M
Megha Jain
Guest
A few years ago, every product wanted a chatbot. Today, every product wants RAG.
Think of RAG as giving an AI access to your company's knowledge base before it answers a question. It first finds relevant information and then uses that information to generate a response.
Open any product roadmap discussion and you'll hear some variation of the same concern:
"Competitors have it."
"Customers are asking about AI."
"What if we lose users because we don't have it?"
The fear is understandable.
Nobody wants to be the company that missed the next platform shift.
But after speaking with product teams and watching the market evolve, I've started noticing a different pattern:
Many organizations are asking whether they need RAG before asking whether RAG creates enough business value to justify itself. Those are very different questions.
Let's assume RAG works exactly as intended. Customers can ask questions. The system retrieves relevant information. Responses are accurate and contextual. Now ask a harder question:
What happens next?
Does the customer buy more?
Do they convert faster?
Do they stay longer?
Do support costs decrease?
Or did they simply have a slightly better search experience? The answer matters because RAG isn't free. There are infrastructure costs.
Unlike traditional features, the bill arrives every single day. Which means the economics matter.
When we look at successful deployments, a pattern emerges. The biggest wins are happening where information retrieval itself is the job.
Perhaps the most cited example is Klarna.
Its AI assistant handled approximately two-thirds of customer service conversations within the first month of launch and was estimated to perform work equivalent to around 700 support agents. Klarna also reported a 25% reduction in repeat inquiries and resolution timesdropping from 11 minutes to under 2 minutes.
Source:
https://www.klarna.com/internationa...of-customer-service-chats-in-its-first-month/
What's interesting is that the value wasn't generated because customers loved chatting with AI. The value came from reducing operational costs. Every answered question replaced expensive human retrieval of information. The ROI was obvious.
The legal industry has become another major beneficiary. Lawyers spend enormous amounts of time searching through regulations, precedents, contracts and historical case information.
Reducing even a few minutes of research per task creates substantial value because the underlying labor is expensive.
This is why companies like Thomson Reuters and LexisNexis have invested heavily in retrieval-powered experiences. In these industries, the economics work because every successful retrieval directly saves expert time. Again, the value isn't conversation.
It's reduced research effort.
Now let's look at the space where RAG adoption is accelerating the fastest: consumer applications.
Here, the argument usually sounds compelling.
And all of that is true.
The real question is whether the outcome justifies the investment.
Customer A opens an online marketplace and searches:
"Running shoes under $100."
They browse products. Compare reviews. Read specifications. Make a purchase.
Customer B asks an AI assistant:
"What's a good running shoe under $100 for someone training for a half marathon?"
The AI retrieves product information, reviews and recommendations before presenting an answer. Customer B may reach a decision faster. But here's the question product leaders should be asking:
How much faster?
And how much additional revenue does that speed create?
If conversion increases from 2.0% to 2.1%, is that enough to justify the ongoing cost of serving millions of AI-powered queries?
Many organizations are implementing RAG without knowing the answer.
Most AI launches report:
Those are adoption metrics. The business metrics are different.
For consumer products, the questions should be:
Without movement on these metrics, RAG may simply be creating a more interesting experience rather than a more profitable one.
Consumer RAG appears most valuable when customers face high uncertainty before purchasing.
Think about:
These are categories where customers actively research before taking action. Reducing uncertainty can directly influence purchasing behavior. In other words, RAG isn't improving search. It's helping customers make decisions. That's where the value lies.
Now consider:
The customer already knows what they want. The challenge isn't discovery. It's execution. In these cases, a conversational layer may create more friction than value. The fastest path often wins. Not the smartest answer.
Whenever I hear a team discussing RAG, I ask three questions:
If the answers are unclear, the conversation is usually being driven by AI FOMO rather than business value. And that's the risk many organizations face today. Not that RAG won't work.
But that it will work just well enough to impress customers, while failing to create enough measurable value to justify its existence. The future won't belong to the companies that add RAG everywhere. It will belong to the companies that can prove exactly where it changes business outcomes.
Continue reading...
Think of RAG as giving an AI access to your company's knowledge base before it answers a question. It first finds relevant information and then uses that information to generate a response.
Open any product roadmap discussion and you'll hear some variation of the same concern:
"Competitors have it."
"Customers are asking about AI."
"What if we lose users because we don't have it?"
The fear is understandable.
Nobody wants to be the company that missed the next platform shift.
But after speaking with product teams and watching the market evolve, I've started noticing a different pattern:
Many organizations are asking whether they need RAG before asking whether RAG creates enough business value to justify itself. Those are very different questions.
The Most Expensive Question in AI Right Now
Let's assume RAG works exactly as intended. Customers can ask questions. The system retrieves relevant information. Responses are accurate and contextual. Now ask a harder question:
What happens next?
Does the customer buy more?
Do they convert faster?
Do they stay longer?
Do support costs decrease?
Or did they simply have a slightly better search experience? The answer matters because RAG isn't free. There are infrastructure costs.
- Embedding pipelines.
- Retrieval systems.
- LLM inference costs.
- Monitoring.
- Evaluation.
- Maintenance.
Unlike traditional features, the bill arrives every single day. Which means the economics matter.
The Places Where RAG Is Clearly Winning
When we look at successful deployments, a pattern emerges. The biggest wins are happening where information retrieval itself is the job.
Customer Support
Perhaps the most cited example is Klarna.
Its AI assistant handled approximately two-thirds of customer service conversations within the first month of launch and was estimated to perform work equivalent to around 700 support agents. Klarna also reported a 25% reduction in repeat inquiries and resolution timesdropping from 11 minutes to under 2 minutes.
Source:
https://www.klarna.com/internationa...of-customer-service-chats-in-its-first-month/
What's interesting is that the value wasn't generated because customers loved chatting with AI. The value came from reducing operational costs. Every answered question replaced expensive human retrieval of information. The ROI was obvious.
Legal and Compliance
The legal industry has become another major beneficiary. Lawyers spend enormous amounts of time searching through regulations, precedents, contracts and historical case information.
Reducing even a few minutes of research per task creates substantial value because the underlying labor is expensive.
This is why companies like Thomson Reuters and LexisNexis have invested heavily in retrieval-powered experiences. In these industries, the economics work because every successful retrieval directly saves expert time. Again, the value isn't conversation.
It's reduced research effort.
The Consumer Product Question Nobody Is Answering
Now let's look at the space where RAG adoption is accelerating the fastest: consumer applications.
- E-commerce
- Travel
- Marketplaces
- Financial comparison products
Here, the argument usually sounds compelling.
- Customers have questions
- Customers need guidance
- Customers want recommendations
And all of that is true.
The real question is whether the outcome justifies the investment.
Imagine Two E-Commerce Customers
Customer A opens an online marketplace and searches:
"Running shoes under $100."
They browse products. Compare reviews. Read specifications. Make a purchase.
Customer B asks an AI assistant:
"What's a good running shoe under $100 for someone training for a half marathon?"
The AI retrieves product information, reviews and recommendations before presenting an answer. Customer B may reach a decision faster. But here's the question product leaders should be asking:
How much faster?
- Five seconds?
- Thirty seconds?
- Two minutes?
And how much additional revenue does that speed create?
If conversion increases from 2.0% to 2.1%, is that enough to justify the ongoing cost of serving millions of AI-powered queries?
Many organizations are implementing RAG without knowing the answer.
The Metric That Matters
Most AI launches report:
- Number of conversations
- Number of prompts
- Number of users interacting with AI
Those are adoption metrics. The business metrics are different.
For consumer products, the questions should be:
- Did conversion increase?
- Did average order value increase?
- Did cart abandonment decrease?
- Did repeat purchases improve?
- Did customer acquisition costs decrease?
Without movement on these metrics, RAG may simply be creating a more interesting experience rather than a more profitable one.
Where Consumer RAG Probably Works
Consumer RAG appears most valuable when customers face high uncertainty before purchasing.
Think about:
- Electronics
- Financial products
- Insurance
- Travel planning
- Real estate
- Healthcare decisions
- Education platforms
These are categories where customers actively research before taking action. Reducing uncertainty can directly influence purchasing behavior. In other words, RAG isn't improving search. It's helping customers make decisions. That's where the value lies.
Where Consumer RAG Is Harder to Justify
Now consider:
- Food delivery
- Ride-sharing
- Utility payments
- Subscription management
- Everyday repeat purchases
The customer already knows what they want. The challenge isn't discovery. It's execution. In these cases, a conversational layer may create more friction than value. The fastest path often wins. Not the smartest answer.
The Framework I'm Using
Whenever I hear a team discussing RAG, I ask three questions:
- How much customer uncertainty exists before action?
- What business metric will improve if uncertainty decreases?
- Is the expected improvement larger than the ongoing cost of serving AI responses?
If the answers are unclear, the conversation is usually being driven by AI FOMO rather than business value. And that's the risk many organizations face today. Not that RAG won't work.
But that it will work just well enough to impress customers, while failing to create enough measurable value to justify its existence. The future won't belong to the companies that add RAG everywhere. It will belong to the companies that can prove exactly where it changes business outcomes.
Continue reading...