Zeta Global published second-quarter results on August 4, 2026 that included a disclosure the company had not made before: a first attempt at quantifying how much of its revenue comes from customers who have actually adopted its artificial intelligence tools. The answer, according to chief financial officer Chris Greiner, is that roughly one fifth of the customer base accounts for about 70% of revenue.

The number arrived inside an otherwise conventional beat-and-raise quarter. Revenue reached $442.8 million for the three months ended June 30, 2026, an increase of 44% against the $308.4 million recorded a year earlier, according to the company's condensed unaudited consolidated statements of operations. That figure exceeded the midpoint of Zeta's own guidance by $23 million, or 5%. Adjusted EBITDA came in at $91.7 million. Net income, under generally accepted accounting principles, was $8.2 million, compared with a loss of $12.8 million in the same quarter of 2025.

Those are the headline figures, and they follow a well-established pattern. What did not follow the pattern was the decision to put adoption metrics on the record. For an industry in which nearly every vendor now describes itself in terms of artificial intelligence, the absence of adoption and monetization data has been conspicuous. Zeta's attempt to supply some is the part of the quarter most likely to matter beyond the company's own shareholder base.

The adoption framework, in detail

Greiner introduced the framework midway through the earnings call, framing it as a response to investor requests. "We're introducing an initial framework to measure adoption and monetization of Zeta's AI, one underpinned by revenue. because moats are ultimately proven by how long customers stay and how much more they spend over time," he said, according to the transcript.

The framework has four components, each expressed as a ratio rather than an absolute figure.

The first concerns the whole customer base. Across all customers, including pilots, proof-of-concept engagements and scaled accounts, the 20% who have what Greiner described as comprehensively adopted the company's AI tools account for roughly 70% of revenue, according to the transcript. Comprehensive adoption, in Zeta's usage, covers audience creation, activation and related functions rather than occasional or single-purpose use.

The second narrows the lens to the largest accounts. Among super-scaled customers, defined by the company as those generating at least $1 million of revenue on a trailing twelve-month basis, 50% have comprehensively adopted the AI tools, and that half drives 75% of super-scaled customer revenue.

The third is a growth comparison. Zeta described the comprehensive adopters as AI super users and said they grew four times faster than the 80% of customers still early in their adoption path.

The fourth links adoption to retention economics. Customers who have most comprehensively adopted the AI tools carry a year-to-date net revenue retention rate 400 basis points above the company-wide figure, and more than 20 percentage points above customers still ramping their adoption, according to the transcript. Zeta calculates net revenue retention by dividing current-year revenue from customers that also generated revenue in the prior year by the prior-year revenue from those same customers, excluding political and advocacy accounts because of their biennial pattern. The company reported a full-year 2025 net revenue retention rate of 120%, up from 114% in 2024, 111% in 2023 and 112% in 2022, according to the supplemental earnings presentation. That 2025 figure excludes LiveIntent.

A separate disclosure sits alongside the adoption ratios. Super-scaled customer relationships now average 56 months, up from 48 months a couple of years ago, based on data going back to 2018, according to Greiner. An eight-month extension in average tenure across the highest-value cohort is a material change in lifetime value assumptions, though the company did not publish the underlying distribution.

None of these ratios are audited, and none appear in the press release financial statements. They are management disclosures, presented on a call, with definitions supplied by the company. That is worth stating plainly, because the framework's usefulness to anyone comparing vendors depends entirely on whether other vendors adopt comparable definitions. At present, they have not.

Where the beat came from

Zeta guided to second-quarter revenue of $419 million to $422 million when it reported first-quarter results on April 30, 2026. The actual figure was $442.8 million. Greiner attributed the gap to faster growth in both super-scaled customer count and average revenue per user, each running ahead of the rates embedded in the company's medium-term Zeta 2028 plan.

Super-scaled customer count reached 197 at June 30, 2026, up from 168 a year earlier, a gain of 29 accounts and 17% growth. The 2028 plan models compound annual growth of 4% to 8% for that metric. Sequentially, the count rose from 189 at March 31, 2026, marking a seventh consecutive quarter of sequential expansion, according to the supplemental presentation.

Super-scaled customer average revenue per user reached $1.8 million, also up 17% year over year, against a 2028 modelled range of 12% to 16%. Greiner said customer gains were strongest in consumer and retail, telecom and healthcare.

Excluding the contribution from Marigold's enterprise business, second-quarter revenue growth was 28%. Marigold contributed $48.2 million in the quarter, down from $55.6 million in the first quarter, according to the reconciliation table in the press release. Stripping that out leaves $394.6 million of revenue against $308.4 million a year earlier. Political candidate revenue was described as not material in the quarter.

The 28% organic figure is the more useful number for anyone assessing underlying demand, and it is the one Zeta itself used when claiming what it calls the rule of 49, defined as revenue growth plus adjusted EBITDA margin summing to 49 or more. On total revenue growth, the same arithmetic produces 64.7, which the company rendered as the rule of 64.

Athena usage, voice and internal code generation

Zeta launched Athena by Zeta into general availability on March 24, 2026, opening the conversational agent to all customers of the Zeta Marketing Platform after early-access testing that began at the company's Zeta Live conference in October 2025.

By the time of the second-quarter call, Greiner said the product had been generally available for 130 days, and that more than 40% of super-scaled customers were monthly active users. Those customers had collectively generated thousands of campaigns through the agent.

Two usage patterns drew specific comment. Chief executive David A. Steinberg said that over the preceding 60 days, users interacting with Athena by voice were engaging with the platform at a level 500% higher than non-Athena users. Greiner supplied a related figure from a different angle: 83% of customer interactions with Athena are now spoken.

That second number is the more consequential of the two, because it describes interface behaviour rather than engagement intensity. If the majority of interactions with an enterprise marketing platform are voice-driven, the design assumptions behind dashboards, saved reports and permissioning models change. Greiner tied the observation to the company's revenue thesis, saying it reinforced the view that natural language will become the primary interface for marketing and business intelligence work and carries a propensity to drive higher platform utilisation.

Zeta also disclosed how far it has taken code automation internally. Steinberg said 89.6% of new code generated in the second quarter was automated, a figure he rounded to 90% in prepared remarks, up from 75% in the first quarter of 2026. The work runs through an internal workflow tool the company calls Spade.

Asked about the effect on development speed, Steinberg said cycles that would once have taken years now take months, and that new products inside the business intelligence line can be assembled in hours. He also noted that Zeta added engineers during the quarter rather than reducing headcount, while adjusted EBITDA margin expanded.

On cost, Steinberg said total internal AI spending remains well under 1% of revenue. Greiner added that unit costs for AI usage have fallen almost 40% even as usage rose across the company. Steinberg attributed that decline to negotiated terms, and separately described an enterprise agreement with OpenAI under which all Zeta global employees have access to OpenAI products, with token consumption above a threshold routed to a manager for approval.

Asked directly whether Athena token costs would eventually flow into cost of revenue, Steinberg said the OpenAI arrangement for that product is licence-based rather than token-metered, giving the company visibility into its cost base.

The ARPU question, and what tenure explains

One analyst pressed on an apparent contradiction. Super-scaled ARPU has moved within a narrow band, between roughly $1.6 million and $1.8 million, across seven quarters, even as the company describes rapid expansion inside its largest accounts.

Greiner's answer pointed to composition. New accounts that cross the $1 million trailing-twelve-month threshold enter the super-scaled cohort at revenue levels far below mature accounts, pulling the average down as the cohort grows.

The supplemental presentation supplies the underlying distribution for full-year 2025. It is measured across scaled customers, a wider group than the super-scaled cohort, defined as accounts generating at least $100,000 on a trailing-twelve-month basis. Accounts are sorted into four bands by how long they have been on the platform, and each band shows how many accounts sit in it, how much revenue they generated in total, and what that works out to per account.

Time on the platformAccountsRevenue, 2025Per account, 2025Per account, 2024
Less than 1 year111$79 million$0.7 million$0.9 million
1 to 3 years217$241 million$1.1 million$1.2 million
3 to 5 years65$135 million$2.1 million$2.0 million
More than 5 years209$815 million$3.9 million$2.8 million

The last two columns carry the argument. An account in its first year is worth roughly $0.7 million a year. The same account, once past five years, is worth roughly $3.9 million, and that mature figure rose from $2.8 million the year before. Newer accounts, meanwhile, are worth less than their equivalents were in 2024, at $0.7 million against $0.9 million.

That combination is what holds the headline ARPU flat. Each new account crossing the $1 million super-scaled threshold enters near the bottom of the curve and dilutes the average, while the accounts already at the top keep expanding. The blended number moves sideways; the two ends move apart.

Concentration follows tenure. The 209 accounts of more than five years generated $815 million, roughly two thirds of the $1,270 million total across all scaled customers, and the presentation states that around 90% of revenue comes from scaled customers who have been with Zeta more than a year. The band of three to five years is conspicuously thin at 65 accounts, against 217 in the band below it and 209 in the band above. The presentation does not explain the gap.

Greiner used slightly different cut-offs when describing the same slide on the call, putting accounts under twelve months at $700,000 and accounts of four or more years at approaching $4 million. The cohort curve, rather than any single ARPU print, is the mechanism the company is asking investors to track.

Pipeline, seller productivity and headcount

Zeta disclosed several pipeline metrics that had not previously been broken out at this level of specificity.

Total sales pipeline at June 30, 2026 was up more than 60% year over year, and up more than $100 million against the position 90 days earlier, according to Greiner. Pipeline creation on a per-seller basis rose more than 100% year over year. Average contract value on deals won in the quarter increased more than 40% against the prior year, and average deal size within the pipeline rose more than 25%.

Against that, quota-carrying headcount reached 198, an increase of one seller sequentially and 11% year over year. Greiner described the gap between pipeline growth and headcount growth as a measure of sales productivity. The 2028 plan models quota carrier growth of 10% to 15% compounded, so the current figure sits within the modelled band while pipeline runs well ahead of it.

Greiner credited two internal programmes. The One Zeta selling initiative, launched roughly fifteen months earlier, is driving sellers to attach more channels and more use cases to individual deals. The Marigold integration, closed roughly six months before the call, is now producing cross-sell activity, particularly around loyalty products sold alongside Zeta's acquire and grow use cases.

Steinberg added a third factor: assigning sellers to specific verticals. Zeta operates across 15 industry verticals, and Steinberg described vertical specialisation as an unlock for per-seller productivity.

The attachment metrics support the claim. Customers using more than one use case rose 90% year over year. Customers using five or more channels rose more than 50%. Cross-sell and upsell deals won within the quarter rose 43%. Revenue grew at double-digit rates across email, CTV and social, and across all three marketing use cases of acquire, grow and retain.

Agency mix, direct revenue and the cost of growth

Direct platform revenue mix fell to 72% in the second quarter, from 75% in both the first quarter of 2026 and the second quarter of 2025, according to the supplemental presentation. Direct platform revenue covers subscription software and utilisation fees generated through channels Zeta owns and operates. Integrated platform revenue is generated through the platform's integrations with third parties.

Greiner attributed the shift to new sales and expansions with agencies, and said that a further large holding company signing had closed between the end of the quarter and the earnings call. Newer agency relationships tend to begin with social channels, he said, and social grew rapidly again in the quarter, lifting the integrated share of the mix.

That mix carries a direct cost consequence. GAAP cost of revenue, which excludes depreciation and amortisation, reached 40.9% of revenue, 10 basis points better sequentially but 300 basis points higher than the 37.9% recorded a year earlier. In dollar terms, cost of revenues rose to $181.0 million from $117.0 million.

This is the same dynamic PPC Land documented after the first quarter, when agency growth lifted Zeta's cost of revenue by 190 basis points and compressed adjusted EBITDA margin by 100 basis points year over year. The second quarter shows the same mechanism operating with a different result at the margin line, because operating expense discipline more than offset the gross cost pressure.

Steinberg put the channel split at roughly 20% of revenue from agencies and roughly 80% from direct enterprise relationships, and said that ratio has not shifted meaningfully.

The 2028 plan targets a direct mix range of 70% to 75%, so the current 72% sits inside the modelled band rather than below it. Actual performance between 2021 and 2025 ranged from 70% to 77%.

Operating efficiency and the margin arithmetic

Adjusted EBITDA of $91.7 million represented growth of 56% year over year and a margin of 20.7%, against 19.1% a year earlier. The company also reported the figure came in $5 million ahead of the guidance midpoint.

The margin expansion is described inconsistently across Zeta's own documents, which is worth noting. The press release and the earnings call both cite 170 basis points of year-over-year expansion. The quarterly cadence table in the supplemental presentation cites 165 basis points. Calculated directly from the reported figures, $91.697 million on $442.766 million against $58.769 million on $308.442 million, the expansion is 166 basis points. The discrepancy is a rounding artefact rather than a substantive disagreement, but it means the figure quoted depends on which document is being read.

Greiner attributed the margin gain to Marigold restructuring actions and integration savings, which improved expense-to-revenue ratios by 30 basis points in research and development, 180 basis points in general and administrative, and 250 basis points in sales and marketing.

The underlying expense lines show the pattern. Research and development rose to $42.2 million from $30.6 million, an increase of 38% against revenue growth of 44%. General and administrative rose to $75.9 million from $62.2 million. Selling and marketing rose to $104.0 million from $86.4 million. Depreciation and amortisation rose to $22.7 million from $17.4 million.

Stock-based compensation totalled $52.1 million in the quarter, against $46.5 million a year earlier. That represents 11.8% of revenue, and it is the single largest reconciling item between GAAP net income of $8.2 million and adjusted EBITDA of $91.7 million. On a non-GAAP basis, net income was $55.8 million, or $0.21 per share, against $34.6 million and $0.14 a year earlier.

Income from operations was $16.9 million, against a loss from operations of $5.1 million a year earlier. For the six months ended June 30, 2026, the company still recorded a loss from operations of $1.9 million and a net loss of $5.1 million, improved from a $34.4 million net loss in the first half of 2025.

Cash generation and the balance sheet

Net cash provided by operating activities reached $69.2 million in the quarter, up 65%. Free cash flow, defined by the company as operating cash flow less capital expenditure and website and software development costs, adjusted for exchange rate effects, reached $58.0 million, up 73%. Free cash flow margin was 13.1%, up 220 basis points, and free cash flow conversion against adjusted EBITDA was 63%, up 610 basis points.

Capital expenditure was $4.8 million and website and software development costs were $6.7 million in the quarter.

For the first half, operating cash flow was $118.9 million against $76.8 million, and free cash flow was $99.6 million against $61.8 million.

Cash and cash equivalents stood at $310.0 million at June 30, 2026, down from $319.8 million at December 31, 2025. Accounts receivable rose slightly to $327.1 million. Total assets were $1,467.8 million.

The liability side moved more. Total liabilities fell to $541.3 million from $698.9 million at year end, driven primarily by current acquisition-related liabilities falling to $24.8 million from $149.0 million as Marigold consideration was settled. Long-term borrowings were $197.5 million. Total stockholders' equity rose to $926.5 million from $804.6 million, with accumulated deficit at $1,064.9 million.

During the quarter, Zeta deployed $29.9 million to repurchase 1.6 million shares. Year to date as of July 30, 2026, repurchases totalled $74.6 million, with approximately $89.4 million remaining under the existing authorisation. Greiner said dilution in the quarter was 0.1% and that the company remains on track for a net dilution target of 3% to 4% for 2026.

Zeta also closed a new $1 billion credit facility comprising a $250 million term loan A and a $750 million revolving facility that remains undrawn.

Asked how the facility changes capital allocation, Steinberg said the company will continue to focus on the same acquisition criteria while acknowledging that smaller deals no longer move the needle at current scale. He drew a clear line against large-scale consolidation. "I believe transformative M&A transforms both companies for the worse. So we are not looking to go do one like huge transformative deal," he said. Small to mid-sized transactions adding people, data sources and products remain the stated approach.

Guidance across the second half

Zeta raised guidance at every line it publishes.

Third-quarter revenue guidance moved to a range of $469 million to $472 million, up $10 million at the midpoint from prior guidance of $461 million. That represents growth of 39% to 40%, or 23% to 24% excluding acquisitions and political candidate revenue. Third-quarter adjusted EBITDA guidance moved to $115.0 million to $116.0 million, up $2.7 million at the midpoint, implying a margin of 24.4% to 24.7%.

Full-year revenue guidance moved to $1,811 million to $1,824 million, up $33 million at the midpoint from $1,785 million. That represents growth of 39% to 40%, or 24% to 25% excluding acquisitions and political candidate revenue. Full-year adjusted EBITDA guidance moved to $404.1 million to $406.3 million, up $7.9 million at the midpoint, implying a margin of 22.1% to 22.4%.

Full-year free cash flow guidance moved to $254.8 million to $255.8 million, up $20.3 million at the midpoint from $235.0 million, representing growth of 55% and a margin of 14.0% to 14.1%.

Full-year GAAP earnings per share guidance moved to $0.09 to $0.11, from $0.02 to $0.04. That is an increase of $0.07 at the midpoint, which the company characterised as greater than 300%.

The quarterly cadence table in the supplemental presentation fills in the fourth quarter by subtraction. Implied fourth-quarter revenue is $508 million, growth of 29%, with adjusted EBITDA of $131.9 million at a margin of 25.9%. Excluding Marigold and political candidate revenue, the quarterly progression runs $341 million in the first quarter at 29% growth, $395 million in the second at 28%, $416 million guided in the third at 23%, and $453 million implied in the fourth at 20%. The full-year figure on that basis is $1,604 million, growth of 25%.

The deceleration in the organic growth line across the second half is a function of both comparison base and stated conservatism, not a change in demand commentary.

What is deliberately excluded from the numbers

Greiner was explicit that the guidance carries what he described as the company's typical cushion of 2% to 5%, built on two specific assumptions: minimal revenue from partnerships that has not already been contracted, and 2028 model growth rates for customer count and ARPU rather than the higher rates currently being achieved.

Asked what Athena usage assumptions are embedded in the organic guide, Greiner said only what the company has line of sight to. "What we built into guidance is what we signed already. So we're not leaning into anything on a go-get perspective on either AI adoption or newly signed partner agreements that are still yet to be closed in the pipeline," he said.

Political candidate revenue guidance was left unchanged at $7 million for the third quarter and $8 million for the fourth, ahead of the United States midterm cycle. Greiner said the figure was held deliberately so that the guidance raise would read as core business strength rather than election spending, and described the assumption as conservative given expected demand. Political programmes spin up late and are paid ahead, which he noted is favourable for free cash flow.

One further item sits outside the numbers entirely. Greiner said the raised full-year GAAP EPS guidance excludes a potential one-time tax benefit from the release of a valuation allowance, which he described as having a reasonable probability of occurring later in the year and as potentially representing material one-time upside.

Steinberg framed the raise in terms of flow-through. "I think it's also important to note, we raised revenue by $33 million. We raised free cash flow by $20 million. So you're seeing a disproportionate percentage of incremental revenue into this business at this point drop to the bottom line," he said.

Verticals, concentration and the 2028 targets

Eight of Zeta's top ten verticals grew more than 20% year over year on a trailing twelve-month basis, with consumer and retail, financial services, automotive and healthcare all accelerating from the first quarter, according to Greiner. Asked which two fell short, he identified advocacy as one of them, and said he expects it to clear the 20% threshold next quarter given second-half activity.

The company operates across 15 industry verticals and publishes a full-year 2025 revenue split: consumer and retail at 24%, insurance and travel and hospitality at 11% each, telecom at 10%, financial services at 8%, technology and media at 7%, and advertising and marketing, education, healthcare and automotive at 5% each. No single vertical dominates, which Steinberg cited when describing the range of Palantir accounts Zeta can approach.

Greiner also placed the quarter within a longer sequence, noting it was the 21st consecutive quarter of revenue growth above 20% excluding acquisitions and political candidate revenue.

The medium-term plan the company calls Zeta 2028 targets revenue of $2.3 billion or more, adjusted EBITDA of $573 million or more at an implied 25% margin, and free cash flow of $371 million or more at an implied 16% margin and 65% conversion. Measured from the 2026 guidance midpoint of $1,818 million, the revenue target implies a compound annual growth rate of roughly 23% across the following two years, a marked step down from the 39% guided for the current year but consistent with the organic trajectory once acquisition contribution laps.

Separately, the Zeta Marketing Platform was named a Leader in the Forrester Wave covering email marketing service providers in the first quarter of 2026, receiving the top score in the strategy category and the highest possible scores in 11 of 26 criteria, including identity resolution, data management, data governance, AI approach, vision, innovation, roadmap and regulatory compliance, according to the supplemental presentation.

Four catalysts, and what each actually delivers

Zeta attributes its repositioning to four events across 2026, three of them partnerships.

OpenAI

The collaboration was announced at CES in January 2026 and covers two distinct things. OpenAI models power Athena's voice capability. Separately, Zeta partners with OpenAI's advertising operations.

Steinberg drew a hard boundary on the data question. "OpenAI powers Athena's voice, the decisions run on our models and no large language models ever touch the data in our data cloud," he said. The decisioning systems he referred to are Zeta's own inference models, which the company says it has been training on its data cloud since 2017.

On the advertising side, Steinberg said Zeta is actively serving ads through the relationship, that it is a meaningful revenue opportunity, and that it is scaling quickly. He also confirmed it is not reflected in forward guidance. The commercial context matters here: OpenAI's advertising business has expanded rapidly through 2026, with self-serve ads manager access broadening to US advertisers in early May and the minimum spend requirement dropped, and further geographic expansion by late June.

Steinberg described a follow-up meeting scheduled for the week after the call, with 20 people assigned to scoping additional work.

Athena

The agent reached general availability on March 24, 2026, and Zeta positions it as the intelligence layer between its data assets and its activation channels rather than as an application sitting on top of the platform. Two agentic applications shipped first: Insights, a conversational analytics tool, and Advisor, a goal-based optimisation tool.

The company launched Athena for agencies at Cannes Lions in June 2026, hosting more than 115 executive meetings and delivering 37 live demonstrations, which Steinberg said produced a record sales pipeline coming out of the event.

Asked which agencies were adopting, Steinberg said the pattern surprised him. He had expected mid-sized agencies to move first. Instead, larger agencies adopted faster and at scale. He cited one case in which a large agency brought Athena into work for a large airline client, and said the resulting campaign is running at a 1,400% return on marketing spend, against the 600% to 700% figure Zeta typically cites from Forrester research.

Snowflake

Zeta said it has deepened an existing relationship with Snowflake and now has more than 100 shared customers, with the Zeta Marketing Platform embedded in those customers' Snowflake environments alongside joint go-to-market activity.

Palantir

The Palantir partnership was announced at Cannes Lions on June 23, 2026, with Zeta projecting more than $100 million in annual revenue from Palantir's eligible enterprise customers over time.

Two milestones were reported on the second-quarter call. The Zeta Data Cloud was fully integrated with Palantir Foundry as of July 31, 2026, adopting Foundry's ontology layer. And Zeta has signed multiple agreements from the initial combined sales motion.

Steinberg described the commercial mechanics without hedging. He said the two engagements signed with Palantir happened in real time, that Palantir introduced Zeta to two of the largest advertisers in the world, and that both meetings converted. He also said the pipeline he is building around Palantir accounts is being tracked separately, outside the company's reported pipeline, because including it would distort the aggregate figure at this stage.

Asked how he sequences that list, Steinberg said he started with enterprises spending $1 billion a year on consumer marketing, and that the qualifying list ran longer than the twenty accounts he had planned to begin with.

Gap, and the marketing cloud replacement cycle

The customer disclosure with the widest read-across for marketing technology buyers concerns Gap Inc.

Steinberg said Gap selected Zeta to help architect its next-generation marketing stack under a multiyear agreement, with Zeta named as Gap's system of record and Athena positioned at the centre of how customer data, decisions and execution come together. The supplemental presentation adds that forward-deployed engineers are embedded onsite.

The displacement detail is specific. Asked about request-for-proposal activity, Steinberg said Gap had been with Salesforce for some time, and that Zeta displaced three other vendors in addition to Salesforce to become the system of record.

He generalised from there, arguing that large organisations no longer want to operate four, five or six separate vendors covering software, professional services, activation, data and customer data platform functions. His technical claim is about latency: platforms that must exit their own environment through application programming interface integrations to reach AI, then reach data sources, then return, incur delay that degrades return on marketing spend.

Greiner said RFP activity is strong and is a component of the greater-than-60% pipeline growth, and that a higher RFP share is part of why average contract values in the pipeline have risen.

Whether a marketing cloud replacement cycle is genuinely underway is a claim the industry has heard before. What is different in this instance is a named displacement at a large multi-brand retailer, with a named incumbent, disclosed on an earnings call.

Who the customers are, and who is not named

Gap is the only customer Zeta names anywhere in the second-quarter materials. It appears three times: in Steinberg's prepared remarks, in his answer to a question from Truist Securities about request-for-proposal activity, and on a slide in the supplemental presentation that places Gap alongside OpenAI and Palantir under the heading of external validation, described there as a flagship customer.

Every other customer reference in the package is anonymised by category. On the call, Zeta described a major sports and entertainment company, a leading energy drink brand, a leading telecommunications provider that expanded into real-time personalisation, and a very large airline reached through a very large agency, the account Steinberg said is running at a 1,400% return on marketing spend. The presentation slides covering Zeta Business Intelligence supply seventeen further descriptors across six use cases, none of them named.

The same applies to the agency disclosure, which is the more consequential omission given that agency relationships drove the three-point decline in direct platform mix and the 300 basis point year-over-year rise in cost of revenue. Greiner said Zeta closed and expanded with another very large holding company between the end of the quarter and the earnings call. He did not identify it, and no analyst pressed for the name.

What Zeta does publish is an aggregate penetration count. As of December 31, 2025, the company reported serving 4 of the largest agency holding companies, alongside 12 of the 15 largest consumer and retail companies, 10 of the 13 largest technology and media companies, 8 of the 12 largest financial services companies, 6 of the 16 largest insurance companies, all 4 of the largest telecommunications companies, 3 of the 6 leading pharmaceutical companies, 2 of the 3 largest airlines, and 1 of the 3 largest automotive companies in the world. That composite produces the 51% of the Fortune 100 figure the company cites.

Those counts carry two caveats. The first is dating: they are measured as of December 31, 2025, seven months before the quarter being reported, and they were not refreshed in the second-quarter package even though the agency channel is where the mix shift occurred. The second is that a penetration count establishes presence, not scale. A holding company that runs a single social budget through the platform and one that has consolidated multiple brands onto it both register as one relationship.

No holding company is identified by name anywhere in the four documents. The only agency-adjacent product named is Zeta's own: Athena for agencies, launched at Cannes Lions in June 2026.

Client confidentiality is standard practice in enterprise software reporting, and the absence of names is not itself unusual. What it does constrain is verification. Analysts cannot independently assess concentration risk inside the agency channel, cannot confirm whether the same holding companies recur across the pipeline metrics Zeta disclosed, and cannot distinguish a holdco relationship measured in low single-digit millions from one measured in tens of millions. Steinberg addressed concentration only at the vertical level, noting that none of the 15 verticals Zeta operates in represents a heavy concentration.

The one customer above 10%

The earnings package says nothing about customer concentration. Zeta's SEC filings do, and what they say has changed.

Through its first years as a public company, Zeta reported that no single customer crossed the 10% threshold. The 2021 registration statement put the top ten customers at under 35% of 2020 revenue with no customer above 10%. The 2023 annual report reported approximately one third of revenue from the top ten and again no customer above 10%. The 2024 annual report reported more than one third from the top ten, with the same statement about the 10% line.

The Form 10-Q for the first quarter of 2026 reverses that. Under concentration of credit risk, the filing states that one customer accounted for more than 10% of total revenues in both the first quarter of 2026 and the first quarter of 2025, and that one customer represented more than 10% of accounts receivable at both March 31, 2026 and December 31, 2025.

The customer is not identified. Neither is the amount.

One accounting policy makes the concentration figure harder to read than it appears. Zeta states in its annual reports that it occasionally holds separate contracts and billing relationships with individual marketing agencies owned by the same holding company, and accounts for each as a separate customer. The 2023 and 2024 filings pair that with a risk factor: if a holding company chose to exert control over its individual agencies and end the relationship, the result could be a disproportionate loss of revenue. Measured at the holding company level rather than the agency level, in other words, concentration would read higher than the disclosed figures suggest.

What is publicly known about the client list

Zeta does name customers, just not in its financial reporting. Its case study library carries Hibbett, First National Bank of Omaha, Nothing Bundt Cakes, Tourism Ireland, Kia, Simon Property Group, White Castle, Chicken Salad Chick, TUI France and the agency Tombras, alongside a video case study with TKO Group Holdings, the parent of UFC, WWE, PBR, IMG and On Location, which describes consolidating email marketing across UFC and WWE on the platform in a three-month implementation.

Two further names surfaced through the Athena launch cycle rather than through Zeta's own materials. Red Roof participated in the Athena beta and has worked with Zeta for four years, according to AdExchanger's reporting at the March 2026 launch. Adweek named Red Roof, TKO and Tombras as early Athena users in its 2026 Tech Stack Awards writeup.

None of those accounts is plausibly the 10% customer. A regional hotel chain, a bakery franchisor and a sporting goods retailer do not individually represent a tenth of a company guiding to $1,818 million of revenue, which would require an account of roughly $180 million a year. Every customer Zeta discusses publicly sits well below the threshold, and the one that clears it appears nowhere in its marketing.

A minor inconsistency runs through the same material. The Q2 2026 earnings deck puts the data cloud at more than 535 million covered individuals. AdExchanger reported over 550 million in March 2026, and Adweek cited more than 545 million in August 2026. The figure varies across Zeta's own communications without an explanation of the basis for each count.

Why the names are withheld

The withholding is permitted, and largely required by nothing more than convention.

Under US accounting standards, an entity must disclose that revenue from a single external customer exceeds 10% of consolidated revenue, together with the amount and the segment reporting it. It does not have to identify the customer. Zeta discloses the existence and omits both the amount and the name, which is common practice among enterprise software companies.

The SEC's disclosure rules bite only where a registrant is substantially dependent on a customer such that losing it would materially harm the business. Zeta has never taken that position. Its risk factor is framed around agency holding companies consolidating their relationships, not around any single named account, and its guidance language emphasises breadth across 15 verticals rather than dependence on any one of them.

Beyond the regulatory floor sits commercial practice. Enterprise marketing contracts routinely carry mutual confidentiality clauses covering the existence of the relationship, and customers frequently treat their marketing stack as competitively sensitive. That is why Zeta's named references skew toward mid-market brands willing to trade a testimonial for visibility, while its largest accounts stay anonymous. Gap is the exception that proves the pattern: a named displacement, disclosed only because Gap itself was willing to be named as part of a publicised transformation programme.

The practical consequence for anyone reading the results is narrow but real. Zeta's revenue quality argument rests on retention, tenure and expansion inside its largest accounts. The concentration disclosure says one of those accounts now exceeds a tenth of revenue. Without a name, an amount, or a vertical, the disclosure establishes that the risk exists while withholding everything needed to size it.

Two other customer counts appear without names. Zeta said it now has more than 100 shared customers with Snowflake. The Marigold enterprise business it acquired brought more than 100 global enterprise brands, according to the September 2025 acquisition announcement. Whether those two sets overlap, and to what degree, is not disclosed.

Zeta Business Intelligence as a fourth use case

Zeta introduced what it calls Zeta Business Intelligence, or ZBI, as a fourth use case alongside acquire, grow and retain.

Steinberg was careful to distinguish it from established business intelligence tooling. "To be clear, this is not static dashboards or old school business intelligence used to explain what happened. ZBI helps predict what happens next and acts on it in real time," he said.

The supplemental presentation groups sample applications into six areas: real estate intelligence, customer experience, market sizing and opportunity assessment, loyalty growth, business expansion, and business measurement. Each is illustrated with anonymised customer types rather than named accounts, spanning specialty bakery franchises, tactical apparel brands, automotive services franchises, athletic footwear, home furnishings, rural lifestyle retail, luxury apparel, animal health, restaurant franchising, energy drinks, pizza franchising and wireless providers.

Two examples were described on the call in more detail. A major sports and entertainment company uses ZBI to understand its fans' share of entertainment spending across live events and streaming, which Steinberg said helps quantify the value of current distribution arrangements and strengthen negotiating positions at renewal. A leading energy drink brand uses ZBI to quantify the incremental value it drives for retail partners.

Steinberg said the sports example was assembled in hours after a meeting with the client's chief marketing officer, and that the client believes the analysis will produce millions of dollars of incremental revenue in streaming rights renegotiation. He also noted the secondary effect Zeta expects: participating in the marketing associated with those streaming rights once the deals are signed.

On go-to-market, Steinberg said the pattern has been inbound. Customers had been using ZBI for months before the public announcement, paying Zeta to develop the applications with them, and the company is now combining that pull with active selling.

Asked separately whether generative engine optimisation is winning incremental budget or simply moving existing spend, Steinberg said the product is entirely additive for Zeta because the company never sold search engine optimisation. He described application programming interface integrations into Claude, ChatGPT and Gemini running through a single interface, allowing brands to monitor and adjust visibility across those systems from one place.

He set expectations for sales cycles at two speeds. Existing clients adding ZBI as a third or fourth use case are moving at the shorter end of Zeta's normal 90 to 180 day product cycle, or faster. New clients, where the entry point is a chief information officer or chief technology officer rather than a chief marketing officer, are expected to take longer.

Asked whether ZBI changes the addressable market framing, Steinberg was direct that it does not replace the existing target. "I want to be clear, we're still looking to get to 7% to 10% of our clients' marketing wallet share. That's not changing," he said, adding that the company sees a path to a $10 billion business at a 30% operating margin on the marketing component alone.

The repositioning argument, and its limits

The supplemental presentation devotes a full section to arguing that Zeta belongs in a valuation cohort of AI infrastructure companies rather than marketing technology companies. The argument rests on three claims the company describes as independent proofs.

The first is economics: 120% net revenue retention for 2025, 78% free cash flow growth in 2025, and sustained revenue growth above 20% at scale. The second is the proprietary data asset. The third is external validation, specifically Palantir's embedding of the Zeta Data Cloud into Foundry.

The presentation then places Zeta alongside a cohort including Palantir, Datadog, CrowdStrike, MongoDB and Snowflake, using last-fiscal-year figures. Zeta's own Rule of 40, calculated as revenue growth plus adjusted EBITDA margin, was 51 for 2025, consistent with 30% growth and a 21.4% margin. The company cites a four-year revenue compound annual growth rate of 30%, net revenue retention within the cohort's 115% to 125% band, a forward free cash flow yield of 5% against a cohort average of 2%, and an enterprise value to revenue multiple of roughly 3 times against a cohort range of roughly 7 to 34 times. Market data is stated as of July 29, 2026, sourced to FactSet, with free cash flow yield derived from guidance where available and consensus estimates where not.

Steinberg's framing on the call was blunter. "Investors have historically viewed Zeta as a marketing technology company. Marketing is where our platform was first applied and where we've built our leadership position, but that description no longer fully encapsulates who we are today," he said.

The limits of the argument are visible in the same documents. Cost of revenue at 40.9% is structurally higher than software infrastructure comparables, a function of media and marketing costs paid to third-party publishers and partners on revenue-share, cost-per-lead, cost-per-click and cost-per-thousand-impression bases. Stock-based compensation at 11.8% of revenue remains the dominant gap between GAAP and adjusted results. And the company's own quarterly cadence shows organic growth decelerating to 20% by the fourth quarter.

Inside the data cloud

The asset underpinning the repositioning is the Zeta Data Cloud, which the company describes as covering more than 535 million individuals globally, resolved from trillions of consumer signals, with 20 years of accumulated governance and model tuning.

Zeta separates the asset into three types. Identities are unique individuals represented by offline personally identifiable information such as a hashed email address, joined to a digital identifier through an authentication event. Identifiers determine how to reach an identity across channels and include email hashes, mobile advertising identifiers and phone numbers. Signals are data in motion and at rest, processed to infer intent, interest and attributes.

The sourcing map is unusually detailed for a public disclosure. Zeta lists seven contributing sources: its demand-side platform, its supply-side platform, its message transfer agent, Disqus, LiveIntent, owned and operated properties including ArcaMax, and third-party ecosystem partners including LiveRamp.

The presentation states that the message transfer agent, LiveIntent and Disqus together contribute more than 75% of identities. The demand-side platform, LiveIntent and Disqus contribute more than 95% of identifiers. Disqus, the demand-side platform and LiveIntent contribute more than 75% of signals.

On permission, the presentation cites counts as of November 2024: 245 million United States individuals providing permission for online tracking by agreeing to publisher terms of service, and 110 million providing permission for email through an opt-in action. The same slide charts the US permissioned population at 231 million in December 2021, 241 million in January 2024 and 245 million in November 2024, across a period covering GDPR, CCPA, Apple's App Tracking Transparency changes, third-party cookie deprecation activity and successive versions of the Transparency and Consent Framework.

Those permission figures are dated. They were current as of November 2024 and have not been refreshed in the second-quarter materials, which is a gap worth noting for anyone assessing the durability claim in mid-2026.

Zeta's consent methodology is not an abstract question. A federal judge in New York allowed an investor lawsuit against the company to proceed on July 13, 2026, over how Zeta described the way it obtained consumer consent for its advertising data. The complaint traces back to allegations first raised in a short-seller report published in November 2024, which the company denied.

That ruling landed in a week when consent frameworks came under pressure from several directions simultaneously, including a separate challenge to the legal architecture underpinning transatlantic data transfers.

Neither the press release nor the earnings call addressed the litigation. The company's standard risk factor language references its ability to collect and use data online, and the standards inbox service providers may adopt, among factors that could affect forward-looking statements.

Why this matters for the marketing community

Three things in this reporting package have practical consequences beyond Zeta's shareholder register.

The first is the adoption framework itself. Marketing technology procurement has spent two years unable to distinguish between vendors whose AI features are used and vendors whose AI features are shipped. Zeta's ratios are self-defined and unaudited, but they establish a template: what share of customers use the tools comprehensively, what share of revenue those customers represent, and what the retention differential is between adopters and non-adopters. Buyers negotiating renewals now have a set of questions to put to other vendors, and a public benchmark against which answers can be compared.

The second is the interface data. If 83% of interactions with an enterprise marketing agent are spoken, the operational assumptions behind team structure, reporting cadence and access control change. That is a workflow question for marketing operations teams, not a procurement question.

The third is the margin mechanics of agency growth. Zeta's direct platform mix fell three percentage points to 72% and cost of revenue rose 300 basis points year over year, both driven by agency relationships that typically start with social channels. The same pattern was visible in the first quarter. Agencies buying media through a platform generate revenue with materially different unit economics than enterprises licensing software, and any vendor pursuing both motions carries that tension. For agencies evaluating platform partnerships, the disclosure indicates where the vendor's incentives sit.

There is also a competitive signal in the Gap disclosure. Enterprise marketing stack consolidation has been forecast repeatedly. A named retailer replacing an incumbent plus three additional vendors, disclosed under oath-adjacent conditions on a public earnings call, is a data point of a different order than a vendor press release. It does not establish a cycle. It does establish one instance.

Finally, the partnership structure deserves scrutiny rather than acceptance. Zeta has three relationships with companies at the centre of enterprise AI spending, and by its own account has excluded revenue from all three from its guidance. That conservatism is defensible. It also means the partnerships remain, for now, a narrative asset rather than a reported one. The next two quarters will show which they become.

Zeta will hold its sixth annual Zeta Live conference on October 8, 2026 at David Geffen Hall, Lincoln Center in New York City, where Steinberg said the company will introduce the next generation of Athena. Lindsey Vonn and Kevin Hart are confirmed as headline speakers.

Timeline

Summary

Who: Zeta Global Holdings Corp. (NYSE: ZETA), the New York-headquartered marketing and data company founded in 2007 by David A. Steinberg and John Sculley. Steinberg, co-founder, chairman and chief executive, and Chris Greiner, chief financial officer, presented the results. Trey Campbell, senior vice president of investor relations, hosted the call in his first quarter in the role. Analysts from Craig-Hallum, RBC Capital Markets, Canaccord Genuity, Citi, Morgan Stanley, William Blair, Goldman Sachs, Truist Securities, KeyBanc Capital Markets, Bank of America, ROTH Capital Partners, Needham, D.A. Davidson and B. Riley Securities participated.

What: Second-quarter 2026 results showing revenue of $442.8 million, up 44% year over year and $23 million above the guidance midpoint; adjusted EBITDA of $91.7 million at a 20.7% margin; GAAP net income of $8.2 million and earnings per share of $0.03; free cash flow of $58.0 million, up 73%. Full-year 2026 revenue guidance rose $33 million to a $1,818 million midpoint, adjusted EBITDA guidance rose $7.9 million to $405.2 million, free cash flow guidance rose $20.3 million to $255.3 million, and GAAP earnings per share guidance moved to $0.09 to $0.11 from $0.02 to $0.04. The company also introduced an initial framework quantifying AI adoption, disclosing that 20% of customers who comprehensively adopted its AI tools account for roughly 70% of revenue, and that among super-scaled customers, 50% of comprehensive adopters drive 75% of super-scaled revenue.

When: Results were released on August 4, 2026, covering the quarter ended June 30, 2026, with the investor call held the same afternoon at 4:30 p.m. Eastern Time. A Form 8-K furnishing the release was filed the same day and signed by Greiner.

Where: Zeta operates from New York City with offices worldwide, and reported across 15 industry verticals. The Palantir Foundry integration completed on July 31, 2026. Zeta Live 2026 is scheduled for October 8, 2026 at David Geffen Hall, Lincoln Center in New York City.

Why: The adoption framework matters because marketing technology buyers have had no comparable public benchmark for distinguishing shipped AI features from used ones. The margin mechanics matter because agency-led growth continues to raise cost of revenue while operating efficiency offsets it, a tension any platform pursuing both agency and enterprise motions carries. And the Gap displacement matters because it puts a named incumbent, a named replacement and a multiyear system-of-record agreement on the public record at a point when enterprise marketing stack consolidation is widely forecast but thinly evidenced.