The Risk of Moving Too Fast into AI Complexity

By Paul Shotton, Advocacy Strategy

Something has shifted in the way AI is sold to public affairs teams. The early conversations were about simple, general-purpose assistants. Increasingly, teams are being introduced to far more advanced things — custom datasets, bespoke knowledge bases, complex platforms with elaborate dashboards, and integrations that promise to do almost everything at once. I have recently heard several examples of teams being walked through these advanced capabilities before they had gone through any structured process of working out what they actually needed.

The instinct behind this is understandable. If the technology is powerful, why not start with the most powerful version of it? But it raises a question worth sitting with: can a team move too fast into complexity? In my experience the answer is yes — and the cost is not only wasted money. It is confusion, disengagement, and a strategy that gets ahead of the organisation trying to deliver it.

To be clear, the problem is not that advanced platforms or custom datasets are wrong. They can be genuinely valuable. A mature team with well-mapped workflows and good data discipline may get a great deal from them. The problem is sequencing — introducing sophistication before the groundwork that makes it useful is in place.

When the technology gets ahead of the organisation

There is a reason capability and value do not move in step. The economist Erik Brynjolfsson has spent years documenting what he calls the productivity paradox: organisations that invest heavily in new technology often see little return until they make the slower, harder, complementary changes — to processes, skills, and ways of working — that let the technology pay off. The technology is the easy part. The organisational adaptation around it is what actually creates value, and it cannot be bought at the same speed.

The same logic applies to AI in public affairs. If a team has not mapped its workflows, identified realistic use cases, assessed its own maturity, clarified what data can and cannot be used, and built a basic level of confidence among its people, then a sophisticated platform simply arrives ahead of the organisation. The capability is real, but there is nothing in place to convert it into value. The dashboards go unused, the custom dataset is never properly maintained, and the team quietly falls back on the simple tools it actually understands.

The cost of buying complexity too early

There is also a straightforward commercial risk. Advanced platforms are expensive, and they are often sold on the strength of their features — the bells and whistles — rather than on the specific problems a team needs to solve. It is entirely possible to pay a great deal for capability that a team is not yet able to use, and to be left with an impressive system and a disappointing return. The features were never really the issue; the readiness to use them was.

In many cases, a more structured and deliberate use of widely available tools such as Claude or ChatGPT will deliver more immediate, practical value at a fraction of the cost. A team that has thought carefully about its prompts, its use cases, and its review process can get a long way with general-purpose tools before a bespoke platform becomes the limiting factor. The sophistication that matters early on is in how the team works, not in what it has bought.

A better question than “what is the most advanced tool?”

The question many teams are encouraged to ask is: what is the most advanced tool we can get? It is the wrong question, because it starts from the technology rather than the need. A more useful question has three parts. What problem are we actually trying to solve? What is our current maturity — honestly? And what is the simplest effective tool or process for this stage? Phrased that way, the answer is often less ambitious, and far more achievable, than the one the market suggests.

This connects to a distinction worth keeping in mind: the difference between using AI as an isolated tool, integrating it into structured workflows, and moving toward more automated or agentic systems. Each step adds capability, but each also demands more — cleaner data, clearer processes, stronger governance, more confident users. Moving up that ladder before the lower rungs are solid is how teams end up with systems they cannot use and people who no longer trust the project.

What happens to the people

The human cost of moving too fast is easy to underestimate. Jump straight into complexity and you tend to produce three reactions at once. The early adopters are delighted — they enjoy the sophistication and may even have driven the purchase. The majority, who were only just becoming comfortable, are now confused and quietly disengage, unsure how the new system fits the work they actually do. And the sceptics, who had reservations to begin with, feel vindicated: this was exactly the over-engineered, top-down imposition they feared. A single premature leap can stall adoption across all three groups at the same time.

Sequence the adoption

The alternative is not to avoid sophistication, but to sequence it. The sensible order starts with the unglamorous foundations: understanding current practice, mapping workflows, identifying use cases, putting basic governance and data hygiene in place, choosing tools that fit, and training people to use them well. Only once that base is solid — and the limitations of the simpler approach are genuinely being felt — does it make sense to move toward more advanced platforms or custom solutions. By then the need is clear, the data is in better shape, and the team has the confidence to use what it buys.

A practical AI policy supports this sequencing by setting the right mindset. It should make experimentation welcome but not uncontrolled, with clear boundaries, sensible guardrails, and an obvious route for escalating questions. A policy like that lets a team move quickly on the simple things while keeping the more ambitious moves deliberate rather than impulsive.

Clarity before sophistication

Sophistication is not the enemy. Premature sophistication is. The teams that get the most from advanced AI are usually the ones that earned their way there — that built clarity about the work, the people, and the data before they reached for the most powerful tool in the room. The order matters more than the ambition.

So the real question is not how advanced your AI can be, but whether your organisation is ready to make it worth the cost. Sophistication should come after clarity, not before it — and it is worth asking, honestly, which one your team is reaching for first.