The Prospecting Data Problem Nobody Talks About

Four days, 1,207 calls, $1,091 spent, zero meetings booked. The data was fine — the targeting wasn't. Why prospecting needs an intelligence layer.

The $1,091 campaign that booked zero meetings

A healthcare RCM company hired an outsourced SDR vendor for outbound cold calling. Two reps. 1,207 calls. Four business days. Around $1,091 spent. The result: zero meetings booked. Seven warm signals — 0.6% of dialed contacts. The instinct is to blame the SDRs, the list, or the script. None of those was the problem.

The data was real. The targeting was off.

We ran a representative sample of the same list through InsightSignal afterward. Across the sample, 95 to 97 percent of the companies were real, verifiable healthcare entities. The data was fine. The list itself was just structurally mistargeted:

• ~40% sat outside the client's stated ICP. Too big, too small, wrong industry, wrong geography, or running a billing platform listed as a disqualifier in the client's own sales playbook.
• Two were direct competitors. Active in the same RCM segment.
• ~13% were DBAs of enterprise health systems. Local facilities where the RCM contract decision happens at the parent level — not where the SDRs were dialing.

This is the part of "the data problem" most enterprise AI conversations skip past. MIT's recent study of 300-plus enterprise AI deployments found that 95% of pilots deliver zero measurable return. McKinsey's 2025 survey found two-thirds of organizations haven't begun scaling AI — with data as the primary blocker. Andy Kohm named the trap directly in a recent Forbes piece on Intelligence as a Service: the conventional wisdom says clean your data before building intelligence on top of it, and that framing is what guarantees the project never finishes.

What "bad data" actually means in prospecting

Every revenue team has the same stack: a CRM, enrichment tools, a sequencer, maybe an intent subscription. On paper, the inputs are there. In practice, BDRs are working with several failure modes at once:

• Stale. That decision-maker changed jobs four months ago. Your sequence is still hitting the old inbox.
• Partial. You know the company name, but not whether they just raised, just hired a VP of Engineering, or just got acquired — the three signals that would have told you when to reach out.
• Ungraded. Two thousand rows of "prospects" with no opinion on which fifty to call first. The list is treated as uniform when it isn't.
• Context-free. You have a name and a title. You don't have why-this-person, why-now, and what-to-say.
• Structurally mistargeted. Even when individual rows are verifiable, the list can be 40% wrong if no one is asking whether the companies actually match the buyer profile.

The output is a system of record masquerading as a system of intelligence — a dashboard waiting for someone to feed it clean data.

The trap of "fix the data first"

The standard fix is a multi-quarter hygiene project. RevOps cleans the CRM. Ops buys a new enrichment vendor. Someone runs a deduplication pass. Six months later, the data is "clean" — and already stale, because prospect data has a half-life measured in weeks, not years. The data problem isn't a one-time project. It's continuous, and it requires intelligence to solve. So we built InsightSignal the other way around: the intelligence layer is the thing that validates, cleans, aligns, and augments data — continuously, automatically. If a platform can't do that, it isn't intelligence. It's a dashboard waiting on a cleanup project that never finishes.

Even the supply side is unbundling

The shift from SaaS to outcome-based intelligence isn't only happening at the application layer. The data-supply layer is unbundling too — in two distinct places:

But API access to fresh data — firmographic or web — still doesn't tell a BDR who to call this hour. It's raw signal supply, not prescription. That gap is where the intelligence layer sits. InsightSignal consumes from whatever source makes sense — a CRM export, a Crustdata or PDL pull, a Perplexity or Tavily query for fresh signals, or a hand-curated seed list — and turns raw signal into graded, evidence-backed recommendations. We're the layer above the new supply-side vendors, not in competition with them.

Seed, Curate, Enhance: how InsightSignal handles it

InsightSignal treats the data layer as the product, not the prerequisite. The pipeline runs in three stages, agentically, on every run:

• Seed. Start with whatever you have — a target persona, an ICP definition, a list of companies, or a vague description of who you want to reach. The seed doesn't need to be clean.
• Curate. Agents qualify and grade every row against the context you provided. Each prospect comes back with a grade, a documented rationale, and the signals that drove it — with source, confidence score, and verification timestamp attached.
• Enhance. Agents fill in missing context and attach the why-now signal that turns a name into an opening line. When fresh evidence matters — a recent funding round, a new hire, a regulatory announcement — agents pull live web signal on demand via search APIs. In the case above, this took rows with verified company phone coverage from 54% to 99%.

Because prospect data is time-varying — people change jobs, companies pivot, intent signals decay — the loop runs continuously. The shortlist a BDR sees today reflects this week's reality, not last quarter's snapshot.

Did the grading actually predict outcomes?

A grading system is only as good as its correlation with reality. We ran InsightSignal's grades against the SDR vendor's actual call outcomes:

• All 7 warm-signal companies the SDRs surfaced were graded A or B by InsightSignal.
• 6 of those 7 were classified as Target with 88-95% intent alignment.
• The 7th was correctly excluded — a $301M county-owned hospital, 6× above the client's revenue ceiling. A warm signal there would have produced an enthusiastic meeting that couldn't close.

86% warm-signal preservation, with one correct override on a structural misfit. The grade ships with its rationale attached, and the rationale is auditable against the client's own playbook.

Closing the loop: Find More Like This

Seed → Curate → Enhance describes one pass. The system is built around a loop, not a line. Your A-grade winners — verified, scored, proven — are the strongest possible seed for the next pass. InsightSignal's Find More Like This feature handles that loop: pick 1-5 winners, and the wizard synthesizes an editable ICP, fetches matching candidates from the data network, and runs the same multi-source verification on what you keep. The output: a new cohort of A and B prospects ready to outreach — plus the seeds for the next run.

The metrics that actually matter

Seat counts and license utilization don't tell you whether a prospecting motion is working. Five metrics do — here's what each looked like in the case above:

• Throughput. A 390-company list collapses to a tiered shortlist: ~53% actionable matches, ~40% noise to remove, ~13% needing DBA-to-parent rerouting.
• Speed. 161 companies verified across three sample sets in 119 minutes — signal-to-recommendation in hours, not weeks.
• Quality. 86% warm-signal preservation, with one structural miss correctly caught. Every prospect carries grade, intent reasoning, and source evidence.
• Coverage. One pipeline spans firmographic fit, persona, intent, billing-platform disqualifiers, DBA resolution, competitor detection, and phone enrichment.
• Trust. Every data point cites source, confidence, and timestamp. Every recommendation is auditable against the buyer's own playbook.

Throughput, speed, quality, coverage, trust. None of those appear on a seat-license renewal page.

What screening cost vs what it saved

Per-seat pricing made sense when the bottleneck was access to information and the unit of value was a human operator. When a single agent does the work that once required a team of analysts, the math breaks. Salesforce's Agentforce charges per conversation. Credit-based pricing in AI-native software grew 126% year over year in 2025. The industry is voting with its pricing pages. Applied to the same four-day campaign:

Screening compounds into permanent verified intelligence; a wasted dial learns nothing. There's no annual seat license to negotiate down, no shelfware risk, and cost-to-first-outcome is measured in dollars, not contracts.

Teams: credits allocated to outcomes

The Teams version, shipping this May, takes outcome-based pricing further. Admins allocate credits per BDR based on performance and outcome. Agents producing pipeline get more runway. Agents who aren't get coached or recalibrated before more spend flows out. Cost control becomes proportional to outcome — at the individual contributor level — instead of a flat dollars-times-headcount line on the renewal. The conversation between a CRO and finance shifts: instead of justifying seat counts, you're justifying outcomes per credit allocated.

The question for revenue leaders

Strip the AI vocabulary away, and the question is simple. Which part of your stack is grading prospects and telling reps where to spend the next hour — and which parts are just storing names you paid someone to scrape? If the answer is the latter, the data problem isn't waiting for a cleanup project. It's waiting for an intelligence layer.

Sources

• Andy Kohm — The Death of SaaS (As We Knew It) And The Rise Of Intelligence As A Service, Forbes Technology Council (Mar 2026)