/
Expert Context For Strategic AI Research with Piers Fawkes
IN CLEAR FOCUS: Marking our 350th episode and 25th season, Piers Fawkes of PSFK explores the future of strategic AI research. As generic LLMs homogenize insights into a "race to the middle," Piers explains how tracking weak signals provides a vital competitive edge for modern brand marketers. Discover how his new venture, Fodda, acts as an advanced context layer, using curated knowledge graphs and digital twins of human experts to deliver traceable AI outputs that preserve human judgment.
Episode Transcript
Adrian Tennant: Hello, I'm Adrian Tennant, host of IN CLEAR FOCUS. This is the first episode of our 25th season, the 350th episode we've published, and it marks seven years of the podcast. Whether you've been with us since the very first season in 2019, or this is the first episode you've chosen to play, thank you for listening. And thank you to every guest who has given us their time. Coming up in this episode of IN CLEAR FOCUS.
Piers Fawkes: We have these urban myths which are around two advertising agencies that went in to pitch a brand, and the brand exec is like, "They just gave me exactly the same pitch." Like the impact of AI is making everything the same.
Adrian Tennant: You're listening to IN CLEAR FOCUS, fresh perspectives on marketing and advertising, produced weekly by Bigeye, a strategy-led, full-service creative agency growing brands for clients globally. Hello, I'm your host, Adrian Tennant, Bigeye's Chief Strategy Officer. Thank you for joining us. Any brand strategist can now ask an AI assistant what's coming next in their category and get a confident, well-argued answer in seconds. But so can every one of their competitors, using the same tools and getting much the same answer. Our guest today has spent more than 20 years producing the kind of intelligence those tools now approximate, and he has a clear view on what separates a distinctive answer from a generic one. In 2004, Piers Fawkes launched PSFK, the trends and innovation firm whose research has informed strategy at Apple, Target, Walmart, and Nike, among many others. Today he's building Fodda, which supplies AI systems with expert-curated source knowledge so their answers come back specific and traceable. Piers believes the edge now belongs to strategists who bring better context to these tools, and who know where their own judgement still decides the outcome. To discuss weak signals, what AI has done to the economics of strategic research, and how marketers can get more from the tools they already have, I'm delighted that Piers is joining us today from Brooklyn, New York. Piers, welcome to IN CLEAR FOCUS.
Piers Fawkes: It's wonderful to be here, Adrian.
Adrian Tennant: Your trends and innovation firm PSFK started in 2004 as a blog after arriving in New York from the UK. What were you trying to do with it?
Piers Fawkes: I was writing, it was a writing project. I was sharing ideas. While I was waiting to get a job. I was waiting to get a visa, was trying to get paperwork, do anything, and so I had a little bit of time on my hands and like everybody who moves to a new country you're obviously filled with this new ideas- inspiration. New York City is obviously full of new ideas, and I wanted to share the things I was seeing with my friends and colleagues. And I also wanted to know what my friends around the world were seeing as well. I had friends who moved to Sydney and friends who moved to Singapore. I had friends back in London and across Europe, and I thought, “How about we create a blog and we can all post interesting things.” I didn't know what those interesting things were going to be but they are ultimately- tends to be, about creative culture consumer culture- brand culture. Um the sort of things that we would write about, and share, and we can. I just started this kind of rather collaborative at first people just sharing ideas and things they seen in their cities.
Adrian Tennant: When PSFK began, finding out what was happening in the world usually meant magazines, newspapers, or emailing people directly. Of course, fast forward to today, now there's an abundance of information arriving through algorithmic feeds. I'm curious, how has that changed what strategists actually know, do you think?
Piers Fawkes: Well, in some ways you still have the same challenges I had, we had, when I started writing. Okay. Cause there were plenty of sources of information around in 2004. But the stuff that magazines like Wired and Fast Company were writing about and the newspapers, especially then they weren't writing about the things that I was seeing ,and I was interested in and I was sharing. Yeah. And there was just like a lack. And a lot of those kind of publications I mentioned have evolved and kind of got with it, but there's always a kind of a lag. And so we were just trying to overcome and just kind of talk about things that were really happening. Which we thought were really interesting, and I guess could be leveraged by a product, a brand or whatever. But I don't think we were like- that was our motivation. Our motivation was to share interesting things with each other. I started when web publishing was democratized. We suddenly have these web blog publishing platforms which allowed anybody to share content. Today we have, multiple blog like platforms, social media platforms, that allow people to share broadly or within niches and the machines are also algorithmically scanning everything. But at the same time, like, even though everything is being discussed, everything is being talked about. What is interesting within that? What are the interesting things? What are the important things that we should leverage? And I think we always had a goal to kind of share things, which would, it sounds grandiose, but to share things that would inspire people to make the world better. It's kind of a theme that we talked about a lot when we try to justify what the heck we were doing. But it's a “We wanted people who work in advertising at people working brands.” To be inspired now and make the world better doesn't always mean like, save the planet. I mean hopefully there's always elements of progressive sustainability and progressive business but it could also be like, make someone smile, make someone happy, do service for people as well. And I think there's a need in our algorithmic world for that sort of injection. You know, I see too many people on the subway getting their news from an algorithmic feed and you begin to wonder, “Well, how is the serendipity?” “Where's the kind of jar or where's something brand new that's going to make you think?” Because you're going to go into work with the same sort of news that your colleagues have. So where are you finding inspiration and unfortunately what I worry about is because algorithmic feeds are so good that you feel satisfied you're giving your news and you're giving your insights and so you feel that you don't really need anymore. You know and so you're going into work content and using the same information everybody else has been in fact there's so much more you could use so many different angles you could use.
Adrian Tennant: Weak signals are central to how PSFK reads a market. Now, for a CPG marketer working from quarterly tracking studies and syndicated reports, how would you describe what qualifies as a weak signal?
Piers Fawkes: Yeah, I think it's important to have lots of different layers of data and so your tracking studies and your consumer pulse surveys are important, just as important. What we try to do is we're looking for signals of change, signals of innovation, new ideas, sometimes directly within an industry. So a consumer goods company might be looking for new ideas directly within its space, but it should also be looking for analogous and even alternative spaces as well and you know it maybe you look at industries the seven same customer maybe you look at projects are projects that are talking to the same people trying to communicate in different ways and so what we spent a lot of time doing is gathering the signals without overly thinking what does this mean. Yeah, we do that actually at intervals. So the weak signals are just the gathering of these new ideas. We categorize and tag those ideas so we can filter them. And then at a cadence, we'll do pattern recognition and analyze and try to look at the themes that are running through those ideas. Yeah.
Adrian Tennant: Three years ago, the economics of trend research really changed. A client could pay $20,000 or so for a custom research engagement or $20 a month for an AI subscription. Piers, what did that do to your business?
Piers Fawkes: Yeah, tell me about it. It is a significant moment in the PSFK journey is the introduction of the $20 chatbot. I think, I mean, it radically changed the business. It caused major disruption. Sometimes I talk about how I kind of lost my job to AI three years ago, on November 22, because of the introduction of ChatGPT. But, you know, at that time, and today, I was shaking my fist at the quality of the data, and the quality of the data and insights that the machine had, the system had. I kind of leaned into the dragon that was eating me or blowing fire on me. And it was like, I need to understand this thing and what it's going to do. And so what I think it did, arguably, is it made my work better because I could expand the research that I did. I think ultimately, because of my background, there's a certain sort of purview and a certain sort of point of view. And I could really only research in English, you know, 20 years ago. And if I needed to research in other markets, I had to adjust my methodology and go into experts research. Done a lot of expert analysis to do multiple and now we live in a world where everything can be read and so I can explore, cultures countries topics that are hard to understand and for example, we all know that china has this amazing automotive innovation taking place. But most of that doesn't get written about in English language articles and newsletters and stuff like that. But you can read about that every day, you know, just by kind of using your machine to translate those stories, find the stories for you. So it made my product better in terms of research. It allowed me to scrunch vast volumes of data to look for things and then layer in new themes as well. So I think eventually I came back with a better product for a different time.
Adrian Tennant: Well, you mention data quality. Marketers of course are now talking about "AI slop" and what we see is a race to the middle. Why do these general purpose large language models tend to converge on similar answers for everyone in a category?
Piers Fawkes: I am not a data scientist, but things that I think about are the LLMs and the LLM training data, which is the kind of corpus of the data that we used to get from these LLMs, was in many ways used to train the machines. It's called training data. So we were amazed we had access to all this data and all this information that the world has ever written in English, and it gave us this great encyclopedia. But it's not necessary for an LLM to have all that data and all that information. So the reason that data existed wasn't to provide knowledge to all of us, it's just to get the system to work. So there's one thing. So going forwards, the LLMs don't necessarily need to have or are going to have the world's encyclopedia within its brain, okay? It's a lot of tokens, a lot of crunching. It's just not necessary. And we already see it. We already see how the big frontier models use web search to supplement their data. Okay. So there are definitely obviously Chinese LLMs that don't have these kind of corpus of knowledge, but still work just as well. Anyway, it's a discussion around “What was that data for? What was that knowledge even for?” And if it wasn't to create the best encyclopedia ever, it was just to train the model, to get the machines to work. So if that's the case, and then no one's making sure that data is amazing. And the second thing is because it's a probabilistic machine, it's all about probability. And it's all about giving you the response that out of all the information it has on average, what the answer is going to be. Okay. So it provides an average and it races to the middle as they say. So it provides an answer that it thinks that you want. I need to expect that everybody wants the majority of people which is useful to many occasions but it might not be useful if you're trying to create a. Billion dollar bet on a brand, on advertising so that's why I need better days.
Adrian Tennant: Piers, you've discussed how you approached artificial intelligence as an opportunity to improve your business. Your latest venture is Fodda, which you describe as a context layer for AI systems. Can you break it down for us? What does that mean?
Piers Fawkes: Yeah context. There's all these new words being bandied around- literally it is providing your eye with better context better in the size of information so it can have a better output so a context layer is basically a database. Or a knowledge graph is what we call it because it's the format of the database and so it allows an AI system to query that data. And because that data is special and curated, it allows that AI system to have better results. The AI system could be an LLM that a human is using, or it could be an agentic system and agentic workflow as well.
Adrian Tennant: Plenty of strategists already upload reports and decks into their existing AI tools before they start work. What does a structured knowledge graph give them that a folder of PDFs can't?
Piers Fawkes: Yeah, well, there's a sort of technical issue around the upload of PDFs and the vectorization. So again, I am no data scientist, but the vectorization of a document, it means basically that it misses, it doesn't have signposts to how the data is connected. It's just basically, the stars in the sky with all the data is up in the sky. Okay. And so it's not a sophisticated way to search data just because you upload the PDF into your system. It's good enough system for many people. With knowledge graphs and hybrid knowledge graphs, or what we're doing is really giving signposts to the AI. We're basically tagging. Data and say, this article is connected with this company. It's connected to this date, and connected with this theme, or it's connected with this trend. And we provide these signposts, which allows the AI to find the data quicker. But more importantly, it allows the AI to see patterns that we wouldn't see otherwise, and make connections within that data. It can, what they call traverse, it looks across the data, looks across the themes, and can identify and spot new ideas and new trends. That's the promise.
Adrian Tennant: Let's take a short break. We'll be right back after this message.
 | Steven Van Belleghem: Hello, I'm Steven Van Belleghem. I'm the author of "Customer Passion: 101 Ways to Meet Customer Expectations and Stand Out," published by Kogan Page. My new book is a practical, evidence-based guide for managers, team leads, and business strategists, and it provides actionable steps for embedding customer-first behavior into everyday operations. "Customer Passion" includes real-world case studies from organizations including Disney, In-N-Out Burger, The Ritz-Carlton, and Mr. Beast. You'll learn how to turn customer-centric principles into measurable performance and how to combine automation with authentic human connection. As an IN CLEAR FOCUS listener, you can save 25 percent on "Customer Passion" when you order directly from KoganPage.com. Just enter the exclusive promo code BIGEYE25 at checkout. Shipping is always complimentary for customers in the US and in the UK. I hope my book helps you close the gap between intention and execution. Thank you. |
Adrian Tennant: Welcome back. I'm talking with Piers Fawkes, founder of PSFK and Fodda, about how marketers can get more differentiated thinking from the AI tools they're already using. Piers, every Fodda response comes back with its sources attached. Now, when you're presenting to a client or a board, why does provenance matter as much as the insight itself?
Piers Fawkes: Well, the ways I think about provenance and evidence is, one- is, you're forcing the system to give you evidence in order to stop the machine from hallucinating. Yeah? Maybe the best thing and the worst thing about AI is its desire to please. And then, or its desire to please creates hallucination. Yeah? I do not know the answer, but I want to please the user, so I'm going to give you an answer that's probably going to make you happy. So it makes things up. We use evidence and provenance to just check itself. You have to give the evidence, and you'll see this in many systems where there are lots of links back so that you can actually verify and validate. All the links are weak signals. We basically do the same thing we did 25 years ago, 20 years ago. We gather lots of weak signals, we put that in a database and then we do in a cadence pattern recognition across all of those signals. And when we do the patterns and we find the trends we wanna make sure we can link back to each one so that we don't have the example evidence to be a link to the trend of the trends be the link to the evidence. It obviously makes the work robust, but again, it allows the AI to like weave through that data. And again, maybe find something that we haven't spotted and hopefully help us get some new ideas.
Adrian Tennant: Well, Piers, we love case studies on in-clear focus. Panama Red connected FODR to their Curiosity engine and produced a non-alcoholic positioning territory for a spirits client. What did the AI get to that a base model wouldn't have?
Piers Fawkes: Panama Red is a interesting case study because they ran this process through a classic LLM and using Fodda database. And I think there's a couple of things that happened. So we talked about probabilistic and that sort of race to the average. And so the machine is going to try to provide you with the most likely scenario. So even if you're pushing and trying to get it to be creative, it's trying to be on average the most creative, rather than alternative and different. So that's one issue. And the second issue is you are using data that everybody else is using. Arguably. So when you're using LLM data to come up with pitch ideas, new ideas, the guy in the room next door, the girl in the room above, I could have come up with the same answer. And we have these urban myths that are beginning to resonate, which are around. So advertising agencies went into pitch a brand and the brand exec is like, they just gave me exactly the same pitch. So we have these myths around like how the impact of AI is making everything the same. Back to Panama Red, the second thing is because you have a different differentiated and curated feat, it provided a context, as I said, it provides context that allowed the system just to have different points and different ideas to bounce off. Now, arguably, the LLM, in its great wisdom, might actually have access to everything that Fodda has, arguably. But one of the reasons you use Fodda is that it's a curated slice. It's basically, we're like, this is the important data that's out there. And what we also have is pattern recognition. We've already done the trends analysis, and so, not overworking the machine. You've got to be careful. Some of the times you overwork the machine, you're like, OK, tell me what the trends are on eggs. It's going to shortcut every single time you try to work. Is that I could go find the evidence, I could do pack recognition, I could report the trans back to you, or I could go look it up. I do a web search for you and it's gonna try to make the easy route to please you as a user as well. So, what we're doing with Panama Red is they have an agentic workflow, which is like, here's the creative brief. We want to come up with a new product for the clients. And it has a number of different inputs and they basically plug Fodda into their process using, they call it MCP, like an API. And they basically call our data with questions in the same way you would ask questions to an LLM. You ask questions to our database and it just has richer context and better context.
Adrian Tennant: Excellent. There's real anxiety in marketing right now about what happens to human expertise when an AI can produce a trend report in minutes. Where does human judgment still decide the outcome?
Piers Fawkes: I mean, I think all of us who have used AI to create analysis have seen the kind of the hollowing effect. So where, you know, I've been preparing a presentation where I crunched all the data and I've asked the AI to develop the analysis, and I read it was fantastic. But when I tried to present it at my first presentation, I realized I didn't know what it meant. I didn't know if I believed it, things like that. Okay. I think that despite all the powers in the amazing things there's an importance to have a human in the loop now. On top of that we have all these humans running around with amazing brains full of expertise that they learned over the careers. They've learned by studying, learning by doing, and that insights forever has helped companies create edges and trade change. People have been forever asking experts for their insights. Everyone listening has probably always done an expert interview, been an expert or given an expert interview. Despite all the market data and the customer data, it just gives you an edge. It gives you a point of view. So what we're trying to do at Fodda is turn that expertise into queryable data, into queryable knowledge groups. So we find people who are particularly niches, Zoom marketers, longevity experts, and we go through a process where we turn their data, their knowledge, into queryable data with their permission, of course. And so AI systems can query that data and ask and get better insights and actually, and get better outputs.
Adrian Tennant: When we were preparing for this interview, Piers, you shared with me that you actually built Fodda's first agency demo yourself, the night before the meeting, with no developer. What's now within reach for a strategist who's never before written code?
Piers Fawkes: It's an interesting question because I built for it in November '25, and the story is: the tech person, the developer partner, had flu, and autumn for twenty five people off work for two weeks because of flu. So my partners in bed, and hear from them, I have a meeting with a massive advertising agency group. And i'm like, “If I can sort of like make this thing look like it works, maybe I should take me to not gonna cancel this meeting.” So I'm like, “I've heard about all this coding, this vibe coding and stuff like that. I wonder if I could do that myself.” And what was great is I would ask my LLM, ChatGPT or Claude, how to code. And then I would use a coder, Claude Code or Antigravity, to actually follow the instructions. So actually, today I do three. I have three systems. I have an LLM, I have a chief technology officer type coder, and then I have the coders themselves. And so we have this sort of interplay between them, and it helped me understand how to kind of build really a chatbot for that meeting. It's amazing what you can build. Now, the thing I'm thinking about today and I'm pondering is: is it the strategist's job to build these systems? I'm beginning to think about who is the experts and who should be building. You know, advertising marketing comes with these defined lines about who does what. Who's a crazy? If it's a strategist, who's production? Is media? But we seem to be in a playful time where we all playing these machines, make creating these kind of little apps and things like that. It's great. I think there's learning, understanding of scope, but just because I can post on Instagram, does that make me a social media expert? And the same way with, I'm beginning to feel with vibe coding, just because I can create an app, does that make me a developer? There's been all this conversation about the death of coding and coders, but I think it's the same thing about lawyers. A couple of years ago, everyone was saying, “We don't need lawyers anymore because we've got chat GPT.” Well, in fact, we need lawyers more than ever because it's made more complicated systems that we need to discuss and think through. And it's going to be that, I feel, in the same way with coding and web development. I think what's going to happen is we're going to need experts who know about security, who know about data, who know about token efficiency, who actually need to do this work. So I think strategists should play with this technology, but see it as a sandbox, as a playground to try to understand the scope. But I am definitely concerned about the kind of pursuits of production-worthy products that are being built by people who aren't the experts. I don't know if strategists should be building this. Strategists should be compiling data, doing the analysis, and coming up with a big idea or something. Isn't that what they're supposed to do? But what I see is a lot of kind of building and hacking and stuff like that. So if it's for playground, it's to understand what this technology can do. Great. but sometimes I worry about what people in agencies are doing or convincing themselves is what their job is.
Adrian Tennant: Right. Back to the swim lanes, in other words.
Piers Fawkes: Mm-hmm. Yeah.
Adrian Tennant: Fodda now includes human agents, digital twins built from named experts through an onboarding process and a recorded interview. Piers, what does that process capture about how somebody actually reasons?
Piers Fawkes: It's a little bit about what they know about, more than how they reason about, I think, these human agents, yeah? And to do that, we have four ways we kind of ingest—to my context, the other word is ingestion, okay? We ingest this personal data. So we do, easy one, deep research. We've all done a deep research project. So yeah, “Tell me about adrian and his love of horses. He's a horse riding expert.” It goes away, looks what you've written about, post about, blah blah. Very straightforward. One is we index, with your permission, your conversation history within your lm. It could be within a project, project within your lm, what could be brought within your lm, and you basically, you, Adrian, say, “I am an expert in horse riding,” and then the machine will run a skill and go, “Okay, when it comes to horse riding, here's the five things you know a lot about: standing on the horse's back, whatever it is.” And so it got this, that. I can't cycle pd of your knowledge around those topic areas, okay? Is you, if you are an expert, you probably been probing and questioning and talking and writing using the other, yeah. Then the third one is we do an interview, so we use the deep research and the indexing. We call it on clothing, actually, because it's... We take that data, and we create a questionnaire. We're like, “Where are the gaps?” The system goes, “Where's the gaps in this data? What don't we know about?” We write a questionnaire, and we put the expert onto an AI interview. So they're interviewed by an AI machine. It's on Google Meet. Google Meet opens up, and they get interviewed for about 10 minutes about different topics. Yes, there are more and more of these kind of AI interviews going on. And so it responds, it reacts to what people say, and we gather more data. And so we can create, ultimately, a knowledge graph around how this person thinks, and also a little bit about their tonal voice as well. So we're not trying to create bots that mimic, but we are trying to understand where the boundaries are in terms of how the expert speaks. So at that point, we're ready to have that knowledge graph queried through father. You can go on security, and your system can query that data. And then the fourth thing that we do is we keep it up to date. So every two weeks, we phone that experts, and we ask some questions. Some of those questions might be generic, like, “You know, how you thinking about what's right today?” But one might be like, “XYZ just won the derby. What do you think about that?” And so bringing kind of relevant contextual and relevant information and newsworthy information. So a Fodda user later might go, “Okay, well, I want to know about kind of the latest horse races and what's happening.” And we have up-to-date knowledge around very niche subjects and points of view. And so we just do that. We continue to kind of build the database and build their knowledge graph, and then ultimately we get more and more experts, kind of fill all the niches.
Adrian Tennant: Fascinating. Great conversation. Piers, if listeners would like to learn more about you, PSFK or Fodda, what's the best way to do so?
Piers Fawkes: Yeah, it's as simple as Fodda.ai. And Fodda is F-O-D-D-A.ai. And when you go to Fodda.ai, there's a button that says “add to Claude.” You click it once, and you go through some steps, but more or less, Fodda gets instantly added to your cloud. And you can start using it for free. Start querying all the things we just talked about and query for free. And I'd love to have you folks try it out. Give me feedback. There's a link so you can give me a call. And I'm very hands-on. I want to know what works, what doesn't work. I try to make this as an excellent product for everybody. So I really appreciate being invited today and sharing and talking about the work we're doing at Fodda.
Adrian Tennant: Piers, thank you very much for being our guest this week on IN CLEAR FOCUS.
Piers Fawkes: Thank you so much.
Adrian Tennant: Thanks again to my guest this week, Piers Fawkes, founder of PSFK and Fodda. As always, you'll find a complete transcript of our conversation with timestamps and links to the resources we discussed on the IN CLEAR FOCUS page at Bigeyeagency.com. Thank you for listening to IN CLEAR FOCUS, produced by Bigeye. I've been your host, Adrian Tennant. Until next week, goodbye.
Timestamps
00:00: Introduction to IN CLEAR FOCUS
01:09: The Role of AI in Brand Strategy
02:23: Introducing Piers Fawkes
02:44: The Origins of PSFK
04:01: Information Overload and Strategy
06:25: Understanding Weak Signals
08:28: Impact of AI on Trend Research Economics
10:48: Data Quality and AI Limitations
13:14: Fodder: A Context Layer for AI
14:31: Knowledge Graphs vs. PDFs
17:12: The Importance of Provenance
19:04: Case Study: Panama Red and Fodda
21:43: Human Expertise in the Age of AI
23:38: Building with AI: A New Frontier
27:12: Ingesting Expert Knowledge
30:35: Getting Started with Fodda
