes a step further by creating those experiments directly in Web Experimentation, building your entire experimentation roadmap for you.
• Outcome – your web manager can do in minutes what would traditionally require a full experimentation team — including a UX researcher, analyst, designer, developer and CRO specialist.
• It’s pretty much an entire CRO team in your pocket!

We've also built an Event Campaign Launch agentic workflow

The work of the marketing, digital and social teams is orchestrated through Opal.

• Creates a landing page or a microsite in your CMS
• Generates images
• Writes the copy
• Creates social posts (and images) to launch the event
• Schedules and posts them directly to the social platforms
• The other cool thing is it connects all those systems, your CMS, (CMP if you have it) your social channels – all together in a simple chat window 

The 10 questions to ask before you turn on AI

If you take one thing from this article, take these ten questions. Share them with your team tomorrow. Print them out and pin them to the wall.

If you cannot answer all ten, that should tell you something. It should tell you that you’re not ready yet. And that is absolutely fine. Answering them IS the strategy.

1. What specific business problem are we trying to solve?
Without a defined problem, AI is a solution in search of a use case. 46% of AI proofs-of-concept are discarded.

2. How do we measure success today, and what does “good” look like?
You cannot demonstrate ROI without baselines.

3. Is our data clean, structured and accessible?
43% of businesses cite data quality as their number one AI obstacle.

4. Who owns this initiative and who is accountable?
Two out of five AI projects stall due to ownership ambiguity.

5. Have we defined governance, brand guardrails and compliance rules?
Ungoverned AI creates risk faster than value.

6. Do our teams have the skills, training and desire to change?
Skills gap is the #1 adoption barrier. Businesses are not investing enough in training. Not even close.

7. Are the processes AI will touch already working?
It may sound obvious, but automating a broken process simply produces broken outputs faster. Fix before you automate.

8. Is leadership actively sponsoring, or just approving the budget?
AI high performers are 3x more likely to have active senior leadership engagement.

9. What is our change management plan?
Integrated change management makes projects 47% more likely to succeed.

10. Are we prepared to start small, learn and iterate?
The expectation of immediate enterprise-wide transformation is the enemy of progress.

The sequence matters

Five things to take away.

1. The problem is the problem. Most organisations are focused on the AI, not on what they’re trying to fix. If you can’t define your use case in one sentence with a number in it, you’re not ready. The technology is not the starting point. The problem is.

2. People and process, not technology, determine whether this works. 70% of AI failures are a people and process problem. For every pound you spend on AI technology, you should probably be spending seven on the humans and workflows around it. Build with your people, not for them. That distinction matters more than any framework.

3. AI without a strategy doesn’t fail quietly. It fails efficiently. Agents amplify strategy when it exists and amplify chaos when it doesn’t. The risk is not that AI doesn’t work. The risk is that it works very well in the wrong direction.

4. Change management is the single biggest predictor of success. Not the model. Not the platform. Not the budget. The people. Projects with excellent change management are seven times more likely to meet their objectives. And yet AI spending is rising at many times the rate of training investment. That gap is not sustainable.

5. The organisations that benefit most are not the fastest to adopt. They’re the ones who think first. We’ve seen this pattern before. Different decades, different technologies, remarkably similar failure rates. The technology got better every time. The failure rate didn’t. Why? Because the order of operations never changed.