The pitch for intent data is simple and compelling: know which companies are actively researching your category before they ever fill out a form, and reach them first. In practice, most teams that have invested in intent data come back with a different story — expensive tooling, sales teams frustrated by low response rates, and a growing stack of "high intent" accounts that never converted into meetings.
The problem is not the concept. The concept is sound. A buyer who is actively researching your category is more receptive than a cold target who has never heard of you. The problem is that the category has a structural quality problem, and most platforms are selling the idea of intent data without actually delivering signal quality that converts.
After analyzing the failure patterns across dozens of B2B teams, the issues cluster into six distinct failure modes. Each has a specific cause — and a specific fix.
The Six Failure Modes
- 1. The Feed Problem — where the signal comes from matters
- 2. The Model Problem — keyword matching is not intent reasoning
- 3. The Feedback Loop Problem — static models that never improve
- 4. The Resolution Problem — account-level is not person-level
- 5. The Timeliness Problem — stale signals miss the buying window
- 6. The Context Problem — buyer stage is not optional information
Failure Mode 1: The Feed Problem
Most legacy intent data platforms operate on a co-op model: thousands of publishers agree to share behavioral signals in exchange for access to the aggregate pool. The theory is that a larger network produces better coverage. In practice, it produces worse signal quality.
A co-op network that pulls signals from thousands of unrelated publishers means that a significant portion of your "intent signals" are coming from sites your buyer would never use during an active evaluation. Someone reading a general technology news site, a free industry newsletter, or a tangentially related resource gets flagged as in-market for your category — not because they are evaluating vendors, but because they read content that contained your target keywords on a site that happens to be in the co-op.
The result is low-quality signals at scale. High volume with low relevance. The fix is curation: a purpose-built signal feed designed for relevance rather than coverage.
How Cursive addresses it:
Cursive's R1 Relevance layer curates a 28,000-domain signal feed specifically optimized for B2B buying signals. This produces a +28% relevance improvement over co-op data — not because the network is smaller, but because every domain in the feed is evaluated for its contribution to genuine buying intent rather than raw traffic volume.
Failure Mode 2: The Model Problem
Even with a better feed, most intent platforms use a shallow model: if someone visits a page containing your target keywords, they are flagged as "in-market." This is keyword-level intent matching, and it treats radically different buyers as equivalent signals.
Consider two people visiting pages about CRM software. The first is a junior marketing analyst reading "What is a CRM?" — a top-of-funnel educational page visited by someone likely doing general research with no near-term purchase intent. The second is a VP of Revenue reading "Salesforce vs HubSpot: feature comparison for enterprise sales teams" — a bottom-of-funnel comparison page visited by someone who is almost certainly in active vendor evaluation.
Keyword-only models flag both as identical "CRM intent" signals and deliver them to your sales team with the same score. The sales team reaches out to both, finds that most are not actually evaluating, and loses confidence in the signal. The problem is not that intent data doesn't work — it's that the model was too shallow to distinguish readiness.
| Signal Type | Keyword-Level Reading | URL-Level Reading |
|---|---|---|
| Blog post: "What is CRM" | High intent — CRM keyword | Low intent — educational, early stage |
| Pricing comparison page: "Salesforce vs HubSpot" | High intent — CRM keyword | Very high intent — active evaluation |
| G2 category review page | High intent — CRM keyword | Very high intent — vendor shortlisting |
| General tech news article mentioning CRM | High intent — CRM keyword | No intent — incidental mention |
How Cursive addresses it:
Cursive's R2 Reasoning layer applies URL-level buyer-stage classification at 6x the depth of keyword-only systems. Rather than matching keywords, the model classifies the specific page visited, its content type, and the company visiting it against a buyer-stage taxonomy. The result is a signal that tells you not just that a company is researching your category, but where they are in their buying journey.
Failure Mode 3: The Feedback Loop Problem
A signal model that never learns is a static map of a changing territory. Buyer behavior evolves. Categories mature. The signals that predicted purchase intent two years ago may not carry the same predictive value today. Without a feedback loop, your intent model is frozen at the state of knowledge that existed when it was last trained.
Legacy intent platforms deliver scores without any mechanism to connect those scores to outreach outcomes. If a company scored 92/100 for intent but never responded to outreach and ultimately did not purchase anything in the category, that information never feeds back into the model. If a company scored 58/100 but converted into a closed-won deal, that information is similarly lost.
Over time, the gap between the model's predictions and reality compounds. Teams add layers of manual filtering to compensate — only reaching out to accounts above a certain threshold, cross-referencing with their own CRM data, and running additional enrichment — all of which adds time and cost to a workflow that was supposed to save both.
How Cursive addresses it:
Cursive's signal model includes a feedback loop that connects outreach outcomes back to signal weights. Accounts that produced meetings or pipeline improve the weight of similar signals. Accounts that did not respond or were misclassified reduce the weight of those signal patterns. The model improves continuously rather than degrading over time.
Failure Mode 4: The Resolution Problem
There is a meaningful difference between knowing that a company is researching your category and knowing which person at that company is doing the research. Most intent platforms resolve signals to the account level: they can tell you that Acme Corp visited pages about your category, but not who at Acme Corp is driving that research.
This creates a targeting problem. Sales teams receive account-level intent and are expected to identify the right contact themselves — usually by guessing based on job title or org chart position. This manual step introduces lag (time spent researching) and error (wrong stakeholder outreach). It also means that the intent signal, which had genuine freshness at the moment of collection, has aged by the time anyone acts on it.
Person-level resolution changes the economics of the workflow entirely. Instead of an account signal that requires manual contact research, you have a person-level signal that includes name, email, job title, and company — ready for immediate outreach.
How Cursive addresses it:
Cursive's Super Pixel provides person-level identity resolution on your own website traffic, identifying up to 70% of anonymous visitors by name, email, job title, and company. For off-site intent signals, Cursive layers its identity graph to resolve account-level signals to person-level contacts where available, dramatically reducing the gap between signal and actionable outreach.
Failure Mode 5: The Timeliness Problem
Active buying cycles have a window. A buyer who is evaluating vendors in a category moves through stages — awareness, active research, vendor shortlisting, evaluation calls, decision — and the window for each stage is finite. In fast-moving categories, the shortlisting stage can close within days of peak research activity.
Most intent data platforms deliver signals in weekly or bi-weekly batches. This means the signals your sales team receives on Monday may reflect research that happened 7-14 days ago. By that point, the buyer may have already had initial calls with vendors they found first, built a mental shortlist, or moved into a later buying stage where they are less open to new introductions.
Timing is a compounding advantage. The first vendor to reach a buyer at peak research intent has a structural edge in the evaluation. Being second or third — even with a better product — requires overcoming the anchoring effect of whoever got there first.
How Cursive addresses it:
Cursive delivers signals in real time, not weekly batches. Website visitor identification via the Super Pixel fires within minutes of a visit, and off-site intent signals are processed and delivered on the same day rather than batched for weekly delivery. This gives your sales team the ability to reach buyers while the research is still fresh.
Failure Mode 6: The Context Problem
Intent is not uniform across industries. A company in B2B SaaS evaluating a sales intelligence platform has a different research pattern than a healthcare organization evaluating the same category. The pages they visit, the content they consume, and the signals they generate differ significantly. A model trained on aggregate data treats all of these as equivalent, which produces false positives within specific verticals and misses genuine high-intent signals in others.
This is the context problem: the absence of vertical awareness in the intent model. Without understanding that the research behavior of a 500-person logistics company looks different from a 50-person fintech startup, even a high-quality signal feed produces signals that feel off to the sales teams working specific verticals.
How Cursive addresses it:
Cursive's R3 Recognition layer trains separate industry cohort models per vertical — B2B SaaS, healthcare, financial services, logistics, and others. The model learns what "high intent" looks like within your specific buyer segment, not in the aggregate. This means the signals you receive are vertically calibrated from day one, without requiring manual tuning.
The Compound Effect of All Six Failures
Each failure mode degrades signal quality independently. Together, they compound. A broad co-op feed delivering high-volume low-relevance signals, processed by a keyword-only model with no vertical awareness, batched weekly, resolved only to account level, and never connected to a feedback loop — this is the typical enterprise intent data stack. It produces exactly the pattern most teams describe: impressive scores, poor conversion.
| Failure Mode | Legacy Platform | Cursive Fix |
|---|---|---|
| Feed quality | Co-op: thousands of unrelated publishers | 28,000-domain curated feed, +28% relevance (R1) |
| Model depth | Keyword matching only | URL-level buyer-stage classification, 6x depth (R2) |
| Feedback loop | None — static model | Outcomes feed back into signal weights continuously |
| Resolution | Account-level (company only) | Person-level via Super Pixel (name, email, title) |
| Signal latency | Weekly or bi-weekly batches | Real-time delivery |
| Vertical awareness | Aggregate model, no cohort segmentation | Industry cohorts per vertical (R3) |
What High-Performing Intent Data Looks Like in Practice
Teams that get strong ROI from intent data typically have three things in common. First, they use intent signals to prioritize, not to source. They are not trying to cold-source net-new accounts from intent lists alone — they are using intent to prioritize which accounts in their existing TAM to contact first, and in what order. This produces higher conversion because the list is already qualified against their ICP before layering on intent.
Second, they combine off-site intent with on-site visitor identification. Off-site intent tells them which companies are researching in their category broadly. On-site visitor identification tells them which of those companies are specifically researching their product. The overlap between these two signals — a company showing off-site intent that also visited your pricing or comparison pages — represents a high-confidence, person-level lead. These convert at substantially higher rates than either signal alone.
Third, they close the loop. They track which intent signals produced meetings, which produced closed-won deals, and they share that data with their intent provider or use it to tune their own internal scoring. The teams that treat intent data as a set-and-forget service get set-and-forget results. The teams that treat it as a signal that needs to be monitored, tuned, and improved over time compound their advantage quarter over quarter.
The Diagnostic Questions
If you are evaluating whether your current intent data stack is working — or auditing a new provider — these are the questions that surface the failure modes quickly:
- Feed: How many domains are in the signal network, and what is the selection criteria? Is it co-op (volume-first) or curated (relevance-first)?
- Model: Does the platform score at the keyword level or the URL level? Can it distinguish between a buyer reading a definition post and a buyer reading a comparison page?
- Feedback loop: How does your outreach outcome data get back to the model? Does the model update based on your results, or is it static?
- Resolution: Does the platform deliver account-level or person-level signals? If account-level, what is the expected workflow for contact identification?
- Latency: How frequently are signals delivered? Real-time, daily, weekly? What is the typical gap between signal generation and delivery?
- Vertical awareness: Is the model trained on aggregate data or vertical-specific cohorts? How does it account for category-specific research patterns?
The answers to these questions will tell you whether the platform can actually deliver on the intent data promise — or whether you are paying for a high-volume signal feed that your sales team will eventually stop trusting.
Next Steps
The intent data category is not broken as a concept. The infrastructure delivering on that concept is broken for most buyers. The path forward is a stack built on a curated feed, URL-level reasoning, vertical cohort models, real-time delivery, person-level resolution, and a feedback loop that connects outcomes back to signal quality.
If you want to see what that looks like in practice on your own traffic, run a free estimate of how many named leads your site is losing or start with Cursive's Super Pixel to begin building a person-level intent layer on your own first-party data.
For a deeper technical explanation of the R4 framework, read the R4 Signal Model breakdown.