AI-generated
Moatsight's own AI workflow wrote this edition. No person edited it.
- Written Nov 10, 2026 from 25 sources, including the 10-Q filed Nov 9, 2026.
- Argued by a bear and a bull critic.
- Before it was published, a separate AI check traced the key figures the company reported to saved copies of its filings.
Business explainer
Loud claims, checked: 1 not supported · 3 partly supported · 3 open
| Verdict | Claim | Who and when |
|---|---|---|
| Partly supportedThe word opted-in was dropped from the 10-K after fiscal 2023, and the company's November 2024 counts give 110 million US email opt-ins against 245 million with digital permission; whether investors were misled is open | Zeta's US data set was not opted-in as the company once described itIn Business explainer | Culper Research, short report of November 13, 2024; plaintiffs in the securities class action |
| OpenThe company says revenue from customers who are also vendors is insignificant; no court, regulator or restatement has decided the claim | Revenue was flattered by two-way contracts with data suppliers, and much of the data came from consumer websites that do not make its use clearIn Business explainer | Culper Research, November 13, 2024 |
| OpenAn allegation in a complaint that no court has decided | The chief executive sold shares through trusts and companies to avoid reportingIn Business explainer | Plaintiffs' amended complaint in the securities class action, May 2025, as PPC Land reported on July 13, 2026 |
| Not supportedOn the present filings: no AI or partnership revenue is reported and revenue is still marketing revenue | Zeta is now an intelligent AI infrastructure company, comparable with Palantir, Datadog or SnowflakeIn What changed | The company, in its release and earnings supplemental of November 9, 2026 |
| OpenNo partnership revenue is disclosed and no year is given | The Palantir partnership can add more than $100 million of annual revenueIn What changed | The CEO, on the November 9, 2026 earnings call |
| Partly supportedOrganic growth rose from 28% to 30%; most of the rest came from Marigold and election spending | Revenue growth of 46% shows the business acceleratingIn What changed | The company's release of November 9, 2026 |
| Partly supportedThe beats are real but run at about the 2% to 5% cushion the CFO says is built into guidance | Twenty-one straight beat-and-raise quarters show momentum beyond planIn What changed | The company's release of November 9, 2026 |
A verdict is the Moatsight view of what the filings show, not a finding of fact.
The core#
Zeta Global sells marketing software to large companies: email and messaging, a customer database, and programmatic ad buying. All three are built on its own file of US consumer identities and the signals attached to them. Revenue is mostly usage-based, so Zeta earns more when its clients send more messages and buy more media through it. Two things decide what the business is worth. The first is how much growth remains once acquisitions and election spending are taken out: about 30% in the third quarter of fiscal 2026, and guided to 22–24% in the fourth. The second is whether that growth turns into cash after stock-based pay, which is still large. Behind both sits a question about the data. A short seller and a securities class action dispute what consent stands behind the file of identities. The company's own November 2024 counts give 245 million US individuals who agreed to publishers' terms of service and 110 million who opted in to email. The "AI infrastructure" story (Athena, Palantir, OpenAI) is so far a story about adoption. No revenue line yet shows it.
What Zeta sells, and to whom#
Zeta reports one operating segment. The Zeta Marketing Platform bundles three jobs a large consumer brand usually buys from different vendors:
- Acquire. A demand-side platform buys display, connected-TV and social ads for a brand, targeted with Zeta's identity data.
- Retain. An email service provider ("Zeta Messaging") sends and personalises the brand's own email, SMS and app messages.
- Grow. A customer data platform ("CDP+") merges the brand's own customer records with Zeta's identity file, so the brand can recognise and score people across channels.
Athena, launched in 2025, is a conversational AI agent laid over the platform. Users can ask it for an audience or a campaign instead of building one by hand. AthenaOS, presented at the company's Zeta Live event in October 2026, extends this to what the company calls enterprise intelligence. The CEO said on the November 9, 2026 earnings call that it is "in alpha now" and that general availability is set for "the first quarter of 2027".
The FY2025 10-K lists the revenue sources as "subscription fees, volume-based utilization fees and fees for professional services". It adds that "a substantial portion of our revenue is derived from usage-based pricing, which is less stable than subscription-based pricing". Clients can usually change their spend or end a contract at short notice. Revenue therefore behaves more like a consumption business than like a software subscription. It rises quickly when clients' campaigns grow, and it can fall just as quickly.
Zeta also reports two revenue types:
- Direct platform: revenue earned entirely through Zeta's own platform and owned channels.
- Integrated platform: revenue earned through integrations with third parties, for example social ads bought for an agency.
Direct revenue carries more operating leverage, because no third party takes a share of the spend.
| Period | Direct platform, % of revenue | Integrated platform, % of revenue |
|---|---|---|
| FY2024 | 70% | 30% |
| FY2025 | 74% | 26% |
| Q3 2025 | 75% | 25% |
| Q2 2026 | 72% | 28% |
| Q3 2026 | 71% | 29% |
The direct share has fallen four points in a year. The CFO said on the August 4, 2026 call that newer agency clients start with social buying, which counts as integrated revenue. The company's own 2028 model allows 70–75%, so the share is near the bottom of that range.
Who buys. In fiscal 2025, consumer and retail brands gave 24% of revenue, travel and hospitality 11%, insurance 11%, telecoms 10% and financial services 8% (10-K). Spend is concentrated:
- The top ten customers gave more than a third of 2025 revenue.
- One customer was above 10% of revenue in fiscal 2025 and in the first nine months of 2026. The filings do not name it.
- "Super-scaled" customers, those paying at least $1 million over twelve months, were 87% of 2025 revenue.
Revenue is almost all American. US clients gave $455.3 million of the third quarter's $492.7 million.
The unit economics of one large customer#
The relationship with a large customer is the engine of the business. The company's own disclosures show that customers who stay longer spend much more. At the end of 2025, scaled customers with Zeta for more than five years averaged $3.9 million of revenue a year. Customers in their first year averaged $0.7 million. About 90% of revenue came from scaled customers who had been with Zeta for more than a year (third-quarter 2026 earnings supplemental).
| Quarter | Super-scaled customers | Super-scaled ARPU, USD million per quarter |
|---|---|---|
| Q4 2024 | 148 | 1.73 |
| Q3 2025 | 180 | 1.6 |
| Q4 2025 | 184 | n/a |
| Q1 2026 | 189 | 1.7 |
| Q2 2026 | 197 | 1.8 |
| Q3 2026 | 212 | 2.0 |
ARPU is the quarter's revenue divided by the number of super-scaled customers at its end. The Q4 2025 quarterly figure is n/a because the company gave only an annual figure for that year. In the third quarter of 2026, both the count (+18%) and ARPU (+22%) grew faster than the 2028 model's ranges of 4–8% and 12–16% a year.
ARPU includes Marigold's enterprise clients, who came with an acquisition in November 2025, so part of the jump is acquired. Net revenue retention, as the 10-K defines it, was 128.0% in fiscal 2025 against 113.6% in 2024: existing customers spent 28% more. The annual figure includes customers acquired with LiveIntent. A version that leaves out LiveIntent, shown in the earnings supplemental, is 120%.
A large share of revenue passes through to media. Cost of revenues (excluding depreciation and amortisation) consists mostly of fees paid to publishers and media owners. It was 39.7% of revenue in the third quarter, so Zeta keeps about 60 cents of each revenue dollar before its own operating costs.
The market, and Zeta's share of it#
The company does not state an addressable market in its 10-K. Its third-quarter 2026 earnings supplemental gives two US market sizes for 2023, citing Gartner, ISBA, Insider Intelligence and Statista:
- Marketing technology: $19 billion, growing 14% a year from 2023 to 2025.
- Data-driven ("intelligence powered") media: $64 billion, growing 12% a year.
On those figures, the markets Zeta sells into grow at roughly 12–14% a year. Zeta's organic growth of 27% guided for 2026 is therefore about twice the market's, which means it is taking share.
The comparison flatters Zeta less than it seems:
- Zeta's revenue includes media spend that passes through it. Part of its growth is clients routing more of their media budgets through Zeta, rather than buying more software.
- The market figures are the company's selection, from 2023, and they are not audited.
Management's medium-term plan, "Zeta 2028", targets revenue of at least $2.3 billion, adjusted EBITDA of at least $573 million and free cash flow of at least $371 million in 2028 (fourth-quarter 2025 earnings release). The CFO described the revenue target on the February 24, 2026 call as a 23% compound annual rate including Marigold. Revenue of $1,855 million guided for 2026 needs only about 11% a year to reach $2.3 billion in 2028, which we calculate. The revenue target now looks like a floor rather than a stretch.
What moves the results#
| Driver | Effect on revenue | Effect on margin |
|---|---|---|
| New super-scaled customers (pilots that convert) | Adds revenue in steps: $87.2 million of the third quarter's $155.5 million increase came from new customers | Sales cost is spent up front, so early quarters carry lower margin |
| Spend of existing customers (more channels, more use cases) | $68.3 million of the third quarter's increase | Highest margin, because no new sales cost |
| Direct versus integrated mix | Neutral to revenue | Integrated (agency social) carries a larger media cost; mix fell from 75% to 71% direct in a year |
| Marigold (acquired November 2025) | About $48 million a quarter, slowly shrinking | Restructuring savings so far; margin not disclosed |
| Political candidate spending | $44.4 million in 2024 and about $20 million guided for 2026; close to nothing in odd years | High-margin, lumpy; makes year-on-year comparisons noisy |
| Stock-based compensation | None | $198.9 million in the twelve months to September 2026, 11.5% of revenue, the largest gap between adjusted and GAAP profit |
The data asset#
Zeta's case for its moat is the file of identities under the platform. The FY2025 10-K says the data set "contains more than 245 million individuals in the U.S. and more than 535 million individuals globally" with "an average of more than 2,500 attributes per individual". It says the principal way Zeta collects data is "directly from the consumers when they register or interact with our platform (such as the DISQUS commenting system and LiveIntent inbox advertising), or with partners' services".
The third-quarter supplemental estimates where the identities come from:
- LiveIntent, an email-advertising business bought in October 2024: 20–25% of the graph.
- Disqus, a comment tool used by publishers: 15–20%.
- Zeta's own ad-buying platform: 15–20%.
- Zeta's email-sending infrastructure: 10–15%.
- Third parties such as LiveRamp: 10–15%.
The description has changed over time:
| 10-K for fiscal year | How the US data set is described |
|---|---|
| 2023 (filed February 2024) | "more than 240 million opted-in individuals in the U.S." in "the industry's largest opted-in data set" |
| 2024 (filed February 2025) | "more than 245 million individuals in the U.S."; the words "opted-in" removed |
| 2025 (filed February 2026) | "more than 245 million individuals in the U.S."; same wording |
The company now separates two kinds of permission. Its third-quarter 2026 supplemental shows "Zeta Data Cloud Counts as of November 2024":
- 245 million US individuals "Providing Permission to Online Tracking by Agreeing to Publisher Terms of Service".
- 110 million US individuals "Providing Permission to Email via Opt-in".
So less than half of the US file has opted in to email. The rest is covered by publishers' terms of service, which permit online tracking. That is the factual core of the dispute described in the next section. US privacy law today is mostly an opt-out regime, and the 10-K says Zeta expects "continued emphasis on an opt-out model". Under that regime the distinction matters less for what Zeta may do with the data than for what it told investors.
Why customers pick Zeta#
The 10-K names no competitor and calls the field "highly fragmented". It says most competitors focus on one use case or one type of data. What Zeta offers is breadth:
- One vendor for acquisition, retention and data.
- A ready-made identity file, so a brand can recognise people it does not yet know.
- AI tools sold on top.
Customers seem to value this. Forrester named Zeta a Leader in its Email Marketing Service Providers Wave in the first quarter of 2026 (as the supplemental reports it). Long-tenured customers keep raising their spend, as the tenure figures above show.
The switching cost is real. A brand that runs its email, its customer database and part of its media through Zeta, joined to Zeta's identity file, would have to rebuild all three to leave. The moat is narrower than "infrastructure", though. The data comes largely from owned publisher tools and an ad exchange. Others could assemble a file like it, but not quickly.
Honest weaknesses#
- Profit after stock-based pay is thin. Free cash flow as the company defines it was $237.0 million in the twelve months to September 2026. After stock-based compensation of $198.9 million, $38.2 million was left, which we calculate. The business has only just begun to earn a GAAP profit: $39.7 million over the same twelve months.
- Usage-based revenue with short contracts. A client can cut spend within a quarter. The beat-and-raise record has not yet been tested by a recession.
- Concentration. One client provides more than 10% of revenue, and the top ten more than a third.
- Marigold is shrinking. Its revenue fell from $55.6 million in the first quarter of 2026 to $48.0 million in the third.
- Control. Class B shares carry ten votes each. Chief executive David Steinberg and his affiliates held 52.1% of the voting power at the end of 2025 (10-K).
Risks the filings name, in plain words#
- Privacy rules and consumer opt-outs. California's Delete Act lets residents ask every registered data broker to delete their data through one state portal. It became enforceable in August 2026, and the 10-K says its impact "will not be fully known until late 2026". Browser-level opt-out signals could raise opt-out rates sharply.
- Data suppliers. If partners stopped supplying data, or Zeta had to stop using a source, its products could be "materially adversely impacted".
- Email gatekeepers. Inbox providers and blocklists decide whether Zeta's messages arrive.
- Third-party AI models. Athena relies on outside models, "including under our partnership with OpenAI". Zeta "cannot control the availability or pricing" of those models.
- Litigation. The securities class action and a stayed derivative suit (see below).
- Debt. A $250 million term loan under a $1.0 billion secured credit facility signed on July 24, 2026, maturing in 2031, secured on substantially all assets. The effective rate was 5.6% for the first nine months of 2026.
What the market is debating#
| Claim | Who made it, and when | What the filings and company documents show | What remains open |
|---|---|---|---|
| Zeta's data set was not "opted-in" as the company described it | Culper Research, short report of November 13, 2024; plaintiffs in the securities class action | The FY2023 10-K said "240 million opted-in individuals"; from the FY2024 10-K the word is gone; the company's November 2024 counts give 245 million with digital permission through publisher terms and 110 million with email opt-in | Whether the earlier wording misled investors is for the court; the July 8, 2026 ruling let the case proceed to discovery without deciding the facts |
| Revenue was flattered by two-way contracts with data suppliers, and much of the data came from consumer websites that do not make clear to visitors how their data will be used | Culper Research, November 13, 2024 | The company answered the same day that revenue from customers who are also vendors is "insignificant" and that the two publishers the report named gave less than 3% of revenue in 2024 to September and less than 1% of data assets; Deloitte's opinions on fiscal 2024 and 2025 were issued and no restatement appears since | Open: no court, regulator or restatement has decided it |
| The chief executive sold shares through trusts and companies to avoid reporting | Plaintiffs' amended complaint in the securities class action, May 2025, as PPC Land reported on July 13, 2026 | No filing by the company addresses it; no separate Section 16(b) suit appears in the filings | Open: an allegation in a complaint, not decided |
The consent wording. The securities class action, In re Zeta Global Holdings Corp. Securities Litigation, was filed in the Southern District of New York on November 22, 2024, a week after the short report. It names the company, the chief executive and the CFO. The company's 10-Q for the quarter ended June 30, 2026 states that the court "denied the motion to dismiss" on July 8, 2026, that "the case will now proceed to discovery", and that the company "is not able to estimate a range of possible loss". A denied motion to dismiss means the court accepted, for now, that the complaint states a claim. It is not a finding that the statements were false.
The documents settle one narrower point. The wording changed, and the company's own counts show that 110 million, not 240 million, US individuals had opted in to email. Verdict: partly supported. The word "opted-in" no longer describes the whole file, and the company dropped it. Whether that misled investors is open.
Two-way contracts and the sources of the data. Culper Research argued that some data suppliers were also customers, which could flatter revenue, and that a large part of the data came from consumer websites that do not make clear to visitors how their data will be used. In its statement of November 13, 2024, the company called the report "riddled with misrepresentations". It said it does not operate "so-called 'consent farms'". It also said the two publishers the report focused on, Apptness and ArcaMax, gave less than 3% of revenue in 2024 to September and less than 1% of its data assets. These are claims of wrongdoing, and no court, regulator or restatement has decided them. Fiscal 2024 and 2025 have since been audited, and the filings disclose no restatement. Verdict: open.
Hidden share sales. This claim comes from the plaintiffs' complaint and has not been decided. The Forms 4 the company's insiders file are the record the filings provide (see the earnings tab). Verdict: open.
What we do not know#
- How much revenue depends on identities covered only by publisher terms of service, rather than on email opt-in or clients' own first-party data.
- What the $70.0 million of "loan advances" made in the third quarter of 2026 are: to whom, on what terms and for what purpose. The 10-Q says only that their interest rate approximates a market rate.
- Marigold's profit margin, and why its revenue fell by 14% between the first and second quarters of 2026.
- Whether Palantir and OpenAI will bring revenue, how much, and when: no filing quantifies either.
- What the class action could cost, and whether a settlement would change how the data may be described or used.