The intent vendor surfaced forty surging accounts a week, and the list went to the SDR team every Monday. By the end of the quarter, the numbers told a different story. A few surging accounts turned into meetings. Many would never buy: too small, the wrong industry, already a customer, or a competitor reading about you. The reps did what reps always do with a list that wastes their time. They stopped opening it.
The instinct is to blame the vendor. More often the problem is the design choice to treat one behavioral signal as a buying decision, without asking whether the account can buy, whether anyone else there agrees, or whether the surge happened last week or last month.
Each of those numbers explains part of why intent alone disappoints. The LinkedIn B2B Institute, drawing on work by Professor John Dawes of the Ehrenberg-Bass Institute, argued in 2021 that only about 5% of B2B buyers are in-market at any moment and the other 95% will not buy for months or years. When the true rate is that low, even a modest false-alarm rate produces more wrong answers than right ones. Forrester's The State of Business Buying, 2026, released in January 2026, found that 13 internal stakeholders and nine external participants now influence the typical B2B purchase decision. And 6sense's 2025 B2B Buyer Experience study, a survey of roughly 4,000 buyers with a supplementary sample of 766, found that buyers first reach out to sellers about 61% of the way through their journey, and that buying groups fill four of the five places on their shortlist on day one. By the time a surge becomes visible, the shortlist is often forming.
There is a fourth problem underneath all three. Adverity's State of Play research on data quality, published in September 2025 from a survey of CMOs in the US, UK, Germany, Austria and Switzerland, found that 45% of marketing data is incomplete, inaccurate or out of date. Intent scores inherit those errors, and a model that leans on a single input has nothing to catch them. This is not a vendor problem. It is a scoring architecture problem.
Diagnosis: why intent-only targeting produces noise
At $3M to $30M ARR, the intent feed is usually wired straight to a rep queue, and four structural weaknesses show up within a quarter.
Intent measures topic research, not an account's decision to buy
Third-party intent is usually an inference: content consumption across a network of publishers, resolved to a company through IP addresses, cookies or registrations, then compared to that company's normal baseline. A surge means someone at the organization read more about a topic than usual. It cannot tell you whether that person is a buyer, a job seeker, a customer or a competitor. Forrester's own guidance on the most common intent data mistakes names using intent signals in a vacuum among them, and recommends pairing the data with the other insights you already hold. Without that pairing, every surge looks the same.
The base rate makes false positives the default
If one account in twenty is actually in-market, then even a signal that is right most of the time will flag more out-of-market accounts than in-market ones, simply because there are nineteen times as many of them. A vendor can quote a reasonable accuracy figure and your reps can still find that most accounts they call are not buying. The way out is to add evidence that narrows the pool before the signal is read.
One reader is treated as a buying group
With 13 internal stakeholders in a typical decision, according to Forrester's 2026 research, a real buying motion leaves traces across several people and several functions. Most intent-only setups score the account on volume, so one very curious person can push an account to the top of the list. Breadth, meaning how many distinct people and roles are showing activity, is usually a better sign of a live evaluation than depth from one person, and most intent-only scores do not measure it at all.
The signal is read without a clock
Forrester describes intent as among the most time-sensitive data types available, with a limited window in which acting on it makes sense. Yet many teams store a weekly surge score as a static field and let it sit. An account that surged six weeks ago keeps its high score, and the rep calls after the shortlist is set. Signal freshness decay covers how quickly each kind of buying signal loses its value and how to set routing windows around it.
The framework: the Blended Signal Score
The Blended Signal Score separates the questions and gives each its own evidence. It is a targeting-focused application of the four-signal account scoring model, and it has three parts that apply in a fixed order.
The first is a fit gate built from firmographic and technographic data: the company size, industry, geography, business model and technologies your closed-won customers share. It is a gate, not a component to be added up. A poor-fit account with a large surge should not outrank a strong-fit account with a moderate one, because the first cannot become a good customer however interested it is. Accounts either pass, pass at a reduced tier, or are excluded. Customers, open opportunities, partners and competitors are excluded here too. The gate is only as good as the list it reads, so keep the target account list refreshed on triggers rather than once a year.
The second is a behavioral blend across independent signal families: third-party intent, first-party engagement (visits to your pricing and product pages, content downloads, event attendance, replies), technographic change (a tool added or dropped that your product works with or replaces) and breadth (how many distinct people and roles are active). The key rule is corroboration: an account is not routed to a rep on one family alone. It needs at least two independent families pointing the same way.
The third is recency decay. Every behavioral signal loses weight with age according to a half-life, so a surge from last week counts in full and one from six weeks ago counts for little. The score then reflects what is happening now, not what happened once.
Written as a formula, with weights offered as a suggested starting point rather than a benchmark:
Blended Signal Score = Fit tier × (0.30 × first-party engagement + 0.25 × third-party intent + 0.25 × breadth + 0.20 × technographic change), where each behavioral term is multiplied by 0.5^(age in days ÷ half-life). Fit tier is 1.0 for a full pass, 0.5 for a partial pass and 0 for an exclusion. A first half-life might be 14 days for third-party intent and 30 days for first-party engagement and technographic change. Routing to a rep requires both a score above your threshold and at least two families contributing. Your own backtest, described below, should replace every one of these numbers.
First-party engagement carries the highest starting weight because it is behavior on your own properties, tied to known people. Third-party intent stays valuable as an early indicator for accounts that have not found you yet.
Illustrative example. These numbers are made up and rounded to show the mechanics; they are not benchmarks, vendor figures or client results. Take a universe of 1,000 accounts and assume, in line with the B2B Institute's estimate, that 50 are in-market this quarter. Suppose an intent feed flags 70% of the in-market accounts (35) and 16% of the other 950 (152). Intent alone routes 187 accounts to reps, of which 35 are real: a precision of about 19%, which means roughly four out of five calls go to accounts that are not buying. Now apply the blended design. Suppose the fit gate passes 40 of the 50 in-market accounts and 300 of the 950 others. Intent flags 28 of those 40 and 45 of the 300. The corroboration rule keeps the 24 in-market accounts that also show first-party engagement or a technographic change, and 9 of the 45 others. The blended score routes 33 accounts, 24 of them real: a precision of about 73%. The rep queue shrinks from 187 to 33, and the share of wasted calls falls from about 81% to about 27%.
The trade-off is real and should be stated to leadership plainly. In this example the blended model surfaces 24 in-market accounts rather than 35. The eleven it misses stay in marketing programs and move up the moment a second signal appears.
Implementation: six steps to a blended score you can trust
This runs on the CRM and data most teams already pay for. Each step ends in a check.
Backtest the signals you already have
Pull the last four to six quarters of opportunities, won and lost, and look at which signals were present on each account in the 30 to 90 days before the opportunity was created. The diagnose-before-you-build playbook covers how to run this kind of read-only analysis. Check: you can say, for each signal family, how often it appeared before real opportunities and how often it appeared on accounts that went nowhere.
Write the fit gate from closed-won, not from the pitch deck
Define the firmographic and technographic criteria your best customers share, plus the exclusions: customers, open opportunities, partners and competitors. Store them as fields and logic in the CRM so the gate runs on every account automatically. Check: a new analyst could apply the gate to any company and reach the same answer.
Map every signal to one account record
Each signal family needs a source, a timestamp and a reliable match to a single account. If intent resolves by domain, engagement by email and technographic change by company name, they may not land on the same record, and corroboration fails. This is an identity resolution problem before it is a scoring problem. Check: for a sample of flagged accounts, all four families resolve to the same account ID.
Set weights, half-lives and the threshold
Start with the suggested defaults above, then adjust them to fit your backtest: raise the weight of families that preceded real opportunities, lower the ones that did not, and set the routing threshold where the queue matches the capacity of your team. Check: at the chosen threshold, the weekly queue is a number your reps can fully work.
Test on past cases before going live
Replay the model on accounts where you already know the outcome. We hold every system to the same bar: tested on around 20 of the client's own past cases, and 85 percent correct or it does not ship. Check: the model makes the right call on past accounts, and every miss has a written reason.
Route with reason codes and review precision monthly
Every routed account carries its reasons, such as "fit tier 1, pricing page visits by three people, intent surge on two category topics, nine days old". Monthly, review how many routed accounts reps accepted and converted. Check: precision is reported monthly, and weights change only on evidence.
Workflow: what each score tier sets in motion
A suggested design. Adapt owners and time limits, but keep rep time for accounts with corroborated, current evidence.
Rule: passes the fit gate, score above threshold, two or more signal families active within their half-lives.
Action: routed to the owning rep within one business day with reason codes and suggested contacts from the active people.
Owner: the account executive or SDR assigned to the account; RevOps monitors time to first touch.
Rule: passes the fit gate with one active signal family, or two families below threshold.
Action: enrolled in targeted marketing programs and light SDR research. Promoted to Tier 1 automatically when a second family appears.
Owner: marketing ops, with SDR support.
Rule: strong behavioral signals but a partial fit, or a strong fit with no current signal.
Action: no rep time. Partial-fit accounts with high activity are reviewed monthly, since a pattern of them can mean the gate is too narrow.
Owner: RevOps reviews the monthly pattern.
Rule: current customers, open opportunities, partners, competitors and clear poor fits.
Action: intent from current customers is not discarded. It goes to the account manager or customer success as a possible expansion or risk signal, never to a new-business rep.
Owner: customer success and account management.
The board narrative
Three statements usually carry the conversation.
We used to send reps every account that showed an intent surge. We now score accounts on fit, corroborating signals and recency together, so rep time goes only to accounts that can buy and where more than one kind of evidence says an evaluation is happening now.
Each month we report the share of routed accounts that reps accepted, the share that became opportunities, and time from signal to first touch. The weights were set by backtesting against our own closed deals.
The rep queue is smaller and more of it converts. We keep paying for intent data, but it now contributes to a decision instead of making it alone, and we can show which signal sources earn their cost at renewal time.
Cross-domain: how the blended score connects to the other systems
A blended score is only as useful as the systems that act on it. The Signal-Based Outbound Engine is the most direct consumer: it takes Tier 1 accounts, the reason codes and the active people, and turns them into timely outreach that references what actually changed at the account. The signal-based outbound engine guide walks through that path from trigger to booked meeting. Speed-to-Lead uses the same score to decide how quickly, and to whom, an inbound request goes.
Further down the funnel, the Pipeline Hygiene Sentinel benefits from knowing which signals sourced an opportunity, because deals that entered on a single uncorroborated surge deserve a harder look before they reach the forecast. On the customer side, the Churn Signal Watchtower applies the same logic to retention: usage, support and engagement signals blended with a clock, rather than one metric treated as the truth. The broader picture sits on the GTM Operations page.
If you are still deciding who should build and own the score, the GTM engineer vs. RevOps manager vs. growth engineer decision tree helps with that call. Our own approach is forward-deployed engineering: backtest against your own closed deals, test the model on past cases, and route live accounts only once it makes the right calls.
Sources: LinkedIn B2B Institute, the 95:5 rule, based on research by John Dawes of the Ehrenberg-Bass Institute (Marketing Week, September 2021). Forrester, The State of Business Buying, 2026 (January 2026, based on Forrester's 2025 Buyers' Journey Survey, as reported by Digital Commerce 360). 6sense, 2025 B2B Buyer Experience study (about 4,000 buyers plus a supplementary sample of 766, November 2025, as summarized by CustomerThink). Adverity, State of Play Research: Data Quality 2025 (CMOs in the US, UK, Germany, Austria and Switzerland, September 2025). Forrester, The 10 Biggest Intent Data Mistakes for B2B Marketing and Sales (blog).
