of the United Kingdom’s capitol city.
No concept is more important to the future of commerce than Artificial Intelligence.
Consequently, no term has become more meaningless.
Issue 003 of Tabula Rasa is my humble attempt to define what the hell we’ve meant, historically, when we say “artificial intelligence” (hereafter AI). I’ll also unpack what we mean today— and what it will mean in the future when we utter the painful cliche that AI will “transform commerce.”
For retailers and brands, the umbrella term of AI has been used to define at least three distinct technological leaps forward, each more significant than the last.
The CRO Masquerade: AI's First Act (2015-2022)
For the last decade, when somebody was selling you AI as a retailer or brand, they were effectively selling conversion rate optimization (CRO) tech in some way, shape, or form. Long before GPT, every software that aimed to increase average revenue per visitor billed itself as “AI-powered”, a hodgepodge of conversion rate hacks that the vendor-industrial complex collectively defined as artificial intelligence.
To be clear, much of this technology was incredibly valuable to commerce– automating reasonably basic CRO and product recommendations testing unlocked billions of dollars of incremental value for retailers and spawned a host of nine figure software exits. Whether or not any of this tech was truly “AI” is a purely semantic discussion; what matters is that the value proposition is fundamentally different from what is now being sold under the AI banner.
The first generation of “AI in commerce” effectively substituted the term artificial intelligence for conversion rate optimization.
Throughout this era, the grand promise of AI was that it would enable true 1:1 personalization at scale, where every visitor to a website or app received a fully customized “experience.” In retrospect, this was the wrong north star for what artificial intelligence in commerce could achieve. The folly of how most brands approach 1:1 personalization is a complex topic that deserves its own column, but said simply, it comes down to this: You can track all the customer data in the world, but most of the time, when somebody has just put a hamburger in their shopping cart, the thing they are most likely to buy with it is french fries.
There are higher and greater uses for AI which are now beginning to manifest with the rise of Chat GPT and other large language models.
Which leads to point number two.
Fast, Cheap & In Control: The Efficiency Era (2022-2023)
The second (current) generation of “AI in commerce” is running the last decade’s playbook radically cheaper, faster, and better.
Before Chat GPT was released in 2022, the grand narratives of the technology industry were meaningless, manufactured culture wars about moving to Miami. If Open AI did nothing else, the company snapped an entire industry out of a dangerous state of ennui left by delusions of grandeur about NFTs and the metaverse. For that, we should be eternally grateful.
For commerce operators, the main upshot of generative AI to date has simply been making us significantly faster. At their Edge Summit earlier this month, Bloomreach unveiled a tool called Loomi, a perfect microcosm of AI in commerce today. Loomi essentially crafts a bespoke customer journey campaign in minutes. Workflows that would previously take savvy marketers or merchandisers days to weeks can now be done in minutes. By and large, however, the actual end shopper-facing experiences that Loomi will deploy are essentially the same set of tactics and widgets we’ve seen on our favorite websites for the past decade.
Zooming out, in this current AI wave, customer service is the first major battleground. Companies like Sierra, Decagon, Maven, and Siena have collectively raised hundreds of millions of dollars alongside incumbents like Zendesk and Gorgias to deploy LLMs for retailers and brands to build more efficient CX workflows. These companies are impressive at the technical level and seem to be delivering actual, transformative results for brands like Simple Modern. (A case study detailing Siena’s work with Simple Modern is here).
But for now, they are all just plugging into the existing commerce schema. It’s the standard chat interface; only now it’s powered by an LLM instead of a human on the other side of an Intercom widget.
In a similar vein, DTC darlings AdoreMe and Curology are working with a company called Writer (rumored to be raising at a $2B valuation) to put generative AI at the center of their current workflows. Those flows being mainly around significantly increasing the scale of content creation that is SEO optimized, on-brand and intelligently assisted by custom LLMs. Remember, we started this series by talking about how the era of arbitraging paid traffic from platforms is ending — making investments in best-in-class organic content paramount.
When somebody has just put a hamburger in their shopping cart, the thing they are most likely to buy with it is french fries.
By and large, the next couple of years of AI adoption in commerce will look a lot like the Simple Modern and AdoreMe case studies above. These are still the DTC-era, performance marketing-powered brands finally achieving escape velocity. There are a plethora of manual, arcane workflows at the center of commerce that generative AI startups will make radically cheaper and faster. “Automatically find, copy, and optimize winning ads” from top brands might not be the most audacious tagline, but a product like Icon is bound to have a near-instant product market fit.
Many software businesses like this will be born in the next year or two and skyrocket to $10M+ in ARR with very lean teams. Much like the first wave of CRO tools, we’re talking about tens of billions of dollars in incremental market capitalization, even if the innovation stops here. But this is not the truly interesting futurism you read this publication for.
Ten years ago, technology couldn’t quite match the scale of imagination, so we solved the most lucrative, lowest-hanging fruit problems first. Now, it feels like retailers’ collective imaginations can’t quite keep up with the technology.
That is slowly beginning to change.
The Fifth Avenue Problem: AI's Next Frontier (2024 - ???)
The next generation of “AI in commerce” will be building the backend data layer that finally brings Fifth Avenue online.
Several years ago, analyst Benedict Evans quipped that while “the Internet lets you live in Wisconsin and buy anything you could buy in New York, it doesn’t let you shop the way you could shop in New York.”
Any current deployment of AI does not solve this problem. At the root of this problem is that the current architecture of eCommerce search simply cannot comprehend how people want to discover products.
The most basic example here is that if you search Sephora or Amazon for Meghan Markle’s favorite lipstick, you get a hodgepodge of seemingly random results. Yet, over the past 5+ years, every major women’s and lifestyle magazine has published the correct answer: Charlotte Tilbury Matte Revolution in Very Victoria. Google is, of course, the only “retailer” that can answer this query effectively, making it even all the more remarkable that they’ve fumbled the bag in commerce over the last decade.
Solving the semantic search problem is effectively the entire thesis of Daydream, a new startup from Stitch Fix and Nordstrom exec Julie Borstein that has raised $50M to craft an AI-powered search engine purpose-built for commerce. Daydream is an interesting concept, but I’d argue that advancements in GenAI will make the types of searches Daydream aims to empower even more decentralized.
In the not-too-distant future, LLMs will go from writing PDP copy and blog content to effectively creating training data and contextual information around each product SKU to help power entirely new ways of engaging with shoppers. This will be the third generation and killer application of AI in retail.
We’re rapidly seeing glimpses of what the future will look like as enterprise retailers–who for 20 years were hesitant to change so much as a pixel– are playing around with tests that change the nature of product search. Both Safeway and Ulta Beauty are working with a partner called ownit.ai to tackle product searches that would have previously been impossible, serving relevant results for queries like ‘quick dinner ideas for vegetarians” and “beach day essentials” and augmenting the context behind displaying certain products with AI-generated content. It’s a bit clunky and overwhelming for now but you can see how this will be compelling as the LLMs fine-tune the details.
For now, Meghan Markle’s favorite lipstick remains a mystery; if only because the commercial nature of search has deemed it to be a set of search terms that are so valuable that it becomes undiscoverable.
The competition to create the first truly game-changing deployment of AI is wide open. Amazon has every conceivable data advantage known to man and their first deployments of Rufus were almost comically bad. While they are rapidly improving, AI is the most obvious place where the innovator’s dilemma may finally come for Amazon. Amazon’s business is essentially geared towards showing as many product recommendations and ads as humanly possible, in an interface that cuts against the grain of where GenAI wants to pull retail.
Still, that isn’t stopping Amazon, a company that generally feels besieged by a “Day 2” mentality everywhere else across its business, from thinking first principles about product discovery. For its part, Walmart is developing proprietary, retail-specific LLMs to power new experiments across search, support, and, yes, personalization.
For AI to truly upend the form factor of commerce, two things will need to transpire in the coming years:
- Enterprise workflows for LLMs will need to be purpose-built for retail and CPG. Currently retail is mostly an afterthought vertical for OpenAI & Anthropic, who have first turned their attention to selling myopically more lucrative opportunities to major players in technology & finance. This creates an interesting opportunity for challenger companies like the aforementioned Writer who deploy their own models and have found some early traction with brands.
- Brands will need to rethink the entire concept of an eCommerce journey long before it feels sensible to do so. Our efforts to train a customer “how” to shop online have hemmed us in on every side from exercising any creativity or fluidity in that experience. If we don’t begin the change now, we may find the next generation leapfrogging the “old Web” format of pages into more open world and ‘gaming-aesthetic experiences.
True Classic, perhaps the last great winner of the “build a brand on Meta era” embodies this perfectly. Last month, President Ben Yahalom called for Generative AI leaders to deploy cutting-edge tools to “elevate every aspect of the company.” Already, they have several AI tests live, the most prominent with a company called Zowie, which prompts users to enter natural language questions and recommends products and sizing guides.
I found it to be an efficient and helpful (if uninspiring) experience.
Most of these early-stage AI deployments will likely fail against the familiar conversion pathways but damn it; it’s fun to see brands trying zany stuff again.
The moment for AI to truly transform commerce will be born out of necessity as much as technological innovation. When Facebook and Amazon ads printed predictable amounts of cash, there was little reason for “AI” to be anything more than CRO for ads.
The fun starts when the easy performance marketing funnels dry up… which is already happening.
No concept is more important to the future of commerce than Artificial Intelligence.
Consequently, no term has become more meaningless.
Issue 003 of Tabula Rasa is my humble attempt to define what the hell we’ve meant, historically, when we say “artificial intelligence” (hereafter AI). I’ll also unpack what we mean today— and what it will mean in the future when we utter the painful cliche that AI will “transform commerce.”
For retailers and brands, the umbrella term of AI has been used to define at least three distinct technological leaps forward, each more significant than the last.
The CRO Masquerade: AI's First Act (2015-2022)
For the last decade, when somebody was selling you AI as a retailer or brand, they were effectively selling conversion rate optimization (CRO) tech in some way, shape, or form. Long before GPT, every software that aimed to increase average revenue per visitor billed itself as “AI-powered”, a hodgepodge of conversion rate hacks that the vendor-industrial complex collectively defined as artificial intelligence.
To be clear, much of this technology was incredibly valuable to commerce– automating reasonably basic CRO and product recommendations testing unlocked billions of dollars of incremental value for retailers and spawned a host of nine figure software exits. Whether or not any of this tech was truly “AI” is a purely semantic discussion; what matters is that the value proposition is fundamentally different from what is now being sold under the AI banner.
The first generation of “AI in commerce” effectively substituted the term artificial intelligence for conversion rate optimization.
Throughout this era, the grand promise of AI was that it would enable true 1:1 personalization at scale, where every visitor to a website or app received a fully customized “experience.” In retrospect, this was the wrong north star for what artificial intelligence in commerce could achieve. The folly of how most brands approach 1:1 personalization is a complex topic that deserves its own column, but said simply, it comes down to this: You can track all the customer data in the world, but most of the time, when somebody has just put a hamburger in their shopping cart, the thing they are most likely to buy with it is french fries.
There are higher and greater uses for AI which are now beginning to manifest with the rise of Chat GPT and other large language models.
Which leads to point number two.
Fast, Cheap & In Control: The Efficiency Era (2022-2023)
The second (current) generation of “AI in commerce” is running the last decade’s playbook radically cheaper, faster, and better.
Before Chat GPT was released in 2022, the grand narratives of the technology industry were meaningless, manufactured culture wars about moving to Miami. If Open AI did nothing else, the company snapped an entire industry out of a dangerous state of ennui left by delusions of grandeur about NFTs and the metaverse. For that, we should be eternally grateful.
For commerce operators, the main upshot of generative AI to date has simply been making us significantly faster. At their Edge Summit earlier this month, Bloomreach unveiled a tool called Loomi, a perfect microcosm of AI in commerce today. Loomi essentially crafts a bespoke customer journey campaign in minutes. Workflows that would previously take savvy marketers or merchandisers days to weeks can now be done in minutes. By and large, however, the actual end shopper-facing experiences that Loomi will deploy are essentially the same set of tactics and widgets we’ve seen on our favorite websites for the past decade.
Zooming out, in this current AI wave, customer service is the first major battleground. Companies like Sierra, Decagon, Maven, and Siena have collectively raised hundreds of millions of dollars alongside incumbents like Zendesk and Gorgias to deploy LLMs for retailers and brands to build more efficient CX workflows. These companies are impressive at the technical level and seem to be delivering actual, transformative results for brands like Simple Modern. (A case study detailing Siena’s work with Simple Modern is here).
But for now, they are all just plugging into the existing commerce schema. It’s the standard chat interface; only now it’s powered by an LLM instead of a human on the other side of an Intercom widget.
In a similar vein, DTC darlings AdoreMe and Curology are working with a company called Writer (rumored to be raising at a $2B valuation) to put generative AI at the center of their current workflows. Those flows being mainly around significantly increasing the scale of content creation that is SEO optimized, on-brand and intelligently assisted by custom LLMs. Remember, we started this series by talking about how the era of arbitraging paid traffic from platforms is ending — making investments in best-in-class organic content paramount.
When somebody has just put a hamburger in their shopping cart, the thing they are most likely to buy with it is french fries.
By and large, the next couple of years of AI adoption in commerce will look a lot like the Simple Modern and AdoreMe case studies above. These are still the DTC-era, performance marketing-powered brands finally achieving escape velocity. There are a plethora of manual, arcane workflows at the center of commerce that generative AI startups will make radically cheaper and faster. “Automatically find, copy, and optimize winning ads” from top brands might not be the most audacious tagline, but a product like Icon is bound to have a near-instant product market fit.
Many software businesses like this will be born in the next year or two and skyrocket to $10M+ in ARR with very lean teams. Much like the first wave of CRO tools, we’re talking about tens of billions of dollars in incremental market capitalization, even if the innovation stops here. But this is not the truly interesting futurism you read this publication for.
Ten years ago, technology couldn’t quite match the scale of imagination, so we solved the most lucrative, lowest-hanging fruit problems first. Now, it feels like retailers’ collective imaginations can’t quite keep up with the technology.
That is slowly beginning to change.
The Fifth Avenue Problem: AI's Next Frontier (2024 - ???)
The next generation of “AI in commerce” will be building the backend data layer that finally brings Fifth Avenue online.
Several years ago, analyst Benedict Evans quipped that while “the Internet lets you live in Wisconsin and buy anything you could buy in New York, it doesn’t let you shop the way you could shop in New York.”
Any current deployment of AI does not solve this problem. At the root of this problem is that the current architecture of eCommerce search simply cannot comprehend how people want to discover products.
The most basic example here is that if you search Sephora or Amazon for Meghan Markle’s favorite lipstick, you get a hodgepodge of seemingly random results. Yet, over the past 5+ years, every major women’s and lifestyle magazine has published the correct answer: Charlotte Tilbury Matte Revolution in Very Victoria. Google is, of course, the only “retailer” that can answer this query effectively, making it even all the more remarkable that they’ve fumbled the bag in commerce over the last decade.
Solving the semantic search problem is effectively the entire thesis of Daydream, a new startup from Stitch Fix and Nordstrom exec Julie Borstein that has raised $50M to craft an AI-powered search engine purpose-built for commerce. Daydream is an interesting concept, but I’d argue that advancements in GenAI will make the types of searches Daydream aims to empower even more decentralized.
In the not-too-distant future, LLMs will go from writing PDP copy and blog content to effectively creating training data and contextual information around each product SKU to help power entirely new ways of engaging with shoppers. This will be the third generation and killer application of AI in retail.
We’re rapidly seeing glimpses of what the future will look like as enterprise retailers–who for 20 years were hesitant to change so much as a pixel– are playing around with tests that change the nature of product search. Both Safeway and Ulta Beauty are working with a partner called ownit.ai to tackle product searches that would have previously been impossible, serving relevant results for queries like ‘quick dinner ideas for vegetarians” and “beach day essentials” and augmenting the context behind displaying certain products with AI-generated content. It’s a bit clunky and overwhelming for now but you can see how this will be compelling as the LLMs fine-tune the details.
For now, Meghan Markle’s favorite lipstick remains a mystery; if only because the commercial nature of search has deemed it to be a set of search terms that are so valuable that it becomes undiscoverable.
The competition to create the first truly game-changing deployment of AI is wide open. Amazon has every conceivable data advantage known to man and their first deployments of Rufus were almost comically bad. While they are rapidly improving, AI is the most obvious place where the innovator’s dilemma may finally come for Amazon. Amazon’s business is essentially geared towards showing as many product recommendations and ads as humanly possible, in an interface that cuts against the grain of where GenAI wants to pull retail.
Still, that isn’t stopping Amazon, a company that generally feels besieged by a “Day 2” mentality everywhere else across its business, from thinking first principles about product discovery. For its part, Walmart is developing proprietary, retail-specific LLMs to power new experiments across search, support, and, yes, personalization.
For AI to truly upend the form factor of commerce, two things will need to transpire in the coming years:
- Enterprise workflows for LLMs will need to be purpose-built for retail and CPG. Currently retail is mostly an afterthought vertical for OpenAI & Anthropic, who have first turned their attention to selling myopically more lucrative opportunities to major players in technology & finance. This creates an interesting opportunity for challenger companies like the aforementioned Writer who deploy their own models and have found some early traction with brands.
- Brands will need to rethink the entire concept of an eCommerce journey long before it feels sensible to do so. Our efforts to train a customer “how” to shop online have hemmed us in on every side from exercising any creativity or fluidity in that experience. If we don’t begin the change now, we may find the next generation leapfrogging the “old Web” format of pages into more open world and ‘gaming-aesthetic experiences.
True Classic, perhaps the last great winner of the “build a brand on Meta era” embodies this perfectly. Last month, President Ben Yahalom called for Generative AI leaders to deploy cutting-edge tools to “elevate every aspect of the company.” Already, they have several AI tests live, the most prominent with a company called Zowie, which prompts users to enter natural language questions and recommends products and sizing guides.
I found it to be an efficient and helpful (if uninspiring) experience.
Most of these early-stage AI deployments will likely fail against the familiar conversion pathways but damn it; it’s fun to see brands trying zany stuff again.
The moment for AI to truly transform commerce will be born out of necessity as much as technological innovation. When Facebook and Amazon ads printed predictable amounts of cash, there was little reason for “AI” to be anything more than CRO for ads.
The fun starts when the easy performance marketing funnels dry up… which is already happening.
No concept is more important to the future of commerce than Artificial Intelligence.
Consequently, no term has become more meaningless.
Issue 003 of Tabula Rasa is my humble attempt to define what the hell we’ve meant, historically, when we say “artificial intelligence” (hereafter AI). I’ll also unpack what we mean today— and what it will mean in the future when we utter the painful cliche that AI will “transform commerce.”
For retailers and brands, the umbrella term of AI has been used to define at least three distinct technological leaps forward, each more significant than the last.
The CRO Masquerade: AI's First Act (2015-2022)
For the last decade, when somebody was selling you AI as a retailer or brand, they were effectively selling conversion rate optimization (CRO) tech in some way, shape, or form. Long before GPT, every software that aimed to increase average revenue per visitor billed itself as “AI-powered”, a hodgepodge of conversion rate hacks that the vendor-industrial complex collectively defined as artificial intelligence.
To be clear, much of this technology was incredibly valuable to commerce– automating reasonably basic CRO and product recommendations testing unlocked billions of dollars of incremental value for retailers and spawned a host of nine figure software exits. Whether or not any of this tech was truly “AI” is a purely semantic discussion; what matters is that the value proposition is fundamentally different from what is now being sold under the AI banner.
The first generation of “AI in commerce” effectively substituted the term artificial intelligence for conversion rate optimization.
Throughout this era, the grand promise of AI was that it would enable true 1:1 personalization at scale, where every visitor to a website or app received a fully customized “experience.” In retrospect, this was the wrong north star for what artificial intelligence in commerce could achieve. The folly of how most brands approach 1:1 personalization is a complex topic that deserves its own column, but said simply, it comes down to this: You can track all the customer data in the world, but most of the time, when somebody has just put a hamburger in their shopping cart, the thing they are most likely to buy with it is french fries.
There are higher and greater uses for AI which are now beginning to manifest with the rise of Chat GPT and other large language models.
Which leads to point number two.
Fast, Cheap & In Control: The Efficiency Era (2022-2023)
The second (current) generation of “AI in commerce” is running the last decade’s playbook radically cheaper, faster, and better.
Before Chat GPT was released in 2022, the grand narratives of the technology industry were meaningless, manufactured culture wars about moving to Miami. If Open AI did nothing else, the company snapped an entire industry out of a dangerous state of ennui left by delusions of grandeur about NFTs and the metaverse. For that, we should be eternally grateful.
For commerce operators, the main upshot of generative AI to date has simply been making us significantly faster. At their Edge Summit earlier this month, Bloomreach unveiled a tool called Loomi, a perfect microcosm of AI in commerce today. Loomi essentially crafts a bespoke customer journey campaign in minutes. Workflows that would previously take savvy marketers or merchandisers days to weeks can now be done in minutes. By and large, however, the actual end shopper-facing experiences that Loomi will deploy are essentially the same set of tactics and widgets we’ve seen on our favorite websites for the past decade.
Zooming out, in this current AI wave, customer service is the first major battleground. Companies like Sierra, Decagon, Maven, and Siena have collectively raised hundreds of millions of dollars alongside incumbents like Zendesk and Gorgias to deploy LLMs for retailers and brands to build more efficient CX workflows. These companies are impressive at the technical level and seem to be delivering actual, transformative results for brands like Simple Modern. (A case study detailing Siena’s work with Simple Modern is here).
But for now, they are all just plugging into the existing commerce schema. It’s the standard chat interface; only now it’s powered by an LLM instead of a human on the other side of an Intercom widget.
In a similar vein, DTC darlings AdoreMe and Curology are working with a company called Writer (rumored to be raising at a $2B valuation) to put generative AI at the center of their current workflows. Those flows being mainly around significantly increasing the scale of content creation that is SEO optimized, on-brand and intelligently assisted by custom LLMs. Remember, we started this series by talking about how the era of arbitraging paid traffic from platforms is ending — making investments in best-in-class organic content paramount.
When somebody has just put a hamburger in their shopping cart, the thing they are most likely to buy with it is french fries.
By and large, the next couple of years of AI adoption in commerce will look a lot like the Simple Modern and AdoreMe case studies above. These are still the DTC-era, performance marketing-powered brands finally achieving escape velocity. There are a plethora of manual, arcane workflows at the center of commerce that generative AI startups will make radically cheaper and faster. “Automatically find, copy, and optimize winning ads” from top brands might not be the most audacious tagline, but a product like Icon is bound to have a near-instant product market fit.
Many software businesses like this will be born in the next year or two and skyrocket to $10M+ in ARR with very lean teams. Much like the first wave of CRO tools, we’re talking about tens of billions of dollars in incremental market capitalization, even if the innovation stops here. But this is not the truly interesting futurism you read this publication for.
Ten years ago, technology couldn’t quite match the scale of imagination, so we solved the most lucrative, lowest-hanging fruit problems first. Now, it feels like retailers’ collective imaginations can’t quite keep up with the technology.
That is slowly beginning to change.
The Fifth Avenue Problem: AI's Next Frontier (2024 - ???)
The next generation of “AI in commerce” will be building the backend data layer that finally brings Fifth Avenue online.
Several years ago, analyst Benedict Evans quipped that while “the Internet lets you live in Wisconsin and buy anything you could buy in New York, it doesn’t let you shop the way you could shop in New York.”
Any current deployment of AI does not solve this problem. At the root of this problem is that the current architecture of eCommerce search simply cannot comprehend how people want to discover products.
The most basic example here is that if you search Sephora or Amazon for Meghan Markle’s favorite lipstick, you get a hodgepodge of seemingly random results. Yet, over the past 5+ years, every major women’s and lifestyle magazine has published the correct answer: Charlotte Tilbury Matte Revolution in Very Victoria. Google is, of course, the only “retailer” that can answer this query effectively, making it even all the more remarkable that they’ve fumbled the bag in commerce over the last decade.
Solving the semantic search problem is effectively the entire thesis of Daydream, a new startup from Stitch Fix and Nordstrom exec Julie Borstein that has raised $50M to craft an AI-powered search engine purpose-built for commerce. Daydream is an interesting concept, but I’d argue that advancements in GenAI will make the types of searches Daydream aims to empower even more decentralized.
In the not-too-distant future, LLMs will go from writing PDP copy and blog content to effectively creating training data and contextual information around each product SKU to help power entirely new ways of engaging with shoppers. This will be the third generation and killer application of AI in retail.
We’re rapidly seeing glimpses of what the future will look like as enterprise retailers–who for 20 years were hesitant to change so much as a pixel– are playing around with tests that change the nature of product search. Both Safeway and Ulta Beauty are working with a partner called ownit.ai to tackle product searches that would have previously been impossible, serving relevant results for queries like ‘quick dinner ideas for vegetarians” and “beach day essentials” and augmenting the context behind displaying certain products with AI-generated content. It’s a bit clunky and overwhelming for now but you can see how this will be compelling as the LLMs fine-tune the details.
For now, Meghan Markle’s favorite lipstick remains a mystery; if only because the commercial nature of search has deemed it to be a set of search terms that are so valuable that it becomes undiscoverable.
The competition to create the first truly game-changing deployment of AI is wide open. Amazon has every conceivable data advantage known to man and their first deployments of Rufus were almost comically bad. While they are rapidly improving, AI is the most obvious place where the innovator’s dilemma may finally come for Amazon. Amazon’s business is essentially geared towards showing as many product recommendations and ads as humanly possible, in an interface that cuts against the grain of where GenAI wants to pull retail.
Still, that isn’t stopping Amazon, a company that generally feels besieged by a “Day 2” mentality everywhere else across its business, from thinking first principles about product discovery. For its part, Walmart is developing proprietary, retail-specific LLMs to power new experiments across search, support, and, yes, personalization.
For AI to truly upend the form factor of commerce, two things will need to transpire in the coming years:
- Enterprise workflows for LLMs will need to be purpose-built for retail and CPG. Currently retail is mostly an afterthought vertical for OpenAI & Anthropic, who have first turned their attention to selling myopically more lucrative opportunities to major players in technology & finance. This creates an interesting opportunity for challenger companies like the aforementioned Writer who deploy their own models and have found some early traction with brands.
- Brands will need to rethink the entire concept of an eCommerce journey long before it feels sensible to do so. Our efforts to train a customer “how” to shop online have hemmed us in on every side from exercising any creativity or fluidity in that experience. If we don’t begin the change now, we may find the next generation leapfrogging the “old Web” format of pages into more open world and ‘gaming-aesthetic experiences.
True Classic, perhaps the last great winner of the “build a brand on Meta era” embodies this perfectly. Last month, President Ben Yahalom called for Generative AI leaders to deploy cutting-edge tools to “elevate every aspect of the company.” Already, they have several AI tests live, the most prominent with a company called Zowie, which prompts users to enter natural language questions and recommends products and sizing guides.
I found it to be an efficient and helpful (if uninspiring) experience.
Most of these early-stage AI deployments will likely fail against the familiar conversion pathways but damn it; it’s fun to see brands trying zany stuff again.
The moment for AI to truly transform commerce will be born out of necessity as much as technological innovation. When Facebook and Amazon ads printed predictable amounts of cash, there was little reason for “AI” to be anything more than CRO for ads.
The fun starts when the easy performance marketing funnels dry up… which is already happening.
No concept is more important to the future of commerce than Artificial Intelligence.
Consequently, no term has become more meaningless.
Issue 003 of Tabula Rasa is my humble attempt to define what the hell we’ve meant, historically, when we say “artificial intelligence” (hereafter AI). I’ll also unpack what we mean today— and what it will mean in the future when we utter the painful cliche that AI will “transform commerce.”
For retailers and brands, the umbrella term of AI has been used to define at least three distinct technological leaps forward, each more significant than the last.
The CRO Masquerade: AI's First Act (2015-2022)
For the last decade, when somebody was selling you AI as a retailer or brand, they were effectively selling conversion rate optimization (CRO) tech in some way, shape, or form. Long before GPT, every software that aimed to increase average revenue per visitor billed itself as “AI-powered”, a hodgepodge of conversion rate hacks that the vendor-industrial complex collectively defined as artificial intelligence.
To be clear, much of this technology was incredibly valuable to commerce– automating reasonably basic CRO and product recommendations testing unlocked billions of dollars of incremental value for retailers and spawned a host of nine figure software exits. Whether or not any of this tech was truly “AI” is a purely semantic discussion; what matters is that the value proposition is fundamentally different from what is now being sold under the AI banner.
The first generation of “AI in commerce” effectively substituted the term artificial intelligence for conversion rate optimization.
Throughout this era, the grand promise of AI was that it would enable true 1:1 personalization at scale, where every visitor to a website or app received a fully customized “experience.” In retrospect, this was the wrong north star for what artificial intelligence in commerce could achieve. The folly of how most brands approach 1:1 personalization is a complex topic that deserves its own column, but said simply, it comes down to this: You can track all the customer data in the world, but most of the time, when somebody has just put a hamburger in their shopping cart, the thing they are most likely to buy with it is french fries.
There are higher and greater uses for AI which are now beginning to manifest with the rise of Chat GPT and other large language models.
Which leads to point number two.
Fast, Cheap & In Control: The Efficiency Era (2022-2023)
The second (current) generation of “AI in commerce” is running the last decade’s playbook radically cheaper, faster, and better.
Before Chat GPT was released in 2022, the grand narratives of the technology industry were meaningless, manufactured culture wars about moving to Miami. If Open AI did nothing else, the company snapped an entire industry out of a dangerous state of ennui left by delusions of grandeur about NFTs and the metaverse. For that, we should be eternally grateful.
For commerce operators, the main upshot of generative AI to date has simply been making us significantly faster. At their Edge Summit earlier this month, Bloomreach unveiled a tool called Loomi, a perfect microcosm of AI in commerce today. Loomi essentially crafts a bespoke customer journey campaign in minutes. Workflows that would previously take savvy marketers or merchandisers days to weeks can now be done in minutes. By and large, however, the actual end shopper-facing experiences that Loomi will deploy are essentially the same set of tactics and widgets we’ve seen on our favorite websites for the past decade.
Zooming out, in this current AI wave, customer service is the first major battleground. Companies like Sierra, Decagon, Maven, and Siena have collectively raised hundreds of millions of dollars alongside incumbents like Zendesk and Gorgias to deploy LLMs for retailers and brands to build more efficient CX workflows. These companies are impressive at the technical level and seem to be delivering actual, transformative results for brands like Simple Modern. (A case study detailing Siena’s work with Simple Modern is here).
But for now, they are all just plugging into the existing commerce schema. It’s the standard chat interface; only now it’s powered by an LLM instead of a human on the other side of an Intercom widget.
In a similar vein, DTC darlings AdoreMe and Curology are working with a company called Writer (rumored to be raising at a $2B valuation) to put generative AI at the center of their current workflows. Those flows being mainly around significantly increasing the scale of content creation that is SEO optimized, on-brand and intelligently assisted by custom LLMs. Remember, we started this series by talking about how the era of arbitraging paid traffic from platforms is ending — making investments in best-in-class organic content paramount.
When somebody has just put a hamburger in their shopping cart, the thing they are most likely to buy with it is french fries.
By and large, the next couple of years of AI adoption in commerce will look a lot like the Simple Modern and AdoreMe case studies above. These are still the DTC-era, performance marketing-powered brands finally achieving escape velocity. There are a plethora of manual, arcane workflows at the center of commerce that generative AI startups will make radically cheaper and faster. “Automatically find, copy, and optimize winning ads” from top brands might not be the most audacious tagline, but a product like Icon is bound to have a near-instant product market fit.
Many software businesses like this will be born in the next year or two and skyrocket to $10M+ in ARR with very lean teams. Much like the first wave of CRO tools, we’re talking about tens of billions of dollars in incremental market capitalization, even if the innovation stops here. But this is not the truly interesting futurism you read this publication for.
Ten years ago, technology couldn’t quite match the scale of imagination, so we solved the most lucrative, lowest-hanging fruit problems first. Now, it feels like retailers’ collective imaginations can’t quite keep up with the technology.
That is slowly beginning to change.
The Fifth Avenue Problem: AI's Next Frontier (2024 - ???)
The next generation of “AI in commerce” will be building the backend data layer that finally brings Fifth Avenue online.
Several years ago, analyst Benedict Evans quipped that while “the Internet lets you live in Wisconsin and buy anything you could buy in New York, it doesn’t let you shop the way you could shop in New York.”
Any current deployment of AI does not solve this problem. At the root of this problem is that the current architecture of eCommerce search simply cannot comprehend how people want to discover products.
The most basic example here is that if you search Sephora or Amazon for Meghan Markle’s favorite lipstick, you get a hodgepodge of seemingly random results. Yet, over the past 5+ years, every major women’s and lifestyle magazine has published the correct answer: Charlotte Tilbury Matte Revolution in Very Victoria. Google is, of course, the only “retailer” that can answer this query effectively, making it even all the more remarkable that they’ve fumbled the bag in commerce over the last decade.
Solving the semantic search problem is effectively the entire thesis of Daydream, a new startup from Stitch Fix and Nordstrom exec Julie Borstein that has raised $50M to craft an AI-powered search engine purpose-built for commerce. Daydream is an interesting concept, but I’d argue that advancements in GenAI will make the types of searches Daydream aims to empower even more decentralized.
In the not-too-distant future, LLMs will go from writing PDP copy and blog content to effectively creating training data and contextual information around each product SKU to help power entirely new ways of engaging with shoppers. This will be the third generation and killer application of AI in retail.
We’re rapidly seeing glimpses of what the future will look like as enterprise retailers–who for 20 years were hesitant to change so much as a pixel– are playing around with tests that change the nature of product search. Both Safeway and Ulta Beauty are working with a partner called ownit.ai to tackle product searches that would have previously been impossible, serving relevant results for queries like ‘quick dinner ideas for vegetarians” and “beach day essentials” and augmenting the context behind displaying certain products with AI-generated content. It’s a bit clunky and overwhelming for now but you can see how this will be compelling as the LLMs fine-tune the details.
For now, Meghan Markle’s favorite lipstick remains a mystery; if only because the commercial nature of search has deemed it to be a set of search terms that are so valuable that it becomes undiscoverable.
The competition to create the first truly game-changing deployment of AI is wide open. Amazon has every conceivable data advantage known to man and their first deployments of Rufus were almost comically bad. While they are rapidly improving, AI is the most obvious place where the innovator’s dilemma may finally come for Amazon. Amazon’s business is essentially geared towards showing as many product recommendations and ads as humanly possible, in an interface that cuts against the grain of where GenAI wants to pull retail.
Still, that isn’t stopping Amazon, a company that generally feels besieged by a “Day 2” mentality everywhere else across its business, from thinking first principles about product discovery. For its part, Walmart is developing proprietary, retail-specific LLMs to power new experiments across search, support, and, yes, personalization.
For AI to truly upend the form factor of commerce, two things will need to transpire in the coming years:
- Enterprise workflows for LLMs will need to be purpose-built for retail and CPG. Currently retail is mostly an afterthought vertical for OpenAI & Anthropic, who have first turned their attention to selling myopically more lucrative opportunities to major players in technology & finance. This creates an interesting opportunity for challenger companies like the aforementioned Writer who deploy their own models and have found some early traction with brands.
- Brands will need to rethink the entire concept of an eCommerce journey long before it feels sensible to do so. Our efforts to train a customer “how” to shop online have hemmed us in on every side from exercising any creativity or fluidity in that experience. If we don’t begin the change now, we may find the next generation leapfrogging the “old Web” format of pages into more open world and ‘gaming-aesthetic experiences.
True Classic, perhaps the last great winner of the “build a brand on Meta era” embodies this perfectly. Last month, President Ben Yahalom called for Generative AI leaders to deploy cutting-edge tools to “elevate every aspect of the company.” Already, they have several AI tests live, the most prominent with a company called Zowie, which prompts users to enter natural language questions and recommends products and sizing guides.
I found it to be an efficient and helpful (if uninspiring) experience.
Most of these early-stage AI deployments will likely fail against the familiar conversion pathways but damn it; it’s fun to see brands trying zany stuff again.
The moment for AI to truly transform commerce will be born out of necessity as much as technological innovation. When Facebook and Amazon ads printed predictable amounts of cash, there was little reason for “AI” to be anything more than CRO for ads.
The fun starts when the easy performance marketing funnels dry up… which is already happening.
THIS ARTICLE IS FOR MEMBERS ONLY
Insights and futurism for executives in eCom and Retail
Those things we shouldn’t say out loud? We say them on the private feed. Bi-weekly “after dark” podcasts and a members-only newsletter, just for subscribers.
Our research reports combine visionary thinking with data-backed findings from our own advisory panel, made up of leaders at brands you know and trust.
Query and prompt our vast archive of research, podcasts, and newsletters with a ChatGPT-like interface. Get exclusive access to Alani™, the AI-powered engine for Future Commerce, powered by BundleIQ.
.png)
