Spark Conversations QA | Saint Lewis (VP of Client Engagement)

CFO Governance, Token Control, and Operationalizing the "Human in the Loop"

How do enterprise marketing leads scale global campaign adaptation across dozens of markets without blowing out operational expenditure? This commercial tension sat at the heart of September’s Spark Conversations roundtable in Westminster, London.

Saint Lewis, VP of Client Engagement at Spark, breaks down the practical realities of enterprise AI adoption. In this Q&A, Lewis details how CFOs can bring unpredictable API token costs under central control, why protecting brand authenticity with Gen Z and Gen Alpha audiences requires strict human-in-the-loop oversight, and how to navigate legal compliance across complex international markets.

Read the full interview below for an operational guide to balancing adaptation speed, cost control, and brand protection at scale.

Q. Saint, in your day-to-day discussions with clients around AI, what’s the split between discussions around strategic workflow versus simply keeping stakeholders informed?

A. It’s a balancing act with AI. From discussions with my industry contacts, the split is very high on keeping stakeholders informed. Especially at a senior level, given the expectation to move quickly and drive efficiency, innovation, and industry leadership.

While discussion is constant right now at operational layers, clients need to show progress and results.

Q. Cost is a constant concern, with justifiable anxiety around the unknown variable costs of "token allocations". Are clients walking blindly into spiraling bills as adoption spreads? Are we simply kicking this problem down the road by continuing to spend?

A. I think the promise that AI will "drive new efficiencies" at pace – saving money and potentially improving profit margins – has encouraged clients to embrace AI in all quarters, initially taking advantage of the "free" access AI tools are providing so they don’t get left behind. This has been a key message we’ve all seen and continue to see when it comes to AI: sink or swim.

That said, while we all get addicted to these tools and drive them through our organizations, there clearly hasn’t been the attention needed on the cost – and what do you do if you're far down the road only to realize you're into a big cost and can’t pull out? It’s a real problem if you’ve committed to transforming your business and saving money, only to realize the transformation cost negates the saving because we’re spending thousands on tokens.

The AI companies have also not made it easy to really know where free tokens end and paid tokens begin, which has caused us all to ask the question. We’ve all experienced how quickly we can burn through tokens and are therefore starting to pause on AI initiatives – certainly until we better understand the cost.

So I don’t think the problem is being kicked down the road. I simply think any CFO and other stakeholders are rightly asking, "What’s the cost going to be?" – and we can’t answer easily.

Q. Another topic that frequently arises is that pockets of strategy are still operating in silos across many organizations. Who is currently missing from the briefing table when campaigns are structured? Can everyone really live upstream?

A. In my opinion, there are two issues driving this:

It’s becoming a race to demonstrate who provides the most value and should have their hands on the tiller to drive this new ship. This will create power plays and politics. We’re seeing senior roles change quickly every day and whole departments disappear in client organizations as people try to prove their value in this new world. So in some cases, not everyone is focused on the same thing and pulling in the same direction.
Everyone has received an "AI KPI" – learn something, do something, and do it quickly. In most cases, it is uncoordinated and has no single goal to drive the business. So there are pockets of experimentation and innovation, but given we’re all moving at speed and the AI landscape is changing so quickly, nobody really knows who or what to bring to the table.

We’re seeing those who understand and control the data get the first seat and a bigger voice in the direction of travel – such as media agencies, because they're about audiences and impact, and theoretically have the data. Can everyone come armed with the same?

Q. Localization is a constant operational challenge. How do you help clients balance automated adaptation speed with genuine regional and cultural nuance? Is it really as simple as prioritization against revenue?

A. I think this still comes down to the most efficient way to do the simple stuff (where all audiences can have the same message, the same formats, etc.) versus where is the complexity that’s harder to automate and why it needs the local touch.

I think as more brands move from central to central-local operating models that can originate an idea centrally (not as a toolkit) but provide authentic, native experiences – especially in digital – this becomes less of an issue.

We’re seeing the likes of Meta working toward a "prompt-based" solution on Facebook and Instagram which would achieve a new type of adaptation on these platforms, minimize wasted spend, and optimize audience impact. You feed in the master, promote the local need, and Meta builds the content, serves it, and optimizes it – other players will follow here.

For printed comms and print-led materials like packaging, the physical cost of delivery and supply chain will still create complexity, but it’s clear there will be AI innovation that makes cost less of an issue.

Q. Industry data shows marketers spend 46% of their time on admin, reviews, and approvals. Where in the client workflow is AI actually removing this friction versus creating new review cycles?

A. Right now, it's not. That’s because of risk (thanks to the hallucinations we’re all too aware of!). A "human in the loop" at various stages of the process is needed as the trust and guarantees we need aren't there right now.

I was at the Creative Operations Summit in May this year where Will Harvey (Global Breakthrough Innovation Manager at Diageo) was saying they have a legal expert checking every 50 or so bottles of a spirit they are shipping featuring an exclusive AI-generated illustration – just to make sure they don’t have an IP issue. No fun when you’re producing thousands of bottles!

So while some admin is reduced, new admin is appearing. It’s the nature of things in our industry. I'm not sure this will change as quickly as expected. Look at Meta hiring people back that they fired thinking AI had a lot of this work covered off.

Q. When one core creative idea has to adapt into thousands of assets, how do you and your teams help clients protect brand quality when usage of AI in the creative process isn’t standardized?

A. Increasingly through a hybrid model of deploying AI-enabled QC and "humans in the loop." This depends on which clients we work with and the level of risk in assets going to market with errors.

Where we are dealing with high complexity in the number of assets, language, types of content, and markets, there are too many elements that can go wrong – so we develop tailored hybrid solutions to mitigate risk.

Until we and our clients can trust AI 100%, we will still employ trained and skilled people for some checks and balances.

Q. One of the benefits of AI is that it dramatically lowers the cost of generating content – yet much of that creative doesn’t see the light of day. Do brand teams still care about creative waste, or are they just scaling asset volume? And does it matter?

A. I think brands care about creative waste to some degree, but if AI can make content that’s deemed "good enough" for the spend, timeframe to market, and results needed, they will push ahead. We’re all seeing some awful content, but it's deemed "good enough," especially in advertising.

Where powerful ideas and authentic brand execution are concerned, brand managers are still rightfully protective and won’t let poor content out. Interestingly, in some cases clients expect several content concepts thanks to AI, whereas others only want the right content concepts irrespective of the use of AI.

I think the reality comes down to audience engagement and effectiveness. Brand leads will go with what delivers results, and at the moment there is skepticism – especially with Gen Z and Gen Alpha audiences – when it comes to AI-generated content.

Q. We know that teams are being given an "AI initiative target," with some campaigns created just to hit that metric. How do you steer clients back toward genuine brand outcomes over internal metrics?

A. It comes down to data and results. If the campaign isn’t hitting the mark because of bad creative or content, the pressure will be there to improve the outcome.

Right now, clients have permission to test and fail, so we will see this continue to happen for a while depending on the level of spend behind the campaign.

Don’t get me wrong. We can play a role in reminding clients about their brand values, quality, and impact, but as with my earlier point, sometimes "good enough" gets through.

Q. End-to-end campaign ownership is heavily fragmented. How is Spark helping clients bridge the gap between in-house teams, agencies, and regional hubs? Where does AI sit in this – is it helping or hindering?

A. It depends on who we’re working with. In some cases, we’re doing this by helping them manage data through AI into our Airtable consultancy and capability – centralizing knowledge and simplifying the visibility of data for all in the campaign ecosystem to have the right insights and make the right decisions through custom dashboards.

Sometimes it’s our know-how and years of experience of knowing what is likely around the corner, using AI to help guide all stakeholders in how to plan and move ahead.