Your AI Agent's Biggest Problem Isn't the Model - It's the Data

Uri Knorovich
Nimble

Agentic AI has shown dazzling promise, but it also raises serious concerns. And some of the most pressing ones aren't about reasoning at all.

A new analysis of 250 real-world queries across common retail tasks, such as product pricing, availability, ratings, shipping and specifications, reveals systemic inefficiency at the heart of web-based AI agents. On average, 97.9% of the data retrieved by agents from live web pages is irrelevant to the query being answered. Specifically, the average page ingested ran nearly 9,000 characters, while the average answer was just 32 characters, resulting in a noise-to-signal ratio of 278:1. Price queries were the most extreme outlier, with noise rates approaching 99.5%.

That's not a rounding error. That's a structural problem.

There Are Two Problems That Compound Each Other

The consequences of this data inefficiency show up in two ways, and they reinforce each other.

Accuracy takes a hit. Large language models are sensitive to the quality of context. When the context window is long, the signal is sparse, and multiple plausible-but-wrong answers compete for attention, models are more likely to pick the wrong one.

Retail pages are a near-perfect illustration: a single product page often contains the current price, a sale price, bundle pricing, prices for related configurations, and historical price references buried in user reviews. The model isn't failing because it's incapable; it's failing because it's working with a mess.

Costs scale with noise. At roughly four characters per token, a 9,000-character page translates to approximately 2,200 tokens per query. Multiply that across thousands of daily queries and multi-step retrieval pipelines where an agent might hit several pages before arriving at an answer, and the token math becomes alarming.

The majority of token spending in a typical agentic system may be attributable to data that never contributes to a correct answer. That's wasteful, of course, and for organizations running AI agents at scale, it's a meaningful cost driver.

It's Not Just a Retail Problem

While this specific dataset centers on retail, it reveals an underlying dynamic that affects virtually every industry that relies on live web data: the web was built for humans, not machines.

Human-readable web pages are packed with navigation menus, ads, related products, legal disclaimers, user reviews and various layout quirks. An AI agent retrieving a page to answer a narrow question doesn't benefit from any of that, but it still has to process all of it.

The same dynamic can just as easily show up in finance (extracting one metric from an earnings report), travel (finding availability from a booking page), real estate (pulling property details from a listing site), or healthcare (identifying specific data from a provider directory).

In every case, the agent is doing more work than necessary to retrieve a very small but specific signal.

The Missing Component? Retrieval Architecture.

There's a common assumption in the AI community that improving agent performance means better models, longer context windows, or more sophisticated reasoning chains. But the data here suggests otherwise.

If 97.9% of the input is irrelevant before the model ever begins reasoning, no model improvement will fully compensate for that. With this the case, the bottleneck isn't reasoning. It's retrieval.

This reframes where engineering attention should be focused for teams building production-grade agentic systems. Signal-to-noise ratio deserves to be treated as a first-class infrastructure metric, alongside latency and accuracy. That shift comes with three practical implications:

1. Prefer structured data over raw HTML. If the answer exists as a discrete field (a price, a rating or a ship date), it should be retrieved as a field, not extracted from a wall of prose. Structured data pipelines deliver more reliable inputs and dramatically lower token overhead.

2. Pull precisely scoped data, not full documents. Agents don't need entire pages. They need the specific slice of information relevant to the query. Retrieval systems should be designed to scope requests narrowly rather than broadly.

3. Evaluate the data layer, not just the model layer. When assessing AI agent infrastructure, the question shouldn't only be which model performs best on the given task. You should also be asking how much of what the model receives is actually useful.

Purpose-built web intelligence APIs designed to return structured fields rather than raw page dumps are increasingly where this gap gets addressed.

Better Data Beats a Better Model Every Time

The good news? This is a solvable problem. It doesn't require a better model or a bigger context window. It requires better data hygiene earlier in the pipeline.

For developers building or scaling AI agents, the most impactful architectural decision may not be which LLM to use; it may be how and what data is fed into it. Treating retrieval as an engineering discipline, with the same rigor applied to accuracy and latency, is what separates agents that work in controlled environments from those that hold up in the real world.

Uri Knorovich is CEO and Co-Founder of Nimble

Related Links

Hot Topics

The Latest

When the Database Is the Problem, Is APM Enough?

Production incidents rarely announce themselves as database problems. They appear as slow transactions, timeouts, rising response times, or an application struggling under a workload it previously handled. APM provides an essential starting point. It can identify a slow transaction path, highlight an affected service, and show that a database dependency is consuming more time than expected. But identifying the database as part of the problem is not the same as explaining what is happening inside it ...

From Cleanup to Prevention: The Future of Cloud-Native Cost Optimization

Cloud teams are under constant pressure to reduce spend without slowing development or increasing operational risk. They are deploying autoscalers, rightsizing workloads, enforcing resource requests, reviewing utilization dashboards, and building FinOps processes around cloud-native environments. Yet the results often disappoint ...

The Real Enterprise AI Risk Isn't Hallucinations

Ask most IT leaders about their biggest concern with AI and you'll hear the same answer: hallucinations ... Today, however, the conversation has shifted ... As organizations move beyond chatbots and experiments, they are increasingly deploying AI agents that perform multi-step tasks. These systems retrieve documents, query databases, call APIs, generate reports, write code, and make recommendations. The issue is not whether the model can reason. The issue is whether the organization can see, verify, and govern the decisions being made along the way ...

Beyond Buy vs. Build: The New Playbook for Data Pipelines

While organizations want to take control of their telemetry, building telemetry pipelines from scratch can be a very daunting, complicated task, even when leveraging open-source standards like OpenTelemetry. It requires specialized knowledge across distributed systems, data engineering, and security. This fragmented approach across systems causes higher operational costs; it puts a strain on resources and reduces efficiency as teams have to work with different interfaces and processes ...

The Corporate Network Is Fading - Here's What Replaces It

For decades, enterprise networks were designed around a simple assumption: work happened inside the office. Applications lived in centralized data centers, employees connected through internal infrastructure, and security focused on protecting the perimeter that surrounded everything ... But the way organizations operate today bears little resemblance to that environment. Cloud platforms host critical applications, employees connect from homes and airports as often as they do from offices, and partners collaborate through shared systems that exist far beyond corporate walls. In short, the corporate network no longer resembles the environment it was designed to protect ...

IT Leaders Must Establish a Network Data Architecture Practice

As an analyst who researches how IT organizations design, build, and operate their networks, I find that network data is a constant source of pain. Network teams struggle with data quality, fragmentation, authority, access, and trust. And these issues undermine everything they try to do. Here are the numbers: Only 45% of network teams are completely confident in the accuracy of their network source of truth, which documents the intent of their network ...

Data Center Outages Are Decreasing but More Costly

The 2026 Global Data Center Survey from Uptime Institute reveals an industry navigating workforce constraints, escalating outage expenses, even as rising costs remain the top concern for management teams ...

Observability Has a Physical Layer Now: What AI Infrastructure Teaches IT Operations

The next observability gap may not be in the code. It may be under the rack. That sounds strange until you think about how AI incidents actually feel in the middle of an investigation ... The application dashboard may be accurate. It may also be stopping at the wrong boundary. AI systems depend on software, but they also depend on a dense physical stack: racks, power paths, thermal margin, maintenance activity and, in many environments, liquid cooling. Those physical dependencies can change slowly before they look like a software incident ...

TLS Certificate Expiration Is Becoming an Observability Problem

Certificate expiration is the rare outage you can see coming. Every TLS certificate carries the date it stops working, so the moment it will begin breaking connections is knowable in advance. That's what makes an expired certificate such a frustrating way to lose a service. What's changing now is how often that date comes around ...

Enterprise Cloud Observability: Choosing the Right Starting Point

Enterprises operate different combinations of workloads across cloud, hybrid and multicloud environments. For business-critical workloads, teams need to consider monitoring and observability early so they can detect health issues, investigate failures, and understand operational impact. Organizations place workloads on cloud platforms based on a combination of technical requirements, economics, existing dependencies, organizational standards, and business priorities. Their monitoring priorities therefore depend on what they operate and where those systems run. Those priorities will not look the same for every organization ...

Your AI Agent's Biggest Problem Isn't the Model - It's the Data

Uri Knorovich
Nimble

Agentic AI has shown dazzling promise, but it also raises serious concerns. And some of the most pressing ones aren't about reasoning at all.

A new analysis of 250 real-world queries across common retail tasks, such as product pricing, availability, ratings, shipping and specifications, reveals systemic inefficiency at the heart of web-based AI agents. On average, 97.9% of the data retrieved by agents from live web pages is irrelevant to the query being answered. Specifically, the average page ingested ran nearly 9,000 characters, while the average answer was just 32 characters, resulting in a noise-to-signal ratio of 278:1. Price queries were the most extreme outlier, with noise rates approaching 99.5%.

That's not a rounding error. That's a structural problem.

There Are Two Problems That Compound Each Other

The consequences of this data inefficiency show up in two ways, and they reinforce each other.

Accuracy takes a hit. Large language models are sensitive to the quality of context. When the context window is long, the signal is sparse, and multiple plausible-but-wrong answers compete for attention, models are more likely to pick the wrong one.

Retail pages are a near-perfect illustration: a single product page often contains the current price, a sale price, bundle pricing, prices for related configurations, and historical price references buried in user reviews. The model isn't failing because it's incapable; it's failing because it's working with a mess.

Costs scale with noise. At roughly four characters per token, a 9,000-character page translates to approximately 2,200 tokens per query. Multiply that across thousands of daily queries and multi-step retrieval pipelines where an agent might hit several pages before arriving at an answer, and the token math becomes alarming.

The majority of token spending in a typical agentic system may be attributable to data that never contributes to a correct answer. That's wasteful, of course, and for organizations running AI agents at scale, it's a meaningful cost driver.

It's Not Just a Retail Problem

While this specific dataset centers on retail, it reveals an underlying dynamic that affects virtually every industry that relies on live web data: the web was built for humans, not machines.

Human-readable web pages are packed with navigation menus, ads, related products, legal disclaimers, user reviews and various layout quirks. An AI agent retrieving a page to answer a narrow question doesn't benefit from any of that, but it still has to process all of it.

The same dynamic can just as easily show up in finance (extracting one metric from an earnings report), travel (finding availability from a booking page), real estate (pulling property details from a listing site), or healthcare (identifying specific data from a provider directory).

In every case, the agent is doing more work than necessary to retrieve a very small but specific signal.

The Missing Component? Retrieval Architecture.

There's a common assumption in the AI community that improving agent performance means better models, longer context windows, or more sophisticated reasoning chains. But the data here suggests otherwise.

If 97.9% of the input is irrelevant before the model ever begins reasoning, no model improvement will fully compensate for that. With this the case, the bottleneck isn't reasoning. It's retrieval.

This reframes where engineering attention should be focused for teams building production-grade agentic systems. Signal-to-noise ratio deserves to be treated as a first-class infrastructure metric, alongside latency and accuracy. That shift comes with three practical implications:

1. Prefer structured data over raw HTML. If the answer exists as a discrete field (a price, a rating or a ship date), it should be retrieved as a field, not extracted from a wall of prose. Structured data pipelines deliver more reliable inputs and dramatically lower token overhead.

2. Pull precisely scoped data, not full documents. Agents don't need entire pages. They need the specific slice of information relevant to the query. Retrieval systems should be designed to scope requests narrowly rather than broadly.

3. Evaluate the data layer, not just the model layer. When assessing AI agent infrastructure, the question shouldn't only be which model performs best on the given task. You should also be asking how much of what the model receives is actually useful.

Purpose-built web intelligence APIs designed to return structured fields rather than raw page dumps are increasingly where this gap gets addressed.

Better Data Beats a Better Model Every Time

The good news? This is a solvable problem. It doesn't require a better model or a bigger context window. It requires better data hygiene earlier in the pipeline.

For developers building or scaling AI agents, the most impactful architectural decision may not be which LLM to use; it may be how and what data is fed into it. Treating retrieval as an engineering discipline, with the same rigor applied to accuracy and latency, is what separates agents that work in controlled environments from those that hold up in the real world.

Uri Knorovich is CEO and Co-Founder of Nimble

Related Links

Hot Topics

The Latest

When the Database Is the Problem, Is APM Enough?

Production incidents rarely announce themselves as database problems. They appear as slow transactions, timeouts, rising response times, or an application struggling under a workload it previously handled. APM provides an essential starting point. It can identify a slow transaction path, highlight an affected service, and show that a database dependency is consuming more time than expected. But identifying the database as part of the problem is not the same as explaining what is happening inside it ...

From Cleanup to Prevention: The Future of Cloud-Native Cost Optimization

Cloud teams are under constant pressure to reduce spend without slowing development or increasing operational risk. They are deploying autoscalers, rightsizing workloads, enforcing resource requests, reviewing utilization dashboards, and building FinOps processes around cloud-native environments. Yet the results often disappoint ...

The Real Enterprise AI Risk Isn't Hallucinations

Ask most IT leaders about their biggest concern with AI and you'll hear the same answer: hallucinations ... Today, however, the conversation has shifted ... As organizations move beyond chatbots and experiments, they are increasingly deploying AI agents that perform multi-step tasks. These systems retrieve documents, query databases, call APIs, generate reports, write code, and make recommendations. The issue is not whether the model can reason. The issue is whether the organization can see, verify, and govern the decisions being made along the way ...

Beyond Buy vs. Build: The New Playbook for Data Pipelines

While organizations want to take control of their telemetry, building telemetry pipelines from scratch can be a very daunting, complicated task, even when leveraging open-source standards like OpenTelemetry. It requires specialized knowledge across distributed systems, data engineering, and security. This fragmented approach across systems causes higher operational costs; it puts a strain on resources and reduces efficiency as teams have to work with different interfaces and processes ...

The Corporate Network Is Fading - Here's What Replaces It

For decades, enterprise networks were designed around a simple assumption: work happened inside the office. Applications lived in centralized data centers, employees connected through internal infrastructure, and security focused on protecting the perimeter that surrounded everything ... But the way organizations operate today bears little resemblance to that environment. Cloud platforms host critical applications, employees connect from homes and airports as often as they do from offices, and partners collaborate through shared systems that exist far beyond corporate walls. In short, the corporate network no longer resembles the environment it was designed to protect ...

IT Leaders Must Establish a Network Data Architecture Practice

As an analyst who researches how IT organizations design, build, and operate their networks, I find that network data is a constant source of pain. Network teams struggle with data quality, fragmentation, authority, access, and trust. And these issues undermine everything they try to do. Here are the numbers: Only 45% of network teams are completely confident in the accuracy of their network source of truth, which documents the intent of their network ...

Data Center Outages Are Decreasing but More Costly

The 2026 Global Data Center Survey from Uptime Institute reveals an industry navigating workforce constraints, escalating outage expenses, even as rising costs remain the top concern for management teams ...

Observability Has a Physical Layer Now: What AI Infrastructure Teaches IT Operations

The next observability gap may not be in the code. It may be under the rack. That sounds strange until you think about how AI incidents actually feel in the middle of an investigation ... The application dashboard may be accurate. It may also be stopping at the wrong boundary. AI systems depend on software, but they also depend on a dense physical stack: racks, power paths, thermal margin, maintenance activity and, in many environments, liquid cooling. Those physical dependencies can change slowly before they look like a software incident ...

TLS Certificate Expiration Is Becoming an Observability Problem

Certificate expiration is the rare outage you can see coming. Every TLS certificate carries the date it stops working, so the moment it will begin breaking connections is knowable in advance. That's what makes an expired certificate such a frustrating way to lose a service. What's changing now is how often that date comes around ...

Enterprise Cloud Observability: Choosing the Right Starting Point

Enterprises operate different combinations of workloads across cloud, hybrid and multicloud environments. For business-critical workloads, teams need to consider monitoring and observability early so they can detect health issues, investigate failures, and understand operational impact. Organizations place workloads on cloud platforms based on a combination of technical requirements, economics, existing dependencies, organizational standards, and business priorities. Their monitoring priorities therefore depend on what they operate and where those systems run. Those priorities will not look the same for every organization ...