On July 1, 2026, HubSpot quietly changed its terms of service. Enrichment data from customer CRMs would be pooled and shared across the platform, opt-out by default, communicated only in an email nobody reads and nowhere inside the product. The change was never even given a real product name, which tells you how much HubSpot wanted it noticed. Four days later, after operators lit it up, HubSpot published a post titled “We Got This Wrong. And We Are Fixing It” and killed it. Enrichment stays opt-in.
The privacy story is the easy read. The more interesting one is that HubSpot keeps trying to be a data company and it keeps not working. Salesforce tried it first with Data.com. HubSpot bought Clearbit, and when Clearbit did not bridge the gap, it announced a partnership with ZoomInfo soon after, all while its terms still effectively ban marketing to anyone who has not opted in. Here is the tell that should end the debate: the very data partner HubSpot now leans on, ZoomInfo, publishes that roughly 70% of B2B contact data decays every year. So the argument for pooling your customers’ data is essentially, “our expensive partner data is not good enough, so we are going to use yours to fix theirs.”
This piece covers what actually happened, why operating systems keep losing this fight, what data sharing is genuinely legitimate, where the B2B data stack sits in 2026, and the layer these platforms are structurally missing. Some of what HubSpot reached for is defensible. The way they reached for it, and their track record of reaching, is the story.
TL;DR
- HubSpot pushed an opt-out terms of service change on July 1, 2026 that would have pooled and shared customer enrichment data, then retracted it on July 5 after operator backlash. Enrichment stays opt-in.
- This is HubSpot’s third data-company attempt (after Clearbit and the ZoomInfo/Breeze integration) and the fourth industry-wide after Salesforce’s Data.com. All have failed.
- The reframe that matters: HubSpot’s own data partner, ZoomInfo, says 70% of B2B contacts decay yearly, so pooling your data is an admission their purchased data is not good enough.
- Some sharing is legitimate: system-observed deliverability signals, honeypots, and global complainer lists. Reselling purchased third-party inference into a competitor’s CRM is not.
- The line is system-observed reality versus purchased inference. One is truth, the other is a bet.
- The real gap is a judgment and orchestration layer that self-heals. Utilities do not clean themselves, and buying more data credits will not make them.
Why this problem exists
B2B data rots faster than anyone budgets for.
HubSpot’s own data partner, ZoomInfo, publishes the decay rate. It is the reason every operating system keeps reaching for the data layer.
~70%
of B2B contact data changes every year
~2%
per month leaves the workforce entirely
~4%
per month changes title, team, email, or employer
So the pitch to pool your data reduces to one sentence: our expensive partner data is not good enough, so we are going to use yours to fix theirs.
1. What HubSpot Actually Did (and Undid)
HubSpot tried to turn every customer’s CRM into a shared data pool, and buried the ask where they hoped nobody would look. The change arrived July 1 inside a terms of service and legal update, never given a real product name, pooling enrichment data across customers, opt-out by default, delivered only by email and never surfaced in the portal. On July 5, HubSpot posted “We Got This Wrong. And We Are Fixing It,” and the company committed to keeping enrichment opt-in.
Why it matters
The original move was worse than the retraction was good. Opt-out by default, buried in a ToS email, invisible in the product: those are three separate decisions that each assume consent rather than earn it. The retraction itself is real, but the lesson may not have landed. HubSpot still believes there is a “trusted prospecting” opportunity here, and the phrase they leaned on, “business-card-level” data, tells you they thought they already had a right to it.
How to read it operationally
- Treat the speed of the reversal as a signal of how weak the ground was, not how responsive HubSpot is. Four days is a retreat under fire.
- Note who moved the needle: working operators and practitioners, not industry analysts. That kind of backlash is the accountability mechanism that actually works here.
- Expect a reintroduction. When it returns, look for whether it is opt-in, shown in the portal, and specific about which fields are in scope.
Watch-outs
- HubSpot’s stated policy is “You control your data. This has always been our policy and will not change.” It was, until it was not, and then it was again. Read the terms, not the reassurance.
- A softer relaunch of the same idea is more dangerous than this clumsy one, because it will be easier to accept without noticing.
2. The Data Business Curse for Operating Systems
Every large operating system that has tried to bolt on a data business has walked away from it or buried it, and the pattern is remarkably consistent. Salesforce bought Jigsaw for $142M in April 2010, renamed it Data.com in 2011, and shuttered it in 2018 after selling pieces to Dun & Bradstreet. Eight years, one failed experiment. HubSpot bought Clearbit for $150M on November 1, 2023, and the tool sits inside HubSpot as a nearly separate platform.
Why it matters
Clearbit inside HubSpot is hard to build workflows on, hard to report on, and hard to take action from, so people are not really using it. Two and a half years after that acquisition, in April 2026, HubSpot announced that ZoomInfo powers its Breeze Prospecting Agent through GTM.AI’s MCP server. Clearbit did not bridge the gap, and the ZoomInfo partnership makes that clear: HubSpot is admitting its own data is not good enough and partnering for someone else’s, all while its terms still effectively ban marketing to anyone who has not opted in. All of this happens with the stock down 70% heading into the UNBOUND rebrand.
How to use it operationally
- Treat any “native data” feature from a utility vendor as unproven until you have tested accuracy against your own list.
- Assume acquired data products degrade after the acquisition. Jigsaw and Clearbit are both nine-figure examples.
- Watch what a vendor integrates, not just what it builds. HubSpot renting ZoomInfo tells you more than any Clearbit roadmap slide.
Watch-outs
- Data quality is a full-time job, and multi-tenant CRMs are structurally bad at it. The org chart and the liability posture both fight it.
- A big acquisition price is not evidence of a working product. Salesforce and HubSpot have spent nearly $300M combined on data with little to show.
The pattern, at a glance
Operating systems keep buying data. It keeps not working.
Two nine-figure acquisitions and one telling partnership, spread across fifteen years.
Salesforce buys Jigsaw → Data.com
$142M in 2010. Renamed Data.com in 2011, sold in pieces to Dun & Bradstreet, closed by 2018. An eight-year failed experiment.
HubSpot buys Clearbit
$150M in November 2023. Sits inside HubSpot as a near-separate platform: hard to build workflows on, hard to report on, so people are not really using it.
HubSpot partners with ZoomInfo
April 2026. ZoomInfo now powers HubSpot’s Breeze Prospecting Agent. Clearbit did not bridge the gap, so HubSpot is renting someone else’s data instead.
3. What Data Sharing Is Legitimate
Not all cross-customer sharing is the same, and the line is cleaner than the backlash suggested: system-observed reality is fair, purchased inference is not. One is truth the platform witnessed. The other is a bet someone else sold, and reselling it into a competitor’s CRM is where this crosses from hygiene into harm.
Why it matters
Some sharing is genuinely good for everyone. System-generated deliverability signals (hard bounces, retirees, people who left the company, addresses killed by M&A) protect every sender when they are pooled. So do honeypot files and global complainer suppression lists, which are just good hygiene, and reach data that flags companies behind security gateways nobody can get through. Pooled data is not exotic here. The grill ad that follows you around the internet for a week is the same mechanic, consumer-side, and nobody storms the barricades over it. The egregious version is different: ZoomInfo saying “this person changed jobs,” a purchased inference you paid for, being pushed into your competitor’s CRM. The platform never observed that. Someone is reselling an asset you already bought.
How to use it operationally
- Opt into pooled deliverability signals. As a practitioner, this is the sharing I would advocate every client and partner turn on.
- Keep purchased, third-party enrichment out of any shared pool. If you paid a vendor to append it, it is not the platform’s to redistribute.
- Handle soft bounces with patience. “Out of office” and “no longer with the company” auto-replies can look like hard bounces for up to six weeks before the record actually turns.
Watch-outs
- Stakes rise fast with business-card data. It is expensive to acquire, and a single field can make or break a deal, so handing it to a competitor because provenance was ignored is not a rounding error.
- The decay that drives all of this is relentless: roughly 70% of B2B contact data changes each year, about 2% per month leaves the workforce entirely, and another 4% or so changes a title, department, email, or employer.
- Observed-versus-inference is the test to apply to any sharing proposal. If the platform did not witness it, sharing it is a bet made with your data.
What’s fair to share
Three tiers of sharing, one clean line.
The test for any proposal: did the platform observe it, or did someone buy it? Observed reality is fair. Purchased inference is not.
Pool it and every sender benefits. Worth opting into.
Expensive to acquire, and a single field can make or break a deal. Only shareable if the platform originated it, never if you licensed it from a third party.
Never shareable, in any state of the system. A competitor jumping your active cycle is the exact harm to prevent.
4. The State of B2B Data in 2026
The typical B2B company runs three data sources, quietly pays for the same records more than once, and does not need the enterprise infrastructure vendors keep selling it. ZoomInfo is still the biggest, but new players are taking share and the category is being unbundled in real time.
Why it matters
Three sources is the industry norm: an aggregator like ZoomInfo, a marketing enrichment tool, and a sequencing tool with data like Apollo. Many of these providers quietly buy from the same upstream originators, People Data Labs, RevenueBase, and 5×5, so you may be paying twice for the same data. Meanwhile Snowflake and CDPs are dramatic overkill for most B2B companies, which run databases of a few thousand accounts, one to three ICPs, segments around 2,000 companies, and buying committees of three to five people. That is not consumer-grade infrastructure territory.
How to use it operationally
- Map your sources against their upstream originators before renewing. If two vendors resell the same 5×5 or PDL data, you are double-paying.
- Right-size the stack. A few thousand accounts and a handful of ICPs do not justify a CDP or a Snowflake build.
- Reassess pricing on the incumbents. New entrants are more cost-effective than ZoomInfo, whose stock is well down.
Watch-outs
- The old guard is exiting. Dun & Bradstreet was taken private by Clearlake in August 2025 at $9.15 a share.
- Cheaper is not automatically better. Test accuracy on your own ICP before you switch, because a low price on stale data is still a bad deal.
5. What CRMs Are Actually Missing
HubSpot’s real problem is structural: a collection of utilities is being sold as if it were judgment and orchestration. Utilities are useful, and you should keep them. They do not configure themselves to your industry, and they do not make decisions.
Why it matters
Out of the box, HubSpot is one-size-fits-all across every industry, company size, and geography. There are no preconfigured B2B workflows, no industry-specific configuration, and no inference layer. Hand it to a junior marketer and it is essentially a toolbox with no assembly instructions. Worse, the utilities do not self-heal: hard bounces are not suppressed automatically, lists do not clean themselves, and nurture flows do not update themselves. That gap, the judgment and orchestration sitting on top of the utility, is where a new category of tools now operates, Outkeep among them. If you were sitting in HubSpot’s boardroom, the advice writes itself: build the judgment layer, and stop trying to be a data company.
Operator moves
- Keep the utility for what it is good at (storage, sending, the system of record) and stop expecting it to supply judgment.
- Pay utility prices for the utility. Sending that once implied a six-dollar CPM runs at ten cents now, so do not let a warehouse charge you data-layer money.
- Build or buy the layer that self-heals: automatic suppression, list hygiene, and nurture logic that updates as the data changes.
- Treat industry configuration as your responsibility or your partner’s, never as something the one-size-fits-all platform will hand you.
Watch-outs
- HubSpot already admitted inbound is dead when it renamed INBOUND to UNBOUND for 2026, in their own words because AI killed the inbound methodology. They have not replaced it with anything coherent yet.
- Turning a utility of this size toward a new model is like turning the Titanic away from the iceberg. The intent can be right and the mass still too great to move in time.
- Do not confuse a system of record with a system of action. The record sits still. The action layer is what keeps the database alive.
Where the value moved
The utility stores and sends. The judgment layer keeps data alive.
Keep the utility for what it is good at. The gap it will not fill is where the value has moved.
The utility layer
Storage and transport
Useful, sticky, and one-size-fits-all. It sits still and waits.
The judgment layer
Orchestration and hygiene
The self-healing layer that acts as the data changes underneath you.
Context on Outkeep’s Approach
Outkeep operates in the gap this article describes: the judgment and orchestration layer that sits on top of the utility rather than a warehouse beside it. We spend our time in deliverability, sending reputation, and the daily reality that a database is a leaky bucket losing several percent of its value every month whether anyone is watching or not.
That is why we read the HubSpot episode as directionally revealing and tactically wrong. Pooling system-observed deliverability signals is defensible and useful. Reselling purchased inference into a competitor’s CRM is not, and no amount of “trusted prospecting” language changes which side of the line it sits on. The work worth doing is closing the feedback loop these platforms leave open, and that work does not happen inside a one-size-fits-all utility.
FAQ for Modern B2B Email Programs
What did HubSpot change on July 1, 2026?
It pushed a terms of service update that would have pooled enrichment data from customer CRMs and shared it across the platform, opt-out by default. The change was never given a real product name. HubSpot retracted it on July 5 and committed to keeping enrichment opt-in.
Why did HubSpot retract it so fast?
Working operators publicly pushed back within days. The change was opt-out by default, buried in a terms of service email, and never shown in the product, which made it indefensible once people noticed.
Is cross-customer data sharing ever legitimate?
Yes, when it is system-observed reality. Pooled hard bounces, retirees, honeypots, and global complainer suppression lists protect every sender. Reselling purchased third-party inference into a competitor’s CRM is not legitimate.
What is the difference between observed data and inference?
Observed data is something the platform actually witnessed, like an email hard-bouncing. Inference is a purchased guess, like a vendor claiming a contact changed jobs. Sharing the first is hygiene, sharing the second is redistributing an asset you paid for.
Why do CRMs keep failing at data?
Multi-tenant utilities are structurally bad at data quality, which is a full-time job. Salesforce’s Data.com (an eight-year failure) and HubSpot’s underused Clearbit are both nine-figure proof.
How many data sources should a B2B company run?
About three: an aggregator, a marketing enrichment tool, and a sequencing tool with data. Check whether they share upstream originators like People Data Labs, RevenueBase, or 5×5, because you may be paying twice.
How fast does B2B contact data decay?
Roughly 70% changes each year. About 2% per month leaves the workforce entirely, and another 4% or so changes a title, department, email, or employer.
What are CRMs like HubSpot actually missing?
A judgment and orchestration layer that self-heals: automatic bounce suppression, list hygiene, and nurture logic that updates as data changes. HubSpot supplies utilities and one-size-fits-all defaults, not industry configuration or an inference layer.
