The project model is not wrong, but it was built for a different pace of decision-making. 

A traditional research project usually begins once the question has already been shaped within the business. By that point, the brief may carry internal assumptions, preferred routes, stakeholder pressure, and a loose idea of what the answer is expected to prove. The agency responds to that brief, designs the work, collects the data, runs the analysis, and delivers the result. 

That works when the question is clear, and the business can wait. It works less well when the question is still moving. A campaign team may need to know whether a message is landing while spend is still live. A brand team may need to understand whether a shift in perception is a real signal or just noise. A leadership team may need to decide whether to hold, adjust, or invest before the next full study is ready. 

In those moments, research cannot behave like an event that happens separately from the business. It needs to be closer to the rhythm of the decision, without losing the rigor that makes the answer useful. 

That does not mean every question needs live tracking or a bigger project. Some need new research. Some need a fast steer. Some need deeper validation. Some need existing evidence brought together properly. And some do not need another project at all; they need a sharper conversation about what the business is really trying to decide. 

This is where agencies need to move earlier in the process. Not to create more work, but to stop weak briefs from becoming expensive projects. If an agency only enters once the method has been decided, it can only improve the work within that frame. If it understands the context before the brief is finished, it can challenge the frame itself. 

An always-on business does not need every question to become a full project. It needs a partner that can help decide what evidence is needed, how quickly it is needed, and whether the answer already exists somewhere in the organization. 

Embedded partnerships don’t have to mean more meetings, more status calls, or an agency being pulled into every internal conversation. It simply means that the agency has enough context to be helpful before the question becomes urgent. 

It understands what has already been tested, what the business is trying to achieve, where stakeholder pressure is coming from, and where the same questions keep returning in different forms. It also knows when the client genuinely needs new evidence, and when the better answer is to connect what already exists. 

That context makes the work faster, because the agency is not starting from zero each time. But more importantly, it makes the work sharper. A partner with context can challenge the brief, not just respond to it. 

That does not mean becoming part of the client’s internal view. John-William talks about internal teams becoming “a bit of an echo chamber,” and agencies have to avoid being absorbed into that same pattern. The best relationships sit somewhere in the middle: close enough to understand the business, but independent enough to challenge it. 

This also changes the workflow. The agency should not only create value at the final presentation. Before the work starts, it can help clarify the question. During the work, it can flag when an early signal matters or when the original question needs adjusting. After the work, it can help the client understand what should be watched next. 

That’s what embedded should mean: not constant contact, but continuity. 

The harder question is not just how agencies work, but how they train people to work this way. 

AI can already take on parts of the process that used to sit with junior researchers: drafting surveys, summarizing open-ends, pulling together background material, producing first-pass analysis, checking patterns, and building slides. Some of that is useful. Nobody needs to defend every slow or repetitive part of the old model. 

But the old model did teach researchers something. 

Checking tables, writing questionnaires, managing fieldwork, and building charts were not just admin tasks. They helped people understand how research can go wrong: a badly worded question, a sample that looks fine until you cut it properly, a chart that seems clear but overclaims the data. 

John-William says the industry needs to think less about training people only in “the doing of writing surveys and collecting data and checking data tables,” and more about helping them understand why those things matter. 

That matters because judgment doesn’t appear just because the tools get better. If younger researchers spend less time on the mechanics, agencies need to be more deliberate about how they learn the craft. They need exposure to messy data, weak briefs, bad questions, difficult stakeholders, and the consequences of acting on evidence that is not strong enough. 

The same applies to agents. An agent might help draft a survey, search past work, summarize a dataset, or produce a first pass of an analysis. But it can’t know by itself whether the question was framed properly, whether the evidence is strong enough, or whether the recommendation is right for the business context. 

“A data point is just a data point. There’s a decision that needs to happen,” says John-William. The agency of tomorrow needs a clearer split: let tools remove repetitive work where they can, but keep humans close to framing, interpretation, challenge, and recommendation. 

This shift also changes the commercial conversation. If AI and automation reduce the time it takes to produce parts of the work, clients will ask why those parts should cost the same. That is fair. Agencies should not pretend nothing has changed. 

But good research has never only been about hours spent. As John-William puts it, “you pay for my experience, not my time.” If someone can get to a better answer faster because they have spent years learning what matters, the value is not just the time on the clock. It is the judgment, risk reduction, and confidence the work gives the client. 

That doesn’t mean agencies can keep old pricing and call it expertise. Some tasks will become cheaper. Some workflows may become subscription-based, especially where platforms, dashboards, or ongoing monitoring are involved. Some work will stay project-based because not every question needs a continuous model. 

But pricing has to become more honest about what is being bought. A repeatable task should not be priced like a strategic call, a platform-supported workflow should not be priced like a fully bespoke project, and a senior recommendation that helps avoid a bad decision should not be valued only by how long it took to write the slide. 

John-William describes the direction as “outcomes-based rather than input-based.” Not every piece of research can be priced perfectly against an outcome, but the principle matters. If the inputs change, agencies need to be clearer about where the value really sits. 

The agency of tomorrow is not just the old agency with better tools. 

If AI is used only to write the proposal faster, draft the questionnaire faster, summarize the data faster, and produce the same final deck faster, the model has not really changed. It has only become more efficient. 

The bigger change is in how agencies stay connected to the business between projects, how they help improve the question before work begins, how they use tools without losing research craft, and how they train people when the old junior ladder is changing. 

Some questions will still need the full project model. But agencies cannot treat every question as if it belongs in the same sequence of brief, fieldwork, analysis, and debrief. 

Always-on businesses need something more flexible, sometimes they need a fast steer, sometimes they need deeper validation. Sometimes they need evidence they already have to be joined up properly, and sometimes they just need a partner to say that a new project is not the answer yet. 

The real shift is not more activity, and not just speed for its own sake, but a way of working that stays closer to the business while protecting the rigor that makes research worth trusting, when it actually matters. 

You can watch or listen to the full podcast episode here:  https://www.research-revolutionaries.com/e14-the-insight-agency-of-tomorrow-why-perspective-matters-more-than-data/