Twelve platforms across three markets, priced on their own published rates as of 26 August 2026. Capability at the top of this market has converged, so routing, recording, forecasting and basic self-service are no longer selection criteria at enterprise scale. What still separates these vendors is who controls the AI meter, what one AI-handled interaction actually costs on published list rates, and what it costs to leave.
Anyone given these ten vendors as one comparable list has a category error in front of them before the evaluation starts. Verint and Calabrio are workforce specialists that never sold a routing engine, and since November 2025 they are one company under one owner. NICE and Genesys are suites that bundle workforce management to defend a seat. Five9 and Talkdesk are contact centre platforms without workforce heritage. RingCentral arrives from business telephony. Salesforce and Microsoft attach a contact centre to a seat you already pay for. Naming that is not a preliminary to the answer. It is part of the answer, and it is why the matrix in this report is built to be read across a row rather than down a column.
The consequence is specific rather than taxonomic. A scoring matrix that gives every vendor a workforce management score will rank Verint above Five9 and then be unable to explain why Five9 won the deal, because the deal was decided on global voice coverage Verint does not sell. A matrix that scores routing will rank Genesys above Calabrio and be unable to explain why the operation with a 41 percent attrition problem bought Calabrio, because that operation was not buying routing. The fix is not a better weighting scheme. The fix is deciding which of the three purchases you are actually making and running the other two as separate evaluations.
Three names the usual shortlist omits belong in the comparison because they take money from the same budget line. Amazon Connect is not a niche developer product; it is the telephony and routing substrate underneath a large number of enterprise contact centres, and it prices as meters with no seat fee at all. Google's Customer Engagement Suite competes for the AI self-service budget at roughly an order of magnitude below the suites. And the Teams-attached route that most Microsoft-standardised organisations actually end up on is a different purchase from Dynamics 365 Contact Center, with a different support and integration position. Leaving all three out is how a suite quote goes unchallenged.
The test applied to every claimed difference in the report: a capability both vendors have is not a differentiator, and a capability only one has is not a differentiator either unless a buyer's outcome changes because of it. Roughly two thirds of what vendors market as differentiation fails that test and is not in the document. What survives is commercial structure, telephony ownership, model provenance and exit cost, none of which appear on a feature comparison sheet. No third-party market share figure appears anywhere, and analyst quadrant placements are treated as evidence about a vendor's analyst relations function rather than about its product.
Every price, product boundary and corporate structure below is a claim about 26 August 2026 and nothing else. Sources run in order of preference: published pricing pages including Internet Archive snapshots where the question is whether a price moved, regulatory text, product and API documentation, company filings and earnings releases, then peer-reviewed and preprint research. Vendor marketing is used, labeled as a vendor claim, and named alongside the product it is selling. Every price quoted is list, and real enterprise deals settle well below it.
Thoma Bravo agreed to acquire Verint on 25 August 2025 at $20.50 per share, an enterprise value of about $2 billion, closed on 26 November 2025, and combined Verint with its existing portfolio company Calabrio. On 18 February 2026 the combined organisation announced it would operate under the single Verint name, with Calabrio solutions continuing inside the Verint CX Automation Platform, and on 24 February 2026 Dave Rhodes, formerly Calabrio's chief executive, was named chief executive of the combined firm. Running Verint against Calabrio on a shortlist in August 2026 is comparing a company to itself under two brands, with one product line eventually losing the roadmap argument.
NICE reported AI and self-service annual recurring revenue of $362 million for the second quarter of 2026, described as 15 percent of cloud revenue. Genesys reported cloud ARR of nearly $2.6 billion for the fiscal year ended 31 January 2026 and AI ARR above $250 million, roughly a tenth. Five9 management put AI revenue at an annual run rate above $150 million against a full-year guide of about $1.27 billion. RingCentral reported 13 percent of ARR from customers using at least one paid AI product, which counts the whole subscription of any customer who buys any AI. The commercial point is not the percentage. It is that four figures measuring different things are presented side by side as though they measure the same one.
Archived snapshots of Five9's own pricing page show all five tiers carrying a published monthly rate in June 2023 (Digital $149, Core $149, Premium $169, Optimum $199, Ultimate $229) and again in May 2024 at materially higher numbers (Digital $175, Core $175, Premium $235, Optimum $290, Ultimate $325). By May 2025 the top three tiers read “Contact Sales” and only Digital and Core carried a price, both $119. As of 26 August 2026 the tiers are Digital, Core, Plus, Pro and Enterprise, two published rates at $119 and $159, and the three highest remain quote-only. The top published rate rising and then disappearing is a disclosure event, not a price cut.
In May 2023 Genesys published CX 1 at $75, CX 2 Digital at $110 and CX 3 Digital plus workforce engagement at $150, alongside a flat AI Experience add-on at $40 per agent per month. On 26 August 2026 the same three tiers publish at $75, $115 and $155, a top-tier CX 4 has appeared at $240, and the flat AI seat charge no longer exists. AI capability is metered in AI Experience tokens at $1.00 each, with an agentic virtual agent interaction consuming 1.2 tokens and AI-based scoring 0.05 tokens per invocation. Three years of AI investment moved the seat price by five dollars and created a variable line item with no ceiling.
The Genesys self-service statistics report defines “Contained in Self-Service” as the total number of interactions that entered the Designer application in Self-Service and were concluded without entering Assisted-Service. Nothing in that definition requires the customer's problem to have been solved, or the session to have ended in anything other than abandonment. Genesys' own marketing has since acknowledged the weakness, writing that a high containment rate might look like a win but might not be, if customers did not get what they needed. No vendor in this set publishes a containment figure alongside a definition of what counts as contained and an independent measurement of it.
Salesforce publishes Agentforce at $2.00 per conversation, or 20 Flex Credits ($0.10) per action at $500 per 100,000 credits. Genesys prices an agentic virtual agent interaction at 1.2 AI Experience tokens, $1.20 at the published token rate. NICE's Ultimate Suite carries $0.25 per session on top of $249 per agent per month. Google prices a generative voice agent at $0.002 per second, about $0.36 for a three-minute call. Amazon Connect prices end-customer self-service at $0.0080 per voice minute on top of $0.018 per voice minute, under ten cents for the same call. The spread is a purchasing decision, not a technical one. The arithmetic is the author's; the underlying rates are the vendors' own.
Microsoft's Copilot Credits Guide of June 2026 states that Copilot Credits are the common currency across Microsoft 365 Copilot experiences including Copilot Cowork, Copilot Studio, Dynamics 365 agents, Power Platform workloads and Work IQ APIs, that they are pooled at the tenant level, and that total cost is the sum of credits consumed across all supported experiences. Pay-as-you-go is $0.01 per credit. The pre-purchase plan runs from 300,000 credits at a 5 percent discount to 300,000,000 at 20 percent, and unused credits expire at the end of the annual term. A contact centre director buying Dynamics 365 Contact Center at $110 per user per month is buying a seat whose AI consumption competes with every other Copilot workload in the company.
Article 50(1), (2) and (5) became applicable on 2 August 2026. Providers of AI systems that interact directly with natural persons, which includes customer-service chatbots, voice bots and AI agents, must make sure the person is informed they are interacting with an AI system unless that is obvious to a reasonably well-informed observer, and the information must be given clearly and distinguishably at the latest at the time of first interaction. The obligation applies to systems already on the market, not only to new ones. Penalties reach 15 million euro or 3 percent of worldwide annual turnover, whichever is higher. Nothing in the published pricing of any vendor here reflects the cost of meeting it, and the duty falls on the deployer rather than the platform.
Traditional quality management samples a low single-digit percentage of interactions, so full-coverage automated scoring is a genuine change in kind. What no vendor in this set publishes is a chance-corrected agreement statistic between its automated scores and trained human reviewers on the same interactions. The nearest adjacent evidence is the largest systematic evaluation of language models used as judges published to date, covering 21 judges from nine providers across roughly 541,000 individual judgments, which found mean Cohen's kappa against human labels of 0.451 on one benchmark and 0.593 on another, and found high test-retest reliability coexisting with severe position bias. Moderate agreement across all interactions may still beat excellent agreement across two percent of them. A buyer can settle that in a fortnight and nobody has published the answer.
On 31 July 2025 Genesys announced a $1.5 billion investment, $750 million each from Salesforce and ServiceNow, with proceeds used to repurchase shares from existing equity holders. The announcement names two existing integrations: CX Cloud, combining Genesys Cloud with Salesforce Service Cloud, and Unified Experience, combining Genesys Cloud with ServiceNow Customer Service Management. A buyer being told by a Salesforce account team that Service Cloud Voice removes the need for a separate contact centre platform is being told that by a shareholder in the separate contact centre platform.
Nobody sells against this number because nobody publishes it in a comparable form, and it is the number that decides whether an AI deflection business case survives contact with a finance team. Take one customer interaction handled entirely by AI, no human, and cost it on each vendor's published list meters. The vendors have chosen units that make the comparison awkward, so the arithmetic below normalises them and shows its working. The NICE and Salesforce figures sit on top of a per-seat fee the Google and Amazon figures do not carry, so this is marginal cost per interaction rather than total cost of ownership.
Three minutes of end-customer self-service at $0.0080 per voice minute on top of $0.018 per voice minute, excluding the DID and any Contact Lens analysis. There is no seat fee at all. The trade is that everything above the primitives is yours to build and yours to keep running.
Metered, no seat fee, fully published
180 seconds at $0.002 per second on a generative Playbooks voice agent, excluding telephony. Google does not sell a contact centre seat and does not own the carrier, so this competes for the AI self-service budget rather than for the platform decision.
Metered, no seat fee, fully published
A per-session charge on the Ultimate Suite tier, sitting on top of $249 per agent per month. NICE's five published suite prices, $110, $135, $169, $209 and $249, are identical in a September 2024 snapshot and on the live page today, a period covering the launch of Mpower and the $955 million Cognigy acquisition.
Per session, on top of the seat
An agentic virtual agent interaction consumes 1.2 AI Experience tokens at the published $1.00 token rate. AI-based scoring runs 0.05 tokens per invocation. Seat tiers publish at $75, $115, $155 and $240, and the flat $40 AI seat add-on that existed in 2023 is gone.
Token meter, no published cap
Per conversation, or 20 Flex Credits at roughly $0.10 per action against $500 per 100,000 credits. The most expensive published AI rate in this set, and the vendors at this end are not overcharging for compute. They are charging for orchestration, integration into the system of record, and the fact that you did not have to build any of it.
Per conversation, on top of the Service Cloud seat
Absent from the comparison because Copilot Credit consumption varies by task and Microsoft publishes no per-interaction rate. Credits run $0.01 each pay-as-you-go and are pooled tenant-wide, so the contact centre's AI burn competes with every other Copilot workload in the organisation.
$110 per user per month, plus a shared credit pool
An operation deflecting two million contacts a year is looking at about $160,000 on Amazon Connect's meters and about $4 million on Salesforce's per-conversation rate. Whether the difference is worth paying depends entirely on whether you have engineers, which is the question the assemble-versus-buy section of the report returns to. The contracting instruction that follows matters more than either number: every one of these meters converts a fixed cost into a variable one, and the variable moves with contact volume, which is the one thing a contact centre cannot control. A seat licence is a bad deal in a surge and a good one in a downturn. A per-interaction meter is the reverse, and no published contract in this set caps it. Ask for the cap. It is the single most valuable thing in the contract-language section and it costs nothing to request.
The shortlist is three shortlists stapled together, and the staple is the buying committee rather than anything about the products. Sorting the vendors by what they were built to sell against explains more about who wins a deal than any feature score does. The report carries a full matrix across routing and telephony, omnichannel and data residency, the workforce layer, model provenance and renewal exposure, published commercial structure, and exit cost, plus nine buying situations with a named call and a named condition that would change it.
Forecasting, scheduling, adherence, recording, evaluation. No primary routing engine, so it sits behind whoever owns the voice path. Best forecasting and scheduling in this set, no published pricing on either brand, and a single owner since November 2025. Buy it when the problem is workforce cost, not when the problem is routing.
Quote-only, both product lines
Suites that give away enough workforce management to make buying a specialist look like duplication. Both own carrier-grade voice, both are genuinely single-queue, both put routing changes behind a trained specialist rather than an operations supervisor. Genesys wins on global voice and complex routing; NICE wins where workforce management complexity outweighs it.
$75 to $240, and $110 to $249, published
Engineered for an operation without a platform team: the customisation ceiling is lower and that is the feature, because a lower ceiling means a shorter implementation and an administrator rather than an engineer. Between them price should decide, and Talkdesk publishes its full ladder while Five9 publishes two tiers of five.
$119 and $159 published; $85 to $225 published
Owns the network, which is the actual asset, and attaches a contact centre to a telephony seat. Deliberately shallow customisation for time-to-value, basic workforce management the vendor is honest about, and the highest single-provider model concentration in this set through its OpenAI relationship.
Partly published, AI partly in the seat
Attach a contact centre to a seat the buyer already owns. Agentic resolution inside the system of record is the one thing Salesforce does that nobody else here does as well, at the highest published AI rate and the highest exit cost in the set. Microsoft's answer splits: Dynamics 365 Contact Center if Dynamics is already the service system of record, the Teams-attached partner route if it is not.
Edition prices published, AI sold separately
Sell primitives and let you attach whatever you like. Unlimited routing customisation because contact flows are code, with no ceiling and no guardrails, an engineer rather than an admin making every change, and total portability of your own data. Reporting is excellent or absent depending on what you built. Only with engineers on staff.
Fully published, metered, no seat fee
Exit cost is the dimension on which this vendor set differs most and the one buyers assess least, and no vendor markets it for the obvious reason. It decomposes into integration depth once live, historical interaction data portability, routing logic portability, and realistic elapsed migration time, which runs from six months on the Teams-attached route to well past a year where the platform is also the system of record. The report also runs four scenarios to the end of 2027 with observable tripwires rather than point forecasts: the meter wins at 40 percent, the suite reabsorbs AI into a fixed seat price at 30, consolidation continues at 20, and regulatory drag at 10. The cheapest tripwire to watch is a pricing page rather than an announcement, because the page changes first.
This isn't a vendor summary. Every sentence is labeled by what stands behind it: verified fact, vendor claim, third-party estimate, my assessment, hypothesis, or scenario. Sources are numbered and clickable. Forward-looking sections use scenarios with observable tripwires, not forecasts. It's the same method behind every market assessment I write.
Forty-seven pages, built from public sources with no client brief and no interviews. Read it in the browser or take the PDF.
Each report here answers a real question, directed and researched against public sources and evaluated against a stated assumption, then delivered as Word and PDF. If you're weighing a platform, sizing a category, or defending a number to a board, tell me the decision behind it and I'll tell you honestly whether a report is the right tool.
Commission an assessment