B2B sales professionals believe that roughly a third of the contact and account records held in their customer relationship management systems are inaccurate, incomplete or outdated, according to research Firmable published on September 9, 2026. The sales intelligence company surveyed 222 B2B sales professionals and put the estimated share at 32%. Three in four of them said they would choose perfect prospect data over the best AI sales assistant on the market, yet only 23% use AI to score or prioritise leads.

In Short

A sales data company asked 222 people who sell to other businesses how reliable the customer records in their sales software are, and they estimated that about a third are wrong, missing details or out of date. Those same records decide who salespeople contact, and they also end up in the customer lists companies send to ad platforms for targeting and measurement. Most of the salespeople said they would rather have clean data than a better AI assistant, a sign that many teams see bad records, rather than a shortage of AI tools, as the thing holding them back.

How the survey was built

Firmable describes itself as an AI-native sales platform for B2B sales teams, selling prospect data, buying signals and what it calls agent-driven actions to customers in the United States, Canada and the Asia-Pacific region. Its website positions the product against ZoomInfo and Apollo and lists an integration with Clay. The findings appeared in a post titled What hyper-growth sales teams do differently, written by Chester Thompson, the company's vice president for the Americas. They reached PPC Land by email on September 22, thirteen days later.

Firmable surveyed 222 B2B sales professionals, "from frontline reps to senior leaders", and compared two groups. Hyper-growth teams were defined as those whose business grew by at least 30% over the past year. Everyone else formed the comparison set. "What set them apart came down to cleaner data, sharper targeting, and a few repeatable habits," Thompson wrote.

The methodology note runs to two sentences, and much of what would be needed to weigh the results is missing. It gives no fieldwork dates, no countries, no count of how many of the 222 respondents fell into the hyper-growth group, no breakdown by company size or sector and no margin of error. Growth appears to be self-reported and unverified. Without the size of each group, it is impossible to judge whether gaps of nine or ten points are robust or an artefact of a small subsample.

Then there is the matter of interest. Firmable sells the thing its headline finding says sellers want. That does not make the numbers wrong, but it does shape which numbers were put at the top.

A third of the database, by the sellers' own estimate

The figure doing most of the work is 32%. According to Firmable, B2B sales teams believe that about a third of the contact and account data in their CRM is "inaccurate, incomplete, or outdated." Three separate failures are folded into that phrase. A mistyped email address, a record missing a job title and a buyer who left the company two years ago are different defects with different causes, and the survey does not say how the 32% divides between them.

Nor is it a measurement. Respondents were asked what they believe, and no records were inspected. The post does not state whether 32% is a mean or a median of individual answers. Promotional material circulated on Firmable's behalf on September 22 went further than the post itself, presenting the number as the share of CRM data that is inaccurate and leaving out the incomplete and outdated categories. Firmable's own wording is the more careful version.

The estimate lands in familiar territory. When Adverity surveyed 200 chief marketing officers in the United States, the United Kingdom, Germany, Austria and Switzerland in 2025, respondents judged 45% of the marketing data used for business decisions to be incomplete, inaccurate or outdated, and none rated their data more than 75% reliable. Gartner has put the average cost of poor data quality to organisations at $12.9 million a year. HubSpot offered a similar diagnosis in July 2026, arguing that prospecting often fails because reps work from lists in which half the contacts have changed jobs. Its proposed remedy, a shared enrichment dataset, did not survive contact with customers: HubSpot withdrew its Contact Discovery terms four days after publishing them, following questions about how customer data would feed the shared pool.

Perfect data against the best assistant

Three in four respondents (75%) said they would rather have perfect prospect data than the best AI sales assistant on the market. As survey design, the question sets an unattainable option - no vendor sells perfect data - against one specific category of AI tool, not AI as a whole. It captures a priority. It does not describe what anyone is buying.

Usage figures say more. Across all 222 respondents, the most common use of AI was drafting outreach emails or messages, at 45%. Summarising calls followed at 43%, researching prospects at 39% and forecasting or reporting at 31%. Two uses tied at the bottom of Firmable's chart, at 23% each: scoring or prioritising leads, and cleaning CRM data. The figures sum to more than 200%, so respondents were evidently able to select several uses.

Firmable singled out the lead scoring figure, suggesting that many teams "still aren't using AI where it could directly shape pipeline quality." The tie beside it is at least as telling. Cleaning CRM data is the task most directly aimed at the 32% problem, and 77% of respondents do not use AI for it. AI, in other words, is used about twice as often to write messages to a database its users consider unreliable as to repair it.

The split between the two groups is where Firmable locates its argument about AI. Nearly three in ten hyper-growth teams (29%) said their use of AI significantly improved sales results over the past year, against roughly one in ten (11%) of everyone else. Firmable describes that as nearly three times more likely; the ratio of the two published figures is about 2.6. Seven in ten hyper-growth teams (70%) said they were confident they use AI better than their competitors. No equivalent figure for slower-growing teams was published, so that number cannot be compared with anything.

Causation runs in an open direction. A business that grew by 30% or more has every reason to credit the tools it adopted along the way; a team missing its targets has little reason to praise its software. The survey records perceived contribution, not measured contribution. Thompson's own framing is cautious on this point: "The teams growing fastest are not betting everything on AI, nor are they ignoring it."

What the fastest-growing teams do with their week

The most concrete operational gap concerns time. Hyper-growth teams spend a median of 65% of their week selling, compared with 53% for everyone else, according to Firmable, leaving less of the week for admin, research and data entry. The difference is 12 percentage points - on a notional 40-hour week, just under five hours.

One detail complicates the picture. Hyper-growth teams were also more likely to conduct structured prospect research before outreach, at 41% against 32%, even though research sits in the non-selling share of their week. The post does not reconcile the two findings; a definitional overlap between research and selling is one plausible explanation.

Firmable also published a list of habits that hyper-growth teams keep consistently. Daily prioritisation led at 58%, followed by CRM data hygiene at 48%, outreach timing at 44%, prospect research at 41%, and list-building, follow-up cadence and automation at 37% each. Comparisons with the rest of the sample were published for only three of those items: outreach timing (44% against 31%), prospect research (41% against 32%) and systematic list-building (37% against 24%). Hyper-growth teams also make more follow-up attempts before abandoning a prospect, a median of four against three.

Data hygiene ranking second fits the theme of the study, although without a comparison figure it cannot be said whether slower-growing teams do it any less.

How they keep score

The two groups measure success differently. Quota attainment is the primary measure for 40% of hyper-growth teams against 31% of everyone else. Among hyper-growth teams, pipeline growth followed at 16%, rep productivity at 15%, sales velocity at 14%, customer retention at 10% and consistent outbound activity at 5%. Retention was the primary measure for 23% of the other group, more than twice the hyper-growth share. Teams in rapid expansion, on this evidence, judge themselves on new revenue rather than on keeping existing customers - a pattern that probably says as much about the stage of the business as about sales technique.

Habits over talent

Respondents across the sample played down individual ability. More than nine in ten (95%) said what sets their top-performing reps apart is something other than raw talent, with better habits and consistency (31%) and better relationships (28%) the most common answers. Some 83% agreed they could grow faster by improving their habits rather than simply working harder, 80% said their best-performing reps follow consistent daily routines, and 67% said inconsistent processes hold their team back more than a lack of effort does. About one in five (21%) named time spent on the wrong prospects as the biggest drag on growth - a finding that loops back to the quality of the records used to choose those prospects.

A companion survey, two respondents smaller

Firmable published a second study on the same day, under the same byline. The agent-ready revenue team describes a survey of 220 B2B sales professionals, two fewer than the habits study, and its methodology note uses the same campaign framing. Firmable does not say whether the two studies drew on a single fieldwork exercise, why the sample sizes differ, or whether one group of respondents is a subset of the other.

Its numbers sharpen the data argument. Only 14% of respondents said an AI agent could act on their current sales data as it stands, without cleanup first. Some 80% said they lose at least 10% of their selling time to finding, cleaning or updating data, and 58% had discovered bad sales data only after contacting a prospect. Just 16% would let an agent take a sales action without reviewing it, with sales leaders more willing than individual contributors (21% against 12%), and half said they personally would be blamed if an agent acted on bad data. About half (51%) described their sales stack as held together by manual workarounds, and 63% juggle three to five separate tools to research, enrich and reach a single prospect. Nine in ten use AI for at least one core selling task.

Different wording, different sample sizes, same direction: adoption is broad, and the records underneath it are not trusted.

Why CRM accuracy reaches the media budget

For advertisers, none of this stays on the sales floor. Customer lists uploaded for targeting and suppression, offline conversions returned to bidding systems and account lists pushed into B2B campaigns all start in the CRM. Google stopped accepting Customer Match uploads through the Google Ads API from April 1, 2026, routing them to the Data Manager API instead. At Google Marketing Live in May, the company described lead intent scores meant to surface the leads most likely to close, and journey-aware bidding that learns from the full lead-to-sale path an advertiser shares with Google Ads. Both depend on the record at the end of that path being right. When HubSpot brought TikTok advertising into its CRM in April 2026, PPC Land noted that customers with incomplete CRM hygiene would get less from the integration than those with accurate contact records and deal histories.

Some of the contamination is created by paid media itself. Lunio, which sells invalid traffic detection, surveyed 131 senior marketing leaders and released the results in July 2026: 22.9% named sales time lost to bot-generated lead forms as the most damaging cost, and 22.1% named corrupted CRM data. The company separately measured LinkedIn's invalid traffic rate at 17.62% in the first quarter of 2026. Firmable's survey does not ask where bad records originate. But a fabricated form submission lands in the same database a rep later works from, and, if uploaded as a conversion, in the signal a bidding system learns from.

B2B contact data also travels well beyond the CRM. Intentsify made more than 700 pre-built B2B audience segments available inside The Trade Desk and DV360 in April 2026, and MNTN opened ZoomInfo's database of 100 million decision-makers to connected television buyers in July 2025. Firmographic and contact records now flow into demand-side platforms as routinely as into sales sequences, which turns an error rate estimated by sales teams into a question about audience quality as well. Italy's data protection authority fined Lusha 2 million euros and ordered it to erase Italian contact data in a decision made public in July 2026. In Europe, perfect prospect data is a legal question before it is a technical one.

Agents raise the cost of a wrong record

The timing of the research matters because the industry is moving from AI that drafts to AI that acts. Intentsify argued in a May 2026 white paper that the central failure mode of AI agents in B2B sales and marketing is poor input data rather than weak model capability, citing a Gartner forecast that more than 40% of agentic AI projects will be cancelled by the end of 2027. In August the same company began scoring entire buying groups rather than individual leads, pitching a readiness score as a replacement for the marketing qualified lead. Clay, which says more than 500,000 go-to-market teams use its platform, added Intentsify's intent data on August 20 to workflows its customers already build around signals such as job changes and funding rounds. A job change is precisely the kind of event that turns an accurate record into an outdated one.

Salesforce is moving the same way. On August 26 it put 37 prebuilt sales skills inside Anthropic's Claude for pilot customers, an arrangement in which the CRM remains the record against which lead scoring, attribution and lifecycle triggers are reconciled. Its own researchers had earlier found, in the CRMArena-Pro benchmark, that leading agents succeeded in 58% of single-turn business tasks and 35% of multi-turn ones. Dreamdata, which added a Model Context Protocol (MCP) server on September 1, puts the average B2B buyer journey at 272 days, 88 touchpoints and 10 stakeholders. Each stakeholder is another contact record, and each is subject to the decay Firmable's respondents describe.

Marketers are hesitant too. StackAdapt found in August that only 6% of marketers act on in-platform AI recommendations almost always, with 49% citing fragmented data pipelines as a barrier to delegating decisions to AI. TransUnion's survey of 100 senior US marketing leaders found only 36% rating their data and process readiness as high. Firmable's 14% figure for agent-ready sales data belongs to the same family of findings.

Firmable is no bystander. A banner across its website says the product now works inside Claude, ChatGPT and Copilot through MCP, placing its own data inside agent workflows that, by its own survey, run on records few sellers trust. Whether a vendor-supplied record is more accurate than the one it replaces is a separate matter. It is also one that a self-reported estimate of 32%, however plausible, cannot settle.

Timeline

Summary

Who: Firmable, a sales intelligence company serving customers in the United States, Canada and the Asia-Pacific region, which surveyed 222 B2B sales professionals ranging from frontline reps to senior sales leaders. The findings were written up by Chester Thompson, Firmable's vice president for the Americas.

What: Respondents estimated that 32% of the contact and account data in their CRM systems is inaccurate, incomplete or outdated. Some 75% said they would choose perfect prospect data over the best AI sales assistant, and 23% use AI to score or prioritise leads, the same share that uses it to clean CRM data. Teams growing at least 30% a year reported significant AI gains more often (29% against 11%) and spent a median of 65% of their week selling, against 53% for other teams.

When: Firmable published the research on September 9, 2026, alongside a companion survey of 220 sales professionals, and circulated the findings to PPC Land by email on September 22, 2026. Fieldwork dates were not disclosed.

Where: On Firmable's website. The countries in which respondents work were not disclosed.

Why: CRM records determine which prospects sales teams pursue, what AI agents act on and which customer lists and conversion signals reach advertising platforms. A self-reported error rate of roughly a third, from a vendor that sells prospect data, points to a data problem that sits upstream of both AI adoption and media performance, though the survey measures perception rather than the records themselves.