How to use AI to create value, not volume

Full session transcript

THIS IS AN AUTOMATED TRANSCRIPT

Noz Urbina 00:03
Hello everybody, welcome to this episode of the Omnichannel X podcast. I am your host, Noz Urbina. I am here today with Rafaela Ellensburg for the first time. Welcome to the show, Rafaela.

Rafaela Ellensburg 00:43
Thank you very much. Yeah, I’m very happy to have Rafaela here. She’s making a name for herself, definitely on LinkedIn, and is becoming much more recognizable in the community of omnichannel, structured content and all the good stuff that we talk about here at Omnichannel X. So Rafaela, do you want to give the good folks a little introduction to yourself and your background?

Rafaela Ellensburg 00:43
Sure, yeah. My background is actually in content, of course. I’ve been working at one of the largest retailers in the Netherlands for nine years now—that’s Albert Heijn. What I established there in the past three years is the content engineering discipline, and actually created a formal role for myself there because I saw the necessity to structure content and make it easily findable across the organization, across the enterprise. From that findability, actually make it easier to reuse across all channels, which allows us to be much more efficient in our content operations and content management. But also from that automation, look into how can we actually personalize the content, which we want to do in an omnichannel way. That’s a bit of the journey I took and where I saw this actually needs to be a formal role to be able to do this full time and actually make sure that it’s carried throughout the organization.

Noz Urbina 01:53
Great. When you say a formal role, what was your role before?

Rafaela Ellensburg 01:58
I actually started at Albert Heijn as a content specialist, creating content for our web and applications. From that role, I was promoted into a team lead content to actually help a group of content specialists make their work more efficient, but also more tied to the strategic goals, because oftentimes it gets lost in creation. I actually helped them to measure how their content performed against the strategic goals, and how they could translate those higher level metrics into the day-to-day performance—content performance that they could track.

Noz Urbina 02:45
Great. Lots of things going on here in terms of roles and names and terminology. It’s really interesting that you leaned into content engineering, because we’ve seen a lot of people leaning into the term content design. We’ve seen people backing away from the term content strategy. I think roles and naming have always been a mess in our industry. I was quite against being called a content engineer for many years, and I’m starting to rethink—maybe it actually might be a good term, because it’s maybe one of the only ones that isn’t going to get sucked into being less about the structure and the strategic role of structuring and metadata, and that’s always core for me.

You were also talking about measurement. I really think that’s so fundamental, and one of the things that we keep forgetting about. We keep talking about how can we produce more efficiently, how can we deliver, but I like that you’re talking about measurement because for me, it comes back to why did we make this? What is the purpose for this, and is it fulfilling its purpose? If you’re not thinking about content type and structure and purpose right in the beginning, you end up producing—I’m not going to say for yourself—but you’re not producing in a way that’s directly tied to market metrics, audience metrics, people fulfilling their audience journeys.

Can you, before we get into the content engineering bit that is the implementation side, back up and tell me a little bit about how you do that bit—the tying it to the strategic metrics?

Rafaela Ellensburg 04:40
What you mostly see is that organizations and enterprises start with the higher level strategic direction that they want to go in. From that direction, you get those concepts like “we want to be more sustainable” or “we want to be more relevant for our customers.” This is every company, right? It’s not just one company that wants this—actually every company goes along those lines.

But then you need to translate those higher level goals to something that you can actually, in a tangible way, measure whether you move the needle in whatever direction you want to go. You want to be able to say, “We actually see that our customers are more engaged with specific concepts that we want to convey as a company.” We want to actually see our audience engage with content that’s about sustainability, or we actually want to see people buy more sustainable products. Those are the metrics that we are trying to move as an organization.

From those higher level goals, translated into those higher level metrics, you actually want to make it tangible that your content is linked to whatever higher level strategic goal that you have. Are you actually able to classify, for example, your content on an aggregate level of sustainability? Are you able to tag your content with sustainability if it’s about sustainability? It’s really about translating those business concepts into the content metadata structure that you bring along with the creation and storage of that content. You see how it’s all tied together, from higher level business concepts all the way down to what a content editor is actually creating to support that goal.

Noz Urbina 06:45
Yeah, and you can do that before you’ve created anything.

Rafaela Ellensburg 06:49
You can, of course. You plan out your content, right? You sit down with a marketer, for example, or campaign director, or whoever else is translating the strategic goals into a marketing plan. From the plan, there’s a phase where you’re trying to conceptually create how the content should support whatever marketing message you want to send out there to support that strategic goal. Then you’re already talking about the concepts with each other—the actual language that you use, the specific words, and how you define them, the meaning behind that. That’s something that also needs to be translated into the metadata, into the descriptive metadata, and applied to the content that is created for that marketing plan.

Noz Urbina 07:48
So you’re starting to stub out, you’re starting to create a brief, but a brief that is described by metadata, which goes into the whole lifecycle of the content. That gives you the ability to then report: from our brief, from the strategic discussion we had, I can now trace all the way through the process. How many pieces do we create against this metadata? How many modules did we create against this metadata? Where did they then go, in terms of audience metadata…

Rafaela Ellensburg 08:20
Yeah, or channels, right? The topic omnichannel—the content should be able to go everywhere, but still hold that meaning that was intended in the first place.

Noz Urbina 08:33
How do you… I know that a lot of our customers, when we show up, are struggling with the measurement of omnichannel, because these channels don’t just give up omnichannel metrics very easily. Can you talk at all about how you aggregate that data to do measurement? So you say, “We’ve got a plan, we’ve got a strategy, it’s on multiple channels,” but these channels don’t make it easy.

Rafaela Ellensburg 08:58
No, they don’t. You have to create it yourself, and the way you do that is on a semantic level. It goes back to the business concepts that I mentioned earlier, where if you want to communicate about your sustainable goals, that should also be translated into the metadata. If you do that, then you actually transform your content not just into a beautiful creative piece, but you actually make sure that it is able to be used as data. You get content as data into the system, and that allows you to measure your content on that aggregated level of that concept that you originated in the strategy.

You can actually trace back all the different content pieces, and all of the different channels that they are on, based on the semantic descriptive metadata that you apply to all of those pieces. Then you could say, “Let’s measure what our sustainability content has done on all of our channels, and which segments within the audience resonated with those pieces of content on that level.”

Noz Urbina 10:16
Let’s try to make this as concrete and maybe relatable for some of the people who are not so familiar with these concepts. Can you articulate a question or an insight that you’re able to get from this approach that you weren’t able to get before?

Rafaela Ellensburg 10:36
Yes, a very straightforward question that you would like to receive as an insight from this way of working with content is being able to say: “Let’s say you have multiple segments within your audience, and of course, you have multiple channels on which that audience is moving. The question that you want to be able to answer is, which segments within my audience are particularly interested and engaged in our sustainability content, and from that question then follows, on which channels are they?”

That enables you to target and personalize the content that audience is seeing without hindering or frustrating other audiences that are not that interested in sustainability. You allow yourself as an organization to bring forward that message to whichever person it resonates with in the market, and you’re able to do it on whichever channel that person is present. You get the relevance, and you get it at scale, at an omnichannel scale—making sure that the right message is sent to the right customer at the right moment and the right channel. That is the marketer’s dream, right? That’s what we all want.

Noz Urbina 12:13
I think this is where now we get into the content engineering part. You’ve got that strategy, you want to be able to answer those kinds of questions, fulfill the marketer’s dream. You’ve compared content engineering to managing traditional supply chains. Could you elaborate on that a little bit?

Rafaela Ellensburg 12:38
Yeah, and I think this will actually help people to understand what I’m talking about. I like to compare content to products. People know products—they know shopping, they know logistics, they know that products are created somewhere and then have to be refined before they get to the stores. It’s something that people can grasp, but we can do the same thing for content.

If you look at a supply chain from the moment that a product is created, it also gets some sort of identifier. It needs to be traceable throughout the system, and this could be for security or risk reasons or even legal reasons, but we need to be able to trace where the product originated, and we do this by identification. At the point where a product is created, it moves into some sort of refinery. We need to refine the product, put our brand on it, put some labels on it, information before it actually gets stored in a warehouse—a central warehouse from which it gets distributed to the stores, where a customer can actually buy the physical product and take it home.

We can apply the same pattern, the same flow or structure or value chain to content as well. Someone is creating content, and if you do it well, you are actually able to place an identifier on whatever piece of content that you’ve created. Oftentimes we do not do this in our discipline—we just create content, store it somewhere in whatever system, without knowing how it’s identified, how it’s tagged, how it’s classified.

Noz Urbina 15:02
Or just a minimal amount—just the file name. People have trouble imagining how many more tags could go on a piece of content. We’re tagging our stuff. We said, “It’s this campaign for this quarter, for this market, for this product set. Look, I tagged it.” But there’s more. There’s so much more. When you go into the body of the content, and you get into the topical nature of the content and how topics relate to other topics, there’s so much more that you can do.

Rafaela Ellensburg 15:31
Yeah, that’s exactly how it works, and that’s also what we see with products. If you would really dive into all of the metadata that is hiding behind products, it will amaze you. It also will amaze you how this enables businesses to work with these products in an omnichannel way, because they can track everything. They know where the product is within the entire value chain. They know when the product is getting bought by a customer. They know how many times a customer has bought that product, and that allows them to actually automate a lot of the workflows that go into distributing and actually selling these products—optimize, automate, but also personalize.

A lot of times when vendors of CMSs talk about personalization, what they actually mean is, “We can personalize the products that you show to the customers.” Of course, that’s the easy part, because all that product data has already been structured and tagged, so the data quality is high, which allows us to do a lot more with how we bring the products to the customer in a scalable way. If we would apply those same principles to our content and actually make our content data quality as high as how we treat our product data quality, we can get those same benefits from our content. That’s how I always try to translate the importance of content structuring for enabling that marketer’s dream that we have for our content.

Noz Urbina 17:18
You tell a nice story there about content engineering and management and how it’s the content supply chain. I really like that a lot. I enjoy those kinds of metaphors and that way of framing it. I think a lot of people can see how that maps to content management and content production lifecycle and distribution lifecycle. But you also talk a lot about other things that have a certain complexity to them. They’re different—I don’t like saying that they’re more complicated, I just think they’re maybe not familiar to as many people—concepts like ontologies, concept models, knowledge graphs, etc.

Can you explain a little bit how these fit into your work, and how does that layer onto this content management story that you just talked about?

Rafaela Ellensburg 18:17
When I dive into the world of semantics and knowledge graphs and taxonomies and ontologies, it’s a world that has only recently become a bit more familiar to me. I just started two years ago diving into these frameworks because at some point there was an end to how much I could do in terms of structuring content on a content management level.

The content is not just a siloed domain. If you look at how it is exposed on our digital screens, it’s actually a combination of different domains coming together. It’s not just content, but it’s also campaigns, and it’s also customers, and the data that resides within these domains needs to talk to each other. That’s how I stepped into the world of semantics, because what I saw was it’s not just content as data, but it’s actually content as linked data, and that linking part is something that we can do across semantics.

When I say, “I have a piece of content, I stored it in my database and I added a tag about sustainability”—that tag actually semantically means something. It carries a definition. What is sustainability, and what is sustainability specifically to our organization? If you look at it from that angle, and you are a company or an organization that not just creates content, but also sells products and also runs marketing campaigns, well, then you can say sustainability is not just about this piece of content. It’s actually also about what it means to our products, or what it means to our marketing campaigns that are also connected to our strategic goals.

If you want to connect those dots between those different domains, then you come into the world of semantics, where actually a knowledge graph could help you store that meaning about a business concept and use it in different systems. You have this single version of truth, basically, where you make an agreement about how you define certain concepts, such as sustainability in your organization. From that anchoring and that grounding of that agreement that you have about the concept, you can actually help all of those other systems to collaborate and communicate with each other in a consistent way. That actually helps you to share that consistent message from your strategy through your content, but also connected to how your products and customers are related to sustainability.

Noz Urbina 21:35
So we’re layering onto content management, concept management.

Rafaela Ellensburg 21:41
Yes, so I like to say concepts, not content. It’s really about… there’s room for both. We have to manage our content, but this power of managing the concepts—the core, fundamental concepts of our domain—they’re going to come up again and again, across campaigns, across product lines. Then how do they relate to each other? Because I can imagine sustainability could be related to the concept of supply chain, for example. If I’m buying food and I care about sustainability, how was that food produced? How is that food sourced? We have related ideas of sourcing, of supply chain, of different regions linked to all of it. All that stuff cuts across—it could cut across any content, multiple systems.

Rafaela Ellensburg 22:36
Multiple systems, yeah, and multiple ways of analyzing whatever happens to that concept within your organization, and whenever you communicate it to your audience. You can actually track whether your efforts in having a marketing strategy and a content strategy around sustainability actually moves those metrics in the right directions—not just how people behave, but also how your organization operates. You also have sustainability metrics within the supply chain, and whatever you do with your products to make them more sustainable also ties into how operations are made more sustainable around it. You actually tangibly achieve that strategic goal by influencing the behavior of people, and how that actually gets people to buy more sustainable products, and you actually become a more sustainable company.

Noz Urbina 23:39
Let’s come back to the practical example idea. What does the knowledge graph allow you to know or track that wouldn’t be there otherwise? If you compare—we did everything we were doing in content management. What were the questions and insights that you wanted to be able to get, or the capabilities you wanted to be able to get that said, “Now we need to move into this knowledge graph stuff”?

Rafaela Ellensburg 24:14
I think it definitely comes back to the cross-domain questions that you have. Sustainability is a concept, and not so much a business domain. But looking at business domains, you could say legal has something to do with sustainability. Content wants to communicate about sustainability. From our products department, we want to sell more sustainable products. Those are all different domains within the business, and they come together at the point where the customer interacts with the business. That’s mostly the content. Everything that we have on our screens, on our phones, on our computers—it’s always a company communicating with us as a customer, and that goes through content.

Content is the ideal place where you can actually see those domains coming together. What the knowledge graph does is connect the dots between those domains. You’re creating relationships between those domains. That’s the power of the knowledge graph—by saying, “My content is related to my products, and I do this in a way that I say that through this piece of content, I want to sell more sustainable products to customers who are interested in sustainability within their lives.” Those are the semantic connections that you make from content to products to customer, where you can actually see those bridges being built between those silos, just by mentioning the goal that you want to achieve.

Noz Urbina 26:11
I think that you’re highlighting the fact that it’s concept and the links, the relation links—the objects, but more information about all the many, many different relationships. You can’t add infinite tags. Sometimes you need to say, “I’ve added these dozen tags, but that actually, because of what I’ve done in my concept management, implies all these other relationships.” I can know that I’ve tagged this thing with 10 tags, but there could be 100 different implications of that, because this sustainability is related to this product, is related to this legal framework, is related to this company policy, is related to this bit of knowledge we have on our knowledge base, etc.

I like to think that we want tagged, linked, connected content. The content management system can only allow us to do what is practical within a piece of content, but we can then manage hundreds of links, or maybe thousands of links and relationships, if we’re managing them separately in this other system.

Rafaela Ellensburg 27:26
Yes. If we are managing them at a higher conceptual level, then suddenly you get all of these different combinations of data points that are coming from the databases. At the top level, it’s just a few concepts that are important and that are also stable for a business—they’re stable concepts. They don’t change that much, but they are related in some way to each other. That is what you do at the top level with the ontology. I’m mentioning this word that people are a bit scared of, but that’s actually what you do in the knowledge graph at the most conceptual level.

From those connections between those concepts, you drill down to a lower level, which is more the taxonomy classification level that is more familiar to people. When I say a taxonomy—for example, you can have a taxonomy about dishes from a certain cuisine. Let’s move to the more relatable world of food. You can have cuisine from Europe, for example. In Europe, of course, you have different countries that also have their own kitchen, basically, where they have specific types of dishes that they create.

For example, if I would move towards the Netherlands as a cuisine—Dutch cuisine—I could say the Dutch cuisine is a part of the European cuisine that we know, or the Western cuisine. Within that Dutch cuisine, I have certain types of dishes that are typically Dutch. You could have pannenkoeken—pancakes. We also have pancakes in other parts of the world, but here we consider them as a typically Dutch dish. Or you could have stamppot, which is actually something that you cannot translate to other languages, because that is a very core Dutch dish that you would find within that dish taxonomy that is connected to the Dutch cuisine.

Here you can already see how you can classify specific dishes and categorize them in specific cuisines. That actually helps you to infinitely add recipes and categorize them into those different cuisines or different dish types. I think that is the part that helps us link our very dynamic data that we have in the databases and connect it to something that stays. A cuisine always stays as a concept, and a dish always stays as a concept. That doesn’t change much, but you can actually add many variants of those dishes and categorize them in that way to make the system understand how we as humans think about those dishes in reality and actually make it more relevant to whoever needs to consume it.

Noz Urbina 30:49
I can envision a lot based on what you’re saying. If we have content like recipes, campaigns, sustainability information, etc., and we have these core concepts like sustainability, cuisines, dishes, and then the products themselves—a potato is a kind of concept, but then you have lots of specific potato products. You could then… if we come to how this manifests in the experience, this ability to get… “I’m looking for potato dishes. I have potatoes, but I want to explore what I could do with them across cuisines.” Give me potato dishes, and then allow me to combine other things that I’m interested in. But then the system knows that I’m interested in sustainability, so it’s going to promote sustainable other recipes that have sustainable ingredients that go with my potatoes.

Then, because it understands all these relationships, if I wanted to swap that… I’m like, “I want to do potato stew, but I’ve got all of these things in my cart. Maybe I want to change my mind now and go, I don’t want to do a European potato stew—I want an African potato stew.” Then something like a recommendation algorithm or an AI can help me swap over all that.

I find that a nightmare—if I’m buying a complex product, like all the dishes for a recipe, or if I was buying a bed, and I have the sheets and the pillows and everything that goes underneath, and I’m like, “Add to cart, add to cart, add to cart, add to cart.” Then I go, “Oh no, I want a different size.” I have to go back, because websites don’t understand the relationship between the larger version, or the ability to switch from the Polish potato stew to an African potato stew. They don’t understand that those things are related, or they can’t help me reimagine what I’m trying to accomplish.

Rafaela Ellensburg 32:55
Exactly, yeah. That’s exactly what it does—to think about data in networks of connected and related concepts instead of just rows and columns in a relational database.

Noz Urbina 33:11
Yes, awesome. In terms of AI, how do you see this helping AI? Do you have any practical thoughts on the relationship between these?

Rafaela Ellensburg 33:24
What I think is that AI actually thrives on structured data, because what we want from AI is mostly—especially if you look at enterprises, organizations—they want reliability and consistency from AI, because we’re not betting or gambling with our business outcomes. We want specific, traceable insights that actually help us to do our work better, but also to be more effective in our work.

I think where AI is actually helped with things like knowledge graphs and structured data is that we give the AI a representation of our reality and what we are trying to achieve. We are telling it, in the way that we structure our data, and then the AI can leverage that and scale it for us. What we can actually do is by working in a more structured way with our data and our content, we actually help this human-machine collaboration to be more human-friendly to begin with, but also more reliable, and to make it able for us to trust whatever the AI is doing with our input.

But for that to happen, our input needs to be of high quality. We as humans actually have work to do to make our data of AI quality—more complete, richer, more consistent and truthful, so that whatever the AI does with that data, it becomes better. You do not get garbage in, garbage out, but you get value in, value out. I think that is where we need to work towards.

Noz Urbina 35:27
I love that you’ve raised several concepts which I would love to go into, but I think we’ve got to wrap it there. We’ve talked about truth, we’ve talked about trust, we’ve talked about using AI and giving it a more reliable input, because these things are statistical engines. They’re just looking at patterns that they see. Do you really want to leave it up to statistical chance and hope that the patterns are there, or do you want to give it the model of the reality? “This is how I want you to see the world, because this is how the world is defined.”

Rafaela Ellensburg 36:07
I think it’s also a bit misleading that we call them large language models, but they’re actually very statistical in nature, and they’re missing a large part of the linguistics that actually goes into how humans understand reality and communicate about reality, which is based in semantics and structures.

Noz Urbina 36:31
Fantastic. I hope that you’re feeling like you want to hear more from Rafaela, because I do, and we’re going to do that over on the Truth Collapse podcast. But before we go over there—this one will be on omnichannel x dot digital, on OmniX podcast, and then over at truth collapse.com we have the Truth Collapse podcast where we talk about how we deal with truth and trust in an AI-mediated world, so please do feel free to follow us over there.

Rafaela, how can people get in touch with you and learn more about you?

Rafaela Ellensburg 37:09
Well, LinkedIn is an obvious choice to connect with me, but recently, I also started my own company, which is called The Content Engineering Agency. You can also go to my website—it’s https://www.the-ceagency.com. That’s also a way to reach out to me.

Noz Urbina 37:32
Fantastic. The-dash-ce-agency.com. Great. Thank you, Rafaela, so much for your time and thank you everybody for joining us.

Rafaela Ellensburg 37:42
Thank you very much. Bye bye, folks.