NICE vs Five9 AI Revenue: $392M ARR, $45M a Quarter, and Seats Still Growing

Five9's AI revenue grew 91% in Q3 2026 and NICE's AI ARR grew 46% to $392 million, and both companies say human agent seats are still growing. On the evidence either vendor publishes, AI is not yet replacing lost seat revenue, because no lost seat revenue has been reported. The comparison of NICE and Five9 AI revenue turns on a narrower question: how much of each number is new money?

The short answer

Is AI revenue replacing lost seat revenue at NICE and Five9? Not yet, because neither reports any lost seat revenue. AI is adding to a seat base that still grows. At Five9 most of that addition looks new. At NICE an acquisition and a broader self-service label sit inside the number.

NICE and Five9 AI revenue, side by side in Q3 2026

Both companies reported within 24 hours of each other, and the headline AI figures are not measured the same way. NiCE, as the company now styles itself, released on 4 November 2026 and Five9 on 5 November. The table below sets out each number with its definition, because the definitions decide what the comparison can support.

Compiled from each company's Q3 2026 release and call. Last verified 7 November 2026. NICE's 12% prior-year share comes from dividing $268 million by annualised Q3 2025 cloud revenue, and it matches what the company said at the time.

Read the growth row with suspicion and the guidance row with care. Five9 raised its AI growth outlook in the same week that NICE cut its cloud growth outlook, and the two moves are easy to read as one vendor winning. The base sizes argue against that reading. NICE's AI ARR is about 2.2 times Five9's annualised AI revenue of about $180 million, though annualising a ramping quarter is a rough shortcut.

The denominators need the same care. Five9's subscription line excludes telecom and professional services, while NICE's cloud line is everything hosted. Against Five9's total revenue of $323.3 million, the $45 million is 13.9%, not 17%. Our view: the closer honest statement is that both vendors sit in the mid-teens, and the labels overstate the difference.

NICE AI revenue: what the AI ARR figure actually contains

NICE's $392 million is the same data series the company once labelled AI and self-service. No definition of self-service has been published in the documents we read.

The label moved, the series did not

On the Q3 2025 call, CFO Beth Gaspich gave CX AI and self-service ARR of $268 million. On the Q2 2026 call she used the same label for $362 million. On the Q3 2026 call she said CX AI ARR, $392 million, up 46%. Dividing $392 million by 1.46 gives $268.5 million, so the year-ago base is the self-service-inclusive figure.

This is not a new rename to hide something. NICE's press release was already using the shorthand AI ARR in November 2025. No speaker explained the change in the transcripts we read, and we do not assert the content of the self-service portion. We do assert that a reader of the 2026 release alone would not know it was ever in the description.

Production only is a real discipline

NICE's scope statement deserves credit. Scott Russell said on the Q3 2026 call: "Our AI revenue is production only." He added that it excludes services, pilots and proofs of concept. That is a stricter rule than many AI disclosures, a pattern we have compared in the ranking of AI revenue disclosure metrics. It also explains why bookings lead revenue here, since Gaspich said the ramp "was a bit slower than we originally anticipated".

Cognigy sits inside the growth rate

NICE closed the Cognigy purchase in September 2025, and it has refused to split the AI number. When Piper Sandler's Ryan Abbott suggested on the Q2 2026 call that AI grew "sub-30" without Cognigy, Gaspich replied that "you can't really look at organic versus inorganic". Her reason was that NICE stopped selling its comparable products after the deal.

That answer is fair and revealing at once. If Cognigy replaced older NICE products, some Cognigy growth is customers moving between SKUs inside one metric. A sensitivity shows the range. The 49% and 43% figures from Q3 2025 imply about $11 million of Cognigy in the $268 million base. Gaspich cited $85 million as Cognigy's exit ARR for December 2026. If all of that were already in the $392 million, the remainder grew about 19%. Treat 19% as a floor from an assumption, not a disclosed figure.

Five9 earnings: what AI revenue includes after the scope change

Five9's AI figure is recognised revenue from named products, which makes it the cleaner of the two metrics, although its scope also changed recently. CFO Bryan Lee told the Q1 2026 call that "this AI revenue figure now includes both enterprise and commercial" and that the disclosure "is different from past periods". The transcripts we read do not fully restate how older quarters were built.

On the Q3 2026 call Lee named AI agents, Agent Assist and AI Insights as the three largest contributors to the 91%, and said AI agents were the biggest ramp. Agent Assist and AI Insights help human agents rather than replace them, so they cannot reduce seat counts directly. That split is not published, so we cannot say how much of the $45 million competes with a seat.

Lee also warned about durability. He attributed part of the jump to several customers deploying "earlier and at a larger scale", which benefits timing. He said those ramps "will create difficult year-over-year growth comparisons for 2027". So the 68%, 78% and 91% sequence should not be extended in a straight line.

Level AI is not in any of these figures

Five9 also agreed to pay about $325 million in cash for Level AI, whose products include AI agents, quality management and coaching. Lee gave about $20 million of ARR, about 50% growth and a multiple of about 11 times next 12 months ARR. Guidance excludes it. From the first quarter it closes, Five9's AI number will carry an acquired piece, which is the same comparability problem NICE has now. It also adds AI quality assurance for contact center customers on other platforms, which is a bet that AI revenue can exist without a Five9 seat underneath it.

Contact center AI seats: what each management said, and what it left out

Both vendors say seats are growing, and neither publishes the number that would let you check. Seat statements appear only on earnings calls, in qualitative language, and in neither company's release. The wider seat compression debate in SaaS pricing has more data points than this category does.

Five9's fixed commitments would delay the signal

Mathradas said on the Q1 2026 call that Five9 has "started to transition" new logos and renewing customers to a committed revenue model. A customer committing to a revenue number can shift spend from seats to AI without the total moving. We think this is sensible commercially and awkward analytically. Under that model, seat compression could run for years before it reaches the revenue line, so revenue growth alone cannot rule it out. It is the seat count that would show it first.

Five9's own legal language is more cautious than its call. Its Q3 2026 release warns that AI "will likely perform an increasing proportion of contact center interactions", and that results suffer if it cannot "replace decreases in subscription revenue from licenses" with AI sales. That sentence is what our review of SaaS risk factor disclosures on AI agents flagged as a rare admission tied to a vendor's own billing base.

NICE is pricing for the efficiency it fears

Russell was unusually candid on the Q3 2026 call. He said: "seat-based revenue is fantastic because it's predictable, but the more efficient the agent is, we don't necessarily get to monetize that volume". That is the seat problem in one sentence, from the vendor exposed to it. NICE's response is outcome pricing tied to containment or resolution, a model with its own meter definition problems.

Both CEOs also offered a volume argument. Russell said personal AI agents are "not just displacing work, it's increasing work", citing a Gartner consumer survey. Mathradas said AI "is going to drive more interactions for this business, not less". We would treat both as hypotheses. More interactions help the vendor only if they are paid for and only if the customer does not cap them.

The new dollars test: replacing, relabeling or adding

AI supplied a little over half of Five9's added subscription dollars and roughly half of NICE's added cloud run rate, which points to addition rather than replacement. This calculation is ours, not either company's, and it uses rounded inputs, so we show the working in full.

For Five9, Lee said subscription was about 83% of $323.3 million, or about $268 million. Remove $45 million of AI and about $223 million remains. Grow that back at the disclosed rates and the year-ago base is about $207 million of CCaaS plus about $24 million of AI. The check works: $24 million of $230 million is the 10% Lee quoted. So AI added about $21 million and CCaaS about $17 million, or 56% from AI.

For NICE, AI ARR added $124 million, from $268 million to $392 million. Annualised cloud revenue rose $239 million as reported, from $2,252 million to $2,491 million. Adding back the $9 million credit against cloud lifts that rise to $275 million. So AI is 45% to 52% of the addition. The rest of cloud, on the same basis, grew about 6% to 8%.

Three readings follow, and they are not equally supported. Replacement has no support yet, since Five9's CCaaS line grew 8%, up from 7% in Q2, and NICE's non-AI cloud also rose. Pure relabeling is not supported at Five9, where the base and the AI line both grew. Partial relabeling is plausible at NICE, given the self-service history and the discontinued products Gaspich described.

Our position: Five9's AI number is mostly addition today, with a fair chance of becoming substitution under fixed commitments. NICE's is real and large, but less of it is proven to be new money. The guidance cut supports that caution, even though NICE tied it to enterprise ramps and standalone Workforce Management rather than to AI substitution.

AI dollars cost more to deliver than seat dollars, for now

Both vendors reported lower gross margin alongside rising AI share. A dollar moving from seat to AI is not yet worth the same to the vendor. NICE's adjusted gross margin fell to 66.5% from 69.9%, or about 67% before the credit. Five9's fell to 61.7% from 62.8%. These moves have different stated causes, and it would be wrong to assign both to AI.

Gaspich attributed NICE's decline mainly to hosting and infrastructure for large-enterprise and Cognigy deployments, incurred before revenue fully ramps, plus partner costs. She expects stabilisation as deployments scale. Lee attributed Five9's to timing-related costs with a large partner, which he expects to start reversing in Q4. Only NICE names AI deployment as a cause, and only by implication.

NICE also announced a reduction of about 5% of its workforce, with charges of $15 million to $20 million. Non-GAAP operating margin was 27.0%, down from 31.5%. The behaviour worth noting is the sequence: margin pressure, then a slower ramp, then a cost cut. That does not prove AI is structurally lower margin. It does mean the claim that displaced seat dollars arrive as equally profitable software has not been shown by either vendor.

Mathradas's broader claim is that the market for CCaaS plus support AI is "nearly 2x the displacement of seats". We could not check that figure against a public method, and we would not plan on it. If you sell or buy in this category, ask what the model assumes about how far support staffing actually falls after deflection.

Where this comparison is weakest

The strongest objection is that none of this can detect seat loss early, so a clean result may simply mean it is too soon. Contact center contracts run for years, and Five9 says its deals often span long terms. A vendor could lose future seats while reporting growth today. Our conclusion is about the reported period, and it should not be read as a claim about 2028.

Seat statements cannot be audited

Every seat claim here is management commentary, some of it answered under analyst questioning and none of it in a filing. "Relatively in line" is not a number. CXone seats are also not all of NICE's cloud base, so a rising CXone count could coexist with shrinkage elsewhere. NICE's standalone hosted Workforce Management is winding down over time, on Gaspich's own description.

Our arithmetic mixes unlike things

Comparing NICE's ARR, a period-end figure, with annualised recognised revenue is imperfect. ARR can lead revenue, and Cognigy's contribution lands differently in each. Five9's split relies on a rounded 83% and a rounded $45 million. A one point change in the subscription share moves the AI share of new dollars noticeably. Treat 56% and 45% to 52% as directional, and do not quote them as reported figures.

The competing view has real force

A bull on these vendors would say our relabeling concern is overdone. Moving customers from older self-service tools to Cognigy still earns higher-value revenue, and that is a genuine upgrade. We accept that, but it is a different claim from net new AI demand. A bear would say we are too generous, and that 8% CCaaS growth for a company at this scale already reflects pressure. We cannot rule out that view either, because neither vendor discloses price per seat.

What contact center buyers should ask at renewal

You should negotiate as if seat compression is your option to exercise, because the vendors are already writing contracts around it. If AI handles more of your interactions, you need the savings to reach your budget and not vanish inside a commitment.

Run the same new dollars test on your own account. Split last year's spend into seats, AI and telecom, then compare it to this year's. If your AI line rose and your seat line did not fall, you bought an addition, whatever the business case said. Our comparison of human and agent ticket costs gives a way to price the resolved interaction before you sign.

Frequently asked questions

Where to start before your next CCaaS renewal

This week, pull your last 4 invoices and split them into seats, AI and telecom. If your vendor cannot itemise AI on the bill, you have found your first negotiating point, and it costs nothing to raise.

Then write to your account manager with one question: what is my concurrent seat count trend, and what happens to my commitment if it falls 20%? The answer will show you more about your contract than either vendor's AI growth rate. Revisit this comparison when both report Q4 2026, because that is the first NICE quarter with Cognigy fully in the base, and the point at which Five9 says it will discuss Level AI's financial impact.

Related reading