Insight
Insight
The AI Edition: Fix the Engine, Not Just the Output
14 Jun 2026
Matt Cull
At Kettner’s, Soho House, the Vanndat team hosted The AI edition: fix the engine, not just the output – a breakfast roundtable bringing together senior leaders from across agency leadership, strategy, finance, operations, technology, data, business intelligence and commercial roles.
The discussion focused on one of the biggest questions facing agency businesses right now: how do we move from experimenting with AI to making it genuinely useful, scalable and commercially valuable?
Rather than centring on the more familiar conversation around AI-generated outputs, the session looked beneath the surface. Around the table, the discussion explored what needs to happen behind the scenes for AI to work properly inside complex organisations: cleaner data, better-connected systems, stronger operational processes, clearer governance, smarter investment decisions and new skills for the people using these tools every day.
The central question was simple:
How can agencies make AI genuinely useful, scalable and commercially valuable, rather than simply impressive at the surface?
AI needs stronger foundations
One of the clearest themes from the discussion was that AI is only as effective as the data, systems and processes it sits on top of.
While clients and stakeholders increasingly expect AI to create speed, efficiency and better work, many organisations are still managing fragmented data, inconsistent processes, legacy systems and operational workarounds that have built up over time.
This creates a clear tension. AI may promise faster delivery and sharper insight but if the underlying data is incomplete, inconsistent or poorly governed, it becomes much harder to scale those benefits safely.
The group discussed the importance of improving data quality, consolidating information across systems and aligning processes across teams, markets and departments. For organisations that have grown through acquisition or operate across multiple regions, this standardisation becomes even more important.
The takeaway was clear: AI transformation cannot be treated purely as a technology project. It is also a data, process and operating model project.
Fixing the engine before accelerating the output
A recurring point throughout the session was that businesses should not simply place AI on top of broken or inefficient systems.
AI can help speed up tasks, support data migration, improve process testing and automate parts of operational delivery. But without the right foundations in place, those efficiencies may be limited or unreliable.
The discussion explored how agencies can use AI not only to produce more, but to understand how work actually moves through the business. This includes identifying bottlenecks, improving data hygiene, measuring process efficiency and creating better visibility across operations.
There was also a strong emphasis on starting earlier. Rather than waiting until a transformation programme is already underway, organisations should begin data quality and process mapping work ahead of time so AI has a stronger base to work from.
In other words, AI should not just make the existing machine run faster. It should help businesses understand where the machine needs to be redesigned.
Commercial models are under pressure
The conversation also explored the commercial impact of AI.
As AI enables agencies to work faster, clients may increasingly expect those efficiencies to be reflected in pricing. This creates pressure on traditional time-based models and raises important questions about how agencies define and communicate value.
The group challenged the idea that faster work should automatically mean cheaper work. If AI helps agencies deliver better thinking, sharper insight, stronger outcomes and more responsive service, then value should not be measured only by time saved.
There was broad agreement that the industry may need to move more confidently towards value-based or outcome-based commercial models. This would require agencies to better demonstrate the impact of their work, not just the hours spent delivering it.
The discussion also highlighted the potential value of operational and finance data. Agencies often hold information that could help clients understand where inefficiencies, rework or hidden costs exist. Used well, this data could become a source of client value, helping to improve not only agency operations but also client-side decision-making.
AI is changing the skills people need
Beyond systems and commercial models, the discussion also turned to people.
One of the most important questions raised was what happens to learning, judgement and experience when AI starts doing more of the tasks junior talent used to learn from.
In finance, operations, strategy and delivery roles, much of people’s judgement has traditionally been built through doing the work: checking, reconciling, spotting errors, fixing problems and learning from mistakes. If AI takes on more of that work, organisations will need to be much more intentional about how they develop critical thinking and quality control skills.
The group discussed the growing importance of prompting, reviewing, challenging and managing AI outputs. As AI agents become more capable, people may increasingly find themselves supervising systems rather than completing every task manually.
This does not make human judgement less important. It makes it more important.
Future teams will need to know how to ask better questions, spot weak answers, identify risk, understand context and decide when something does not feel right. Training will need to evolve accordingly.
Innovation needs guardrails
Another theme was the balance between innovation and governance.
AI is creating new opportunities for people across organisations to build tools, automate workflows and experiment with new ways of working, even without traditional technical backgrounds. This entrepreneurial energy can be hugely valuable, but it also creates new risks around compliance, data security, consistency and scalability.
The discussion recognised the need for clear governance, particularly when AI touches client data, finance systems, enterprise tools and regulated processes. At the same time, there was a clear concern that governance should not become a reason to stand still.
The challenge for established organisations is to create the right guardrails: enough structure to keep experimentation safe and aligned, but not so much friction that innovation slows down.
Smaller or newer businesses may be able to move faster because they have fewer legacy systems and simpler operating models. Larger organisations need to find ways to combine their scale, governance and client relationships with the speed and adaptability that AI now demands.
The need for bolder investment decisions
The group also discussed whether businesses are being ambitious enough in the AI investments they approve.
Many AI business cases are still built around improving existing systems or making current processes more efficient. While that can be valuable, it may not go far enough.
The bigger opportunity may be to rethink the model entirely: how work is delivered, how teams are structured, how value is measured, how client relationships are managed and how agencies create differentiation in an AI-enabled market.
This requires a different kind of investment conversation. Rather than asking only how AI can reduce cost within today’s structures, businesses may need to ask what future operating model they are trying to build.
What comes next?
The session made clear that AI can speed things up. But speed alone is not the goal.
The more important question is where that speed should take the industry.
A few clear themes emerged from the discussion:
Data quality needs to come before AI scale.
AI business cases should look beyond existing operating models.
Commercial models need to reflect value and outcomes, not just time saved.
People will need stronger skills in critical thinking, quality assurance and AI management.
Governance should enable safe progress, not prevent it.
Operational insight can become a meaningful source of client value.
AI should not be treated as a layer placed on top of broken systems.
The message from the roundtable was clear: agencies have a major opportunity to use AI not just to produce work faster, but to build smarter, more adaptive and more valuable businesses.
To do that, the industry needs to look beneath the surface.
It needs to fix the engine, not just the output.
At Kettner’s, Soho House, the Vanndat team hosted The AI edition: fix the engine, not just the output – a breakfast roundtable bringing together senior leaders from across agency leadership, strategy, finance, operations, technology, data, business intelligence and commercial roles.
The discussion focused on one of the biggest questions facing agency businesses right now: how do we move from experimenting with AI to making it genuinely useful, scalable and commercially valuable?
Rather than centring on the more familiar conversation around AI-generated outputs, the session looked beneath the surface. Around the table, the discussion explored what needs to happen behind the scenes for AI to work properly inside complex organisations: cleaner data, better-connected systems, stronger operational processes, clearer governance, smarter investment decisions and new skills for the people using these tools every day.
The central question was simple:
How can agencies make AI genuinely useful, scalable and commercially valuable, rather than simply impressive at the surface?
AI needs stronger foundations
One of the clearest themes from the discussion was that AI is only as effective as the data, systems and processes it sits on top of.
While clients and stakeholders increasingly expect AI to create speed, efficiency and better work, many organisations are still managing fragmented data, inconsistent processes, legacy systems and operational workarounds that have built up over time.
This creates a clear tension. AI may promise faster delivery and sharper insight but if the underlying data is incomplete, inconsistent or poorly governed, it becomes much harder to scale those benefits safely.
The group discussed the importance of improving data quality, consolidating information across systems and aligning processes across teams, markets and departments. For organisations that have grown through acquisition or operate across multiple regions, this standardisation becomes even more important.
The takeaway was clear: AI transformation cannot be treated purely as a technology project. It is also a data, process and operating model project.
Fixing the engine before accelerating the output
A recurring point throughout the session was that businesses should not simply place AI on top of broken or inefficient systems.
AI can help speed up tasks, support data migration, improve process testing and automate parts of operational delivery. But without the right foundations in place, those efficiencies may be limited or unreliable.
The discussion explored how agencies can use AI not only to produce more, but to understand how work actually moves through the business. This includes identifying bottlenecks, improving data hygiene, measuring process efficiency and creating better visibility across operations.
There was also a strong emphasis on starting earlier. Rather than waiting until a transformation programme is already underway, organisations should begin data quality and process mapping work ahead of time so AI has a stronger base to work from.
In other words, AI should not just make the existing machine run faster. It should help businesses understand where the machine needs to be redesigned.
Commercial models are under pressure
The conversation also explored the commercial impact of AI.
As AI enables agencies to work faster, clients may increasingly expect those efficiencies to be reflected in pricing. This creates pressure on traditional time-based models and raises important questions about how agencies define and communicate value.
The group challenged the idea that faster work should automatically mean cheaper work. If AI helps agencies deliver better thinking, sharper insight, stronger outcomes and more responsive service, then value should not be measured only by time saved.
There was broad agreement that the industry may need to move more confidently towards value-based or outcome-based commercial models. This would require agencies to better demonstrate the impact of their work, not just the hours spent delivering it.
The discussion also highlighted the potential value of operational and finance data. Agencies often hold information that could help clients understand where inefficiencies, rework or hidden costs exist. Used well, this data could become a source of client value, helping to improve not only agency operations but also client-side decision-making.
AI is changing the skills people need
Beyond systems and commercial models, the discussion also turned to people.
One of the most important questions raised was what happens to learning, judgement and experience when AI starts doing more of the tasks junior talent used to learn from.
In finance, operations, strategy and delivery roles, much of people’s judgement has traditionally been built through doing the work: checking, reconciling, spotting errors, fixing problems and learning from mistakes. If AI takes on more of that work, organisations will need to be much more intentional about how they develop critical thinking and quality control skills.
The group discussed the growing importance of prompting, reviewing, challenging and managing AI outputs. As AI agents become more capable, people may increasingly find themselves supervising systems rather than completing every task manually.
This does not make human judgement less important. It makes it more important.
Future teams will need to know how to ask better questions, spot weak answers, identify risk, understand context and decide when something does not feel right. Training will need to evolve accordingly.
Innovation needs guardrails
Another theme was the balance between innovation and governance.
AI is creating new opportunities for people across organisations to build tools, automate workflows and experiment with new ways of working, even without traditional technical backgrounds. This entrepreneurial energy can be hugely valuable, but it also creates new risks around compliance, data security, consistency and scalability.
The discussion recognised the need for clear governance, particularly when AI touches client data, finance systems, enterprise tools and regulated processes. At the same time, there was a clear concern that governance should not become a reason to stand still.
The challenge for established organisations is to create the right guardrails: enough structure to keep experimentation safe and aligned, but not so much friction that innovation slows down.
Smaller or newer businesses may be able to move faster because they have fewer legacy systems and simpler operating models. Larger organisations need to find ways to combine their scale, governance and client relationships with the speed and adaptability that AI now demands.
The need for bolder investment decisions
The group also discussed whether businesses are being ambitious enough in the AI investments they approve.
Many AI business cases are still built around improving existing systems or making current processes more efficient. While that can be valuable, it may not go far enough.
The bigger opportunity may be to rethink the model entirely: how work is delivered, how teams are structured, how value is measured, how client relationships are managed and how agencies create differentiation in an AI-enabled market.
This requires a different kind of investment conversation. Rather than asking only how AI can reduce cost within today’s structures, businesses may need to ask what future operating model they are trying to build.
What comes next?
The session made clear that AI can speed things up. But speed alone is not the goal.
The more important question is where that speed should take the industry.
A few clear themes emerged from the discussion:
Data quality needs to come before AI scale.
AI business cases should look beyond existing operating models.
Commercial models need to reflect value and outcomes, not just time saved.
People will need stronger skills in critical thinking, quality assurance and AI management.
Governance should enable safe progress, not prevent it.
Operational insight can become a meaningful source of client value.
AI should not be treated as a layer placed on top of broken systems.
The message from the roundtable was clear: agencies have a major opportunity to use AI not just to produce work faster, but to build smarter, more adaptive and more valuable businesses.
To do that, the industry needs to look beneath the surface.
It needs to fix the engine, not just the output.