With the advent of AI tools, government leaders are excited about new opportunities to use new technology to run more efficient and effective public engagements.

Lacking any background in engagement or in AI, a public professional finishes a course that teaches them that good engagement should be inclusive, purposeful, and tied to a real decision. They leave with solid frameworks and good examples. 

Then an opportunity arises to go out and solicit public input, but they are short-staffed. Their supervisor is reluctant. They don’t know which platform to use. They face questions the course either never quite answered or they can’t remember, what do I do next?

That question is why we built our public engagement coach.

The coach is a tool for public professionals who want to plan real engagement under real conditions.

The coach is a tool for public professionals who want to plan real engagement under real conditions. It is not another course. It picks up after the course, at the point where most people get stuck, and helps a practitioner turn what they learned into a plan they can act on.

The knowing-doing gap

Governments have invested heavily in training officials to run more inclusive and effective engagement. They introduce better methods and stronger examples of communities shaping public decisions.

But training does not always translate into changes in practice, and researchers have long studied why. In 1988, two organizational scholars, Timothy Baldwin and Kevin Ford, published a review titled "Transfer of Training" in the journal Personnel Psychology. 

Their central point was that learning something in a training setting does not guarantee a person will use it back on the job. Whether the training transfers depends on the learner, the training design, and, above all, the conditions of the workplace they return to. That last part is the knowing-doing gap in a sentence. 

In public engagement, the gap is easy to see. A practitioner understands the principles but still faces a tight deadline, limited staff, political pressure, and uncertainty about who needs to be involved. The frameworks feel abstract when the real question is narrow: what should I do, for this project, with these constraints?

When support is missing at that moment, people fall back on what is familiar. They run a survey, hold a town hall, or hire a consultant, not because those methods fit the goal, but because they are easy to defend and organize. Processes reach the wrong people, collect input too late to matter, or gather feedback with no plan to use it. For communities, it feels performative. For staff, it feels like being stuck.

Turning engagement principles into real-world practice is an implementation problem. Most tools are built to solve knowledge problems.

Turning engagement principles into real-world practice is not a knowledge problem. It is an implementation problem. And most learning tools are built for the first, not the second. They explain concepts, summarize lessons, and answer questions. They were never designed to help someone decide which residents, agencies, and internal decision-makers to involve first, or whether the goal is consultation, co-creation, or deliberation.

What the coach does

The coach starts where a real project starts: with the practitioner's own challenge, in their own words. Someone can begin by typing, "I have six weeks to plan an engagement process and no idea where to start."

From there, the coach asks targeted questions instead of dictating what to do or offering a generic checklist. 

What decision will public input shape? Who is most affected by it? What kind of input would actually change the outcome? What constraints are fixed, and where is there room to move?

As we built the tool, we drew on engagement from the GovLab and Nesta framework to identify the planning decisions behind effective engagement. The aim is to help the practitioner make clearer decisions about their own project, not to hand back a tidier explanation of theory.

It should feel less like filling out a form and more like talking a project through with someone who knows the structure of good engagement.

It should feel less like filling out a form and more like talking a project through with someone who knows the structure of good engagement.

How we built it

The harder problems were technical, and most of them came down to one thing: keeping the advice grounded and specific instead of generic.

We built the coach as a multi-agent workflow rather than a single chatbot. Every message first goes to an orchestrator, a router that reads what the practitioner said and the state of their session, then hands the message to one specialist. 

Take a town administrator planning a new community center. They want residents to help decide what services and amenities it should include, but in past projects, the Spanish and Arabic-speaking families never showed up, and that input is exactly what a center for the whole community needs. 

So they type what is on their mind: "How do I engage non-native English speakers in our community?" The coach does not hand back a list. It asks them to get specific about who they are trying to reach and what they already know about those residents. 

They say they want to reach Spanish and Arabic-speaking families, but are not sure of the best way to connect. 

The coach pushes once more, toward practical partners, and they arrive at an answer of their own: work through local schools and mosques, and reach families over WhatsApp. 

With that, they have a real outreach strategy, the coach marks that the planning question is addressed, and a pro tip drawn from expert interviews that reinforces that trusted local institutions usually work better than official notices alone.

• A coach agent does the main work, walking the practitioner through nine planning questions drawn from the GovLab framework, one at a time. 

• A retrieval agent steps in to pull real examples and case studies when someone is stuck or asks how others have handled a similar problem. 

• A third agent suggests which question to take on next. 

Splitting the work this way lets each step stay focused, and a shared session tracks which questions are still open, so the conversation never loses its place.

Underneath sits a retrieval system. Instead of relying on the model's general knowledge, the coach draws from a curated library of engagement frameworks, methods, and real cases. 

When a practitioner describes a project, the system retrieves the material that fits, and the coach reasons from that, so its recommendations point to established practice rather than to plausible-sounding invention. This is what keeps the advice trustworthy when a practitioner is about to act on it.

The library also holds something the literature cannot offer: tacit knowledge, the lessons experienced practitioners carry in their heads but rarely write down.

We interviewed engagement experts around the world, anonymized the conversations, and tied each lesson to the planning question it addresses

To capture it, we interviewed engagement experts around the world, anonymized the conversations, and tied each lesson to the planning question it addresses. Research can tell you what good engagement looks like, but the interviews tell you what actually goes wrong in practice and what people who lived through it did about it.

The recurring design decision was restraint. It would have been easy to ask more questions, surface more options, and produce longer plans. But a tool for busy staff has to stay light.

So the coach asks for a follow-up only when an answer is too broad to act on, surfaces examples only when they help, and works toward a usable plan rather than a complete one.

Why this matters now

As AI spreads across government work, there is a real risk that new tools simply make information easier to access without making action easier to take. That would not close the knowing-doing gap. It could widen it, giving people more summaries and more guidance while they still struggle to apply any of it.

The better path is to build tools that support judgment, not just retrieval.

The better path is to build tools that support judgment, not just retrieval. In public engagement, that means helping someone move step by step from a vague challenge to a clearer direction, reason through tradeoffs, and meet the moment where training usually stops.

Public engagement is just our first test, because the knowing-doing gap appears anywhere people move from learning principles to applying them to a real project. Imagine an urban planner being coached through planning a new park, or a workforce leader thinking through how to design an apprenticeship program. If the approach works here, we want to try it in domains like these next.

The success of public engagement is not decided in the classroom. It is decided when a practitioner returns to their desk and has to choose what to do next.

Because the success of public engagement is not decided in the classroom. It is decided when a practitioner returns to their desk and has to choose what to do next.

That is where support has been missing. And that is where the coach can help.