Customising LLMs for Public Affairs

By Paul Shotton, Advocacy Strategy

The next step is not agentic AI

Most public affairs teams do not yet need agentic AI. They need to get better at customising LLMs.

For many professionals, the first phase of AI has been exploratory. They have used tools like ChatGPT or Claude to summarise documents, improve drafts, brainstorm ideas, or structure notes. In many teams, that is still where things stand. A few early adopters are using the tools regularly. Others are interested but unsure how far to go.

That is why I think the next important step is not immediately more advanced autonomy. It is moving from general-purpose use to shaping LLMs around the way work is actually done.

At a basic level, that means defining tone, language, level of detail, formatting preferences, and audience. But it can also go much further. It can mean shaping the model around recurring tasks, standard outputs, team preferences, workflows, and the information environment in which the team operates.

A broader issue, seen through a public affairs lens

In one sense, this is not only a public affairs issue. Many professional teams are now facing the same question: how far should they move from general experimentation into more structured AI use?

But public affairs teams experience this question in a very particular context. They are often small. They work with politically, commercially, and strategically sensitive information. They depend on both internal and external inputs. And they need to translate intelligence into reporting, prioritisation, coordination, and action.

That combination makes customisation more relevant, but also more demanding.

Why public affairs is a special case

Public affairs is not generic knowledge work. It involves monitoring, analysis, prioritisation, internal reporting, stakeholder work, drafting, coordination, and campaign support. It also often involves information that requires careful handling: internal strategy, client materials, member inputs, and politically sensitive analysis.

That means the usefulness of an LLM depends much more heavily on whether it can be shaped around real workflows and constraints.

And those workflows vary significantly depending on the type of organization.

Different organizations, different customization questions

Trade associations need to think not only about their own internal processes, but about how they manage input from members, committees, and boards. They may also need to think about how AI supports collaboration, structures member contributions, and helps translate multiple views into usable internal analysis.

Consultancies face a different set of questions. They may have more variable workflows across monitoring, engagement, drafting, and project management. They may also see strong opportunities for efficiency, but they have to balance that against clear confidentiality boundaries across clients and accounts.

In-house teams face another reality again. They may have more access to broader organizational resources, systems, and support functions. But they also have to operate within company-wide policies on data, privacy, systems, and governance. Their choices may be shaped as much by wider corporate decisions as by the needs of the public affairs function itself.

So while the broader question is common across many professions, the practical implications are very different in public affairs.

Capability rises with complexity

This, for me, is the key point.

The more you customise an LLM, the more useful it can become. But the more you customise it, the more complexity you create.

More tailoring can improve quality, consistency, and fit with real work. But it also requires clearer workflows, more structured information, stronger review processes, better internal alignment, and more maintenance over time.

The investment is not only financial. It is also time, management attention, capability, and discipline.

That is why the most advanced setup is not always the right next step.

Maturity matters more than features

This is where maturity becomes critical.

Some public affairs teams are still exploring. Some have only one or two early adopters. Some have weak information management. Some have poorly defined workflows. Some still produce inconsistent outputs across the team.

In those situations, moving too quickly into a sophisticated AI environment can create more friction than value.

This is particularly important in public affairs because teams are often relatively small compared with legal, sales, or marketing functions. That can be a weakness, because there may be limited internal expertise and less capacity to absorb a complex setup. But it can also be a strength. A smaller team may find it easier to align practice, test use cases, and onboard new approaches if the setup is well chosen.

So the real issue is not size alone. It is whether the team’s current maturity matches the level of complexity being considered.

Audit before you implement deeply

That is why I think many public affairs teams should audit their current practice before investing heavily in AI customization.

How is work actually done today? Which workflows are clear and repeatable, and which are still informal? Where are the bottlenecks? How consistent are outputs? Where is information stored? Which use cases would create real value? And what level of complexity is the team realistically ready to absorb?

Those questions matter across all three organizational models.

In a trade association, the issue may be weak internal reporting, unclear handling of member input, or the challenge of structuring collaborative contributions. In a consultancy, it may be the tension between efficiency and confidentiality across accounts. In an in-house team, it may be the difficulty of aligning intelligence, prioritisation, and reporting across business units, functions, or regions.

In each case, the first step is not simply choosing a tool. It is understanding the workflow well enough to know where AI can genuinely help.

The bridge before agentic AI

This is where customising LLMs becomes important. It is the stage between simple experimentation and deeper operational adoption.

It is also, in many cases, the stage that comes before more agentic AI. Once a model is shaped around role, task, workflow, and context, the boundary starts to blur. But that does not mean teams should rush ahead. It means they should build the foundations properly first.

A more disciplined next step

So my own view is straightforward. For many public affairs teams, the next meaningful step is not chasing the most advanced AI environment. It is understanding their own practice well enough to customise LLMs in a way that fits their maturity, supports real use cases, and creates usable value.

The real question is not how advanced the tool looks. It is whether the team has the workflow clarity, information discipline, and internal readiness to make good use of that level of customization.

This version is stronger because it makes the public affairs angle more precise without pretending the whole issue is unique to public affairs.

The next improvement I would suggest is to make the opening even more punchy and slightly more “thought leadership” in tone, while keeping the rest grounded.