{{locationDetails}}
{{locationDetails}}
Facts and pricing verified .
Thirteen of these 22 platforms publish no plan price at all. Most vendors that do publish one price the AI on its own line, separate from the software, and meter it, though Zendesk folds AI Agents into its $55 Suite Team seat and Tidio bundles a Lyro allowance into several of its plan tiers. That line runs from $0.10 for a Salesforce Flex Credit action up to $4.00 for a Heavy task at eesel AI, which charges no seat fee at all, with Salesforce Agentforce at $2 a conversation on top of a platform you already license. Zendesk's advertised $19 tier excludes AI Agents outright. So the first number you find on a pricing page is almost never the number you pay.
Now the other half of the problem. Some of the biggest valuations here sit on the thinnest public evidence. Sierra carries a $15.8B valuation behind roughly 15 reviews. Decagon sits at $4.5B on 18. Cresta, Maven AGI, eesel AI, Pylon, Crescendo, Lorikeet and Wonderful have no rated review base at all. That's capital moving a great deal faster than evidence.
Both of those gaps land on you, because you have to defend this choice twice. Once to finance, with an annual number. Once to security, with a certification answer. Zendesk tops this ranking on breadth, which is the easy headline, though breadth isn't what either of those conversations turns on. Every price, plan name, certification and reviewer count below comes from what the vendors and the review platforms publish. Vendor outcome figures are labeled as vendor figures, because in this category every headline percentage is marketing until an auditor says otherwise.
So start with what everyone actually publishes. Nine columns, 22 vendors, and one of them decides your budget. Which one? The AI billing unit. The three compliance columns after it decide whether your security reviewer will even let you keep going.
A quick glossary first. ISO 42001 is the AI governance and management standard, and six of the vendors here hold it. HIPAA is the American health privacy law, and PCI DSS is the card data security standard. SOC 2 Type II, the general security control report, sits in the entries below because most of the vendors here document it, though several do not.
| Tool | Published price | AI billing unit | Free tier | Independent reviews | ISO 42001 | HIPAA | PCI DSS | Also |
|---|---|---|---|---|---|---|---|---|
| 1. Zendesk | From $19/agent/mo | Per seat, then per automated resolution | No | 4.3/5, 7,142 G2 | Yes | - | - | - |
| 2. Intercom (Fin) | From $29/seat/mo | Per seat plus $0.99 per Fin outcome | No | 4.5/5, 2,900+ G2 | Yes | Yes | - | AIUC-1 |
| 3. Salesforce Agentforce | $2/conversation or $0.10/action | Per conversation or per action | No | 4.3/5, 1,194 G2 | - | - | - | Einstein Trust Layer PII and card masking |
| 4. Ada | Quote only | Conversation or resolution based | No | 4.6/5, 173 G2; 4.7/5, 15 Capterra | - | Yes | Yes | AIUC-1 |
| 5. Cognigy | Quote only | - | No | 4.6/5, 13 G2 | - | Yes | - | ISO 9001 |
| 6. Parloa | Quote only | - | No | 4.5/5, 37 G2 | - | Yes | Yes | DORA |
| 7. Yellow.ai | Free tier; paid tiers quote only | $0.99 per resolution past the free allowance | Yes | 4.4/5, 107 G2 | - | Yes | Yes (v4.0.1) | - |
| 8. Sierra | Quote only | Per resolved outcome | No | 4.1/5, roughly 15-17 G2 | Yes | Yes | Yes | FedRAMP, EU AI Act |
| 9. Tidio (Lyro AI) | $0 to $300/mo+ | Plan tier plus $0.50 per Lyro conversation | Yes | 4.6/5, 1,879+ G2 | - | On custom solutions | - | - |
| 10. Decagon | Quote only | - | No | 18 G2 reviews | - | Yes | Yes | EU AI Act |
| 11. Cresta | Quote only | - | No | - | Yes | - | Yes | ISO 27701 |
| 12. Maven AGI | Quote only | - | No | - | Yes | Yes | Yes (v4.0 Level 1) | - |
| 13. eesel AI | $0.40 per Regular task | Per task, no seats | No | - | - | On Enterprise | - | SOC 2 Type II underway |
| 14. Lorikeet | From $1,500/mo | Per resolved ticket, $0.95-$1.50 | No | - | - | BAA on Scale | - | - |
| 15. Gorgias | From $10/mo; AI add-on from $250/mo | Ticket-volume tier plus AI add-on | No | 2.2/5, 151 Trustpilot | - | On request | - | - |
| 16. Freshworks Freshdesk (Freddy AI) | From $19/agent/mo | Per seat plus $49 per 100 AI sessions | No | 1.5/5, 407 Trustpilot | - | - | - | - |
| 17. Kustomer | Quote only | $0.60 per engaged conversation; $40/user/mo for rep agents | No | 6 Trustpilot reviews | - | Yes | - | EU-U.S. Data Privacy Framework |
| 18. Pylon | Quote only | - | No | - | Yes | BAA on enterprise | - | - |
| 19. Kore.ai | Quote only | - | No | 3.9/5, 16 PeerSpot | - | Yes | - | Runtime governance and audit trails |
| 20. Crescendo | Quote only | - | No | - | - | Yes | - | - |
| 21. Forethought | Quote only | - | No | 2.5/5, 8 Trustpilot | - | - | - | SOC 2 Type II for Enterprise |
| 22. Wonderful | Quote only | - | No | - | - | - | - | - |
Seven rows give you no rating at all. Is that an oversight? No, it's the finding. Cresta, Maven AGI, eesel AI, Lorikeet, Pylon, Crescendo and Wonderful give you nothing to reference-check before you sign. Sierra and Decagon are barely better, at roughly 15 to 18 reviews each, and Sierra says its customer list covers 40% of the Fortune 50. Where that column is thin, every check lands on you.
One caution on that column. A G2 score is buyers rating a purchase. A Trustpilot score is usually end customers rating the bot that would not let them reach a human, or merchants rating the vendor's own support desk. Gorgias at 2.2 and Freshdesk at 1.5 are buyers on billing and support. Kustomer's six and Forethought's eight read largely like end customers rather than buyers, though at that sample size you cannot be certain who is writing. Compare within a platform, never across.
Ratings alone won't settle this. Four names still stand out, each with the thing that will slow you down attached.
Three of those four will not quote you without a call. That is the second half of the underwriting problem. Evidence is one half, cost is the other, and only eight vendors let you do that arithmetic tonight.
Below is each published meter run at two monthly volumes, plus the annual figure your CFO will ask for. Seat fees are excluded, so treat these as the AI line alone, sitting on top of whatever your helpdesk already costs you. One caveat: these run the meter gross. Freshdesk's 500 included sessions and Yellow.ai's free allowance come off the top before you bill, and Freshdesk's allowance alone is worth $245 a month.
| Vendor | Published unit rate | 1,000 units/mo | 5,000 units/mo | Annual at 5,000/mo | What the unit is |
|---|---|---|---|---|---|
| eesel AI | $0.40 per Regular task | $400 | $2,000 | $24,000 | eesel publishes the ladder itself: 100 tickets $40, 1,000 tickets $400, 2,500 tickets $1,000. Heavy tasks bill at $4.00 |
| Tidio (Lyro AI) | $0.50 per Lyro conversation | $500 | $2,500 | $30,000 | Billed on top of, or instead of, your flat tier |
| Freshdesk (Freddy AI) | $49 per 100 AI Agent sessions | $490 | $2,450 | $29,400 | The $19/agent Growth plan gives you 500 sessions before the meter starts |
| Kustomer | $0.60 per engaged conversation | $600 | $3,000 | $36,000 | Add $40 per user per month if your reps use AI Agents for Reps |
| Intercom (Fin) | $0.99 per Fin outcome | $990 | $4,950 | $59,400 | On top of your $29 to $132 seats. Standalone Fin carries a minimum monthly commitment |
| Yellow.ai | $0.99 per additional resolution | $990 | $4,950 | $59,400 | Your free tier covers 1 AI agent and 500 chat sessions a month first |
| Lorikeet | $0.95-$1.50 per resolved ticket | $1,500 plan floor | $4,000 (Scale floor) | $48,000 (Scale floor) | You're billed only for tickets it resolves. Start is $1,500/mo under 5,000 tickets, Scale $4,000/mo from 5,000 to 20,000 |
| Salesforce Agentforce | $2 per conversation | $2,000 | $10,000 | $120,000 | Or Flex Credits at $500 per 100,000, where one agent action costs 20 credits ($0.10) |
Eight vendors out of 22. That's your whole modelable population before you take a sales call. Zendesk bills AI Agents per automated resolution without publishing the rate. Gorgias puts its AI Agent behind a $250/mo add-on instead of a meter. The other twelve give you nothing to multiply.
Look at the spread. The AI line alone runs $24,000 to $120,000 a year at 5,000 resolutions a month, before a single seat. So which unit do you model? The one the contract bills, not the one the homepage markets.
The contract that does the billing depends on where you buy. Three routes are on this list. They cost very different things, and they decide who you're even allowed to ask.
| Route | Who it is | What the AI line costs | What you give up |
|---|---|---|---|
| Incumbent meter | Zendesk, Intercom, Freshdesk, Salesforce Agentforce, Kustomer | Zendesk steps from $19 to $55/agent, then bills per automated resolution. Freshdesk adds $49 per 100 sessions above 500. Intercom adds $0.99 per outcome to $29 to $132 seats. Agentforce is $2 per conversation, and worth it only if the record already lives in Salesforce. | Leverage. The AI rides on a platform you have already signed for, and the meter is priced accordingly. |
| Independent layer | eesel AI, Tidio (Lyro Connect), standalone Fin, Ada, Maven AGI, Lorikeet | eesel is $0.40 a task with no seats across nine named helpdesks. Lyro Connect adds the agent with no migration. Fin runs standalone on Salesforce at $0.99 an outcome with a minimum commitment. Lorikeet starts at $1,500/mo. | One vendor to hold responsible. Two contracts, two support paths, and on three of the names in this row, eesel AI, Maven AGI and Lorikeet, no public review base at all. |
| Rip and replace | Crescendo | Quoted, after a white-glove onboarding that its own material puts at roughly 30 days. | The stack you run today, and any chance of a quiet, low-stakes pilot. |
So which route wins? The bolt-on is usually the cheaper experiment, though not at every vendor in that row. It runs on one ticket type, on your existing helpdesk, at a rate you can total before the call, and eesel bills $0.40 a task with no floor under it, while Lorikeet starts at $1,500/mo billed annually. The incumbent wins on procurement speed, not on price: the paper is signed, the security review is done, and the AI line is an amendment rather than a new vendor. Tip: price the bolt-on first, since it is the route most likely to hand you a number without a meeting.
One more filter, and it's brutal. Tidio, eesel AI, Gorgias and Yellow.ai let you see a price today. Intercom, Freshdesk and Lorikeet publish a meter above a seat or a floor, and Agentforce publishes a meter with no platform price under it. Zendesk publishes the seat ladder but not the resolution rate above it. Kustomer is the odd one out, publishing its AI meter at $0.60 but not the platform tier underneath it, so it lands on both lists.
Everyone else needs a demo cycle before any number exists. If your process forbids quote-only vendors, you've just cut this field to ten, and to the eight in the cost table above if you require the AI rate itself to be published.
Zendesk wins on surface area, and you feel it on day one. You get AI Agents, a Copilot for your team, Intelligent Triage, Auto Assist, generative AI for voice, a unified knowledge graph and omnichannel routing, all sitting on the ticketing core you probably already know.
The Marketplace runs to 1,800+ apps, so whatever else you run, something likely connects to it. Zendesk reports customer automation rates of 80% at Best Egg and 80% at TeamSystem, the second with a 99% cut in repetitive emails.
Now the part that trips up your budget. Is $19 the number you take to finance? No. AI starts at $55, and the advertised entry plan is email and ticketing only. There's no permanent free plan either, so your evaluation window is a 14-day trial.
G2 reviewers praise the consolidated omnichannel view and an interface that stays intuitive day to day. Price is the recurring reservation among four-star reviewers. Setup complexity is the other, along with the extra training new agents need.
Zendesk won't tell you what a resolution costs. Fin does, and it arrives with deployment numbers attached. Intercom reports a 76% average resolution rate across 12,000+ customers, many of them above 85%, and 2 million resolutions a week.
It covers your chat, email, voice, Slack and social in one inbox. It'll also run standalone on Salesforce or another helpdesk if you would rather not move.
Every seat tier adds the same $0.99 per Fin outcome. What happens in your busiest month? The bill climbs with the ticket count, which is exactly the month you least want a surprise on the invoice.
Reviewers describe conversations that don't feel robotic, with roughly 80-90% of answers landing correctly and little repeated clarification. The complaints cluster just as tightly. Per-outcome pricing that gets unpredictable and expensive at scale, occasional hallucinations, and trouble with complex or multi-part queries.
Fin will run on someone else's helpdesk. Agentforce won't. It's an add-on to a platform, and its value comes from Data Cloud and the Salesforce record underneath the agent. Powerful if that's where your customer data lives. Close to pointless if it isn't.
You build agents from subagents, actions and instructions in Agentforce Builder, then deploy to portals and messaging channels with handoff to humans.
You can see the price before a call, which is rarer here than it should be. You pay $2 per external customer conversation, or you buy Flex Credits at $500 per 100,000, where one agent action consumes 20 credits and costs $0.10.
Salesforce reports up to 72% deflection of routine inquiries, 32% faster resolution and an estimated 33% lift in CSAT, the customer satisfaction score. Those are the vendor's own figures.
G2 named Agentforce the #1 Best Agentic AI Product for 2026. Then read the reviewers of sibling Agentforce products, who cite a steep learning curve, complicated implementation and high overall cost.
Complexity should buy you something. With Ada it buys your builders more to work with than most of this field offers. One Reasoning Engine drives behavior across voice, email, chat, WhatsApp, SMS, Instagram and in-app. Around it sit Playbooks, Coaching, Simulations and a Developer Toolkit with an MCP Server, MCP being the Model Context Protocol that connects agents to outside systems.
Ada publishes an 84% automated resolution rate and claims agents 8x more productive than human ones, alongside 550+ enterprise agents deployed, 6.4 billion interactions handled and 99.9% platform uptime.
The reviewer picture is unusually warm for a tool you can't price yourself. Ada sits at 4.6/5 across 173 G2 reviews and 4.7/5 across 15 on Capterra, with non-technical staff citing a low barrier to entry.
Those same reviewers say adding new channels or features can eat real time and resources. Language coverage outside English lags too. Pricing runs through a consultation booking, so budget a demo cycle into your timeline.
Ada spreads across every channel you run. Cognigy narrows: it leans hard into the voice-heavy contact center. The integration list gives it away: Avaya, Amazon Connect, Genesys, NiCE CXOne, Microsoft and 8x8, plus 30+ omnichannel connectors and an Extension Framework.
Knowledge AI runs on retrieval-augmented generation, which simply means the agent answers from your documents instead of from memory. Cognigy's own homepage names Toyota running 25+ agents on it, and Lufthansa handling 16M+ automated conversations a year.
Those are big logos, and the company behind them has changed hands. NiCE closed a $955M acquisition of Cognigy on September 8, 2025, the largest in its 40-year history, and NiCE projects the portfolio will grow annual recurring revenue 80% year over year in 2026. Deep pockets, then, behind a publicly traded parent. Also a roadmap now set by a new owner, which is worth a contract clause.
The 13 G2 reviewers rate it 4.6/5. They describe a friendly visual builder for basic bots, dense documentation, and real complexity once you want deep customization.
Cognigy has to win over your contact center. Parloa is built to win over the committee behind it. The certification file targets regulated buyers, and Parloa states the platform is built to handle ID checks, payments and medical information.
The connector bench is just as serious: Genesys, Twilio, AWS Connect and Cisco on telephony, Salesforce, SAP and ServiceNow on the record side, plus support for bringing your own model, speech-to-text and text-to-speech. That last part keeps you off a single vendor's rails underneath.
Money and momentum back up what you're buying into. A $350M Series D at a $3B valuation, per its January 15, 2026 funding announcement. More than $560M raised in under four years, revenue the company puts past $50M, and customers including Allianz, Booking.com, HealthEquity, SAP and Swiss Life.
Parloa also states it is the sole customer-facing voice AI contact center platform with SAP Endorsed App Premium Certification. Its own case study puts BarmeniaGothaer's switchboard workload down 90% and NPS, the net promoter score, up 179%, which is the kind of number your CFO will ask you to repeat.
Two honest caveats. There's no self-serve signup, so evaluation starts with a demo. And 37 G2 reviews is a thin base for a $3B company. That doesn't disqualify it. It moves the diligence onto your reference calls.
Parloa makes you start with a demo. Yellow.ai flips that. It's one of only two platforms here with a permanent free plan you can actually run. It gives you 1 AI agent and 500 chat sessions a month, then charges $0.99 per additional resolution. Enough to run a support desk? No. Enough to prove a use case, yes, and that's roughly the point.
Above it, the Basic tier carries no published price and Enterprise is quoted. Reviewers flag exactly that opacity as a cost-escalation risk.
What you get is broad. A no-code agent builder. Nexus Vox voice AI, which Yellow.ai says supports 500+ languages and clones a voice in 10 seconds. Email and text channels, 20+ analytics dashboards with sentiment analysis, and 150+ prebuilt integrations covering Salesforce, Zendesk and Genesys among others.
G2 reviewers put it at 4.4/5 across 107 reviews, 96% of them at four or five stars, and single out the builder as easy to use without heavy engineering. The company reports 650+ global brands and is going public through a $550 million SPAC merger with Bluerock Acquisition Corp.
A free tier gets you started. It won't get you past your risk committee. Sierra earns this pick on the two things that committee reads first: a certification file that runs from SOC 2 and ISO 27001 through ISO 42001, HIPAA and PCI DSS, and the fact that it is the only platform in this ranking that names FedRAMP at all.
Behind that sits a $950M Series E in May 2026 at a $15.8B valuation, roughly $1.6B raised, and a customer list Sierra says covers 40% of the Fortune 50, including Rocket Mortgage, Gap Inc., SoftBank, Vanguard and Rivian.
The product targets work that finishes, not chat that ends, which is what you want if the tickets in your queue involve money moving. Insurance claims processing, mortgage origination, returns.
Ghostwriter turns your standard operating procedures and transcripts into an agent without code. Agent Studio, Explorer, Context Engine and Monitors handle building, analytics and oversight across voice, chat, email, SMS, WhatsApp and ChatGPT in 59 languages.
So where's the catch? In everything you can't check. Pricing is outcome-based with no rate card, the public pricing route is a demo request, and the review base is roughly 15 to 17 entries at 4.1/5. Those reviewers report a steep learning curve, complex setup, and context loss that turns longer conversations repetitive or generic.
Not every queue needs FedRAMP. Tidio comes at this from live chat, not enterprise CX, and for a small ecommerce or SaaS team that's a feature. The free plan is genuinely free, and the ladder above it is published as far as Plus, with Premium quoted.
Lyro Connect is the underrated piece. It plugs the agent into Zendesk, Intercom, Salesforce and 20+ other platforms with no migration, next to 120+ native integrations including Shopify and HubSpot.
Tidio says Lyro resolves about 67% of inquiries on average, and names Under Armour, ADT Security, The Body Shop and Stanley among 300,000+ businesses.
Trustpilot reviewers praise the support team and how fast you can get running. Then they raise harder things: account suspensions with little explanation, unexpected price increases and difficulty canceling. Weigh that before you put it in front of revenue.
More than $500M raised since founding in August 2023. Eighteen reviews. That gap is the story here, and it sits next to a $4.5B valuation reached in January 2026 after tripling in under six months.
The product covers web chat, SMS, WhatsApp, mobile, voice and email, with memory that carries across channels. Watchtower handles quality monitoring, Duet Autopilot handles self-improvement, and Decagon Assist gives your human agents live translation.
Decagon publishes 80% deflection at Duolingo, 95% cost reduction at ClassPass and 70% resolution at Chime.
The 18 G2 reviewers describe implementation as fast and hands-on, call setup quick and intuitive, and report smooth integration into an existing Zendesk stack. They also say response times slow and escalations rise when ticket volume spikes, and that some features still feel unfinished.
Read that second half twice. An AI agent gets judged during a volume spike, which is precisely the moment those reviewers describe going wrong. The pricing route is a demo form, and there's no free trial behind it to fall back on.
Start with what you can't check. The review platforms carry nothing on Cresta, so every claim below is the vendor's own and your reference calls do all the work.
Is most of your volume voice, with your people on a floor and a supervisor walking it? Then keep reading. Cresta ships five products, AI Agent, Agent Assist, Conversation Intelligence, Quality Management and Knowledge Agent, so assist and quality tooling matter here as much as automation does.
The integration bench leans telephony, with Five9, Genesys and Avaya alongside Amazon Connect and Cisco. That list identifies the buyer better than any marketing line does.
Cresta states it is among the first companies certified to ISO/IEC 42001, the AI management standard, which is a first-mover claim nobody else here makes. It has raised $270M+ including a $125M Series D in 2024, and reports $100M in annual recurring revenue in 2026 and more than 600 employees since 2017.
The published outcomes are Cresta's own. Snap Finance is cited at 5.5x higher containment and a 23% CSAT increase, Cox at a 20% revenue increase, Brinks Home at a 30-point NPS gain.
United Airlines, Marriott, Cox, Intuit and CVS are named customers, so the references exist if your rep will share them.
Cresta bets on your contact center floor. Maven bets on the systems behind it, and the premise is that you shouldn't have to move your data to automate your support. Its Graph of Record connects to your systems of record without migration, and it plugs into Salesforce, Zendesk, Freshdesk and Intercom directly, with a public developer API on top.
Charters scope what an agent is allowed to do. Intelligent Fields pull structured data out of transcripts, and one reasoning engine spans chat, email, voice and web with simulation, testing, monitoring and governance wrapped around it.
For your security reviewer, this is a strong file, and the ISO/IEC 42001 line on it is what your AI governance policy will ask for. Maven says it powers 90+ companies, including publicly traded enterprises. No public review base exists to test that against, which is the trade you are making.
The numbers you do get are Maven's own: 93% autonomous resolution at Mastermind inside six weeks, and 90% of incoming queries handled at Tripadvisor.
Cisco Investments and Dell Technologies Capital are both investors and named customers. Useful signal, and not the same thing as a reference call you run yourself. There is no funding total, headcount or uptime commitment on the site.
After two booking forms in a row, eesel comes as a relief. It publishes one of the clearest price pages here. Light tasks are free. Regular tasks cost you $0.40. Heavy tasks cost $4.00. eesel even does the arithmetic for you: 100 tickets a month is $40, 1,000 is $400, 2,500 is $1,000.
No platform fee, no monthly minimum, no per-seat charge. A $50 usage credit with no credit card lets you test before you talk to anyone, and an annual commitment takes 25% off. Enterprise costs you a $1,000/mo flat platform fee for higher knowledge limits, SSO and HIPAA support with a business associate agreement.
It also goes where you already work: nine helpdesks by name, including Zendesk, Freshdesk, Intercom, Gorgias and Salesforce, plus seven knowledge sources, six ecommerce platforms and seven deployment channels down to WhatsApp, Instagram DMs and SMS.
Here's the trade. eesel's SOC 2 Type II certification is underway, not finished, and its public review presence is close to nonexistent, with a single reviewer reporting the blog-generation feature stopping mid-task and losing context.
Cheap and transparent, and unproven in public. For a pilot on one defined ticket type, it's a fair bet. For your entire front line, ask harder questions first.
$0.95 to $1.50, and only when the ticket is actually resolved. That single line does more for your cost forecast than most of this market manages, across chat, email, SMS and voice. The floor is real, though, and it's annual.
It aims squarely at fintech and healthtech, and the compliance file carried on both paid tiers is built for exactly that buyer. Deterministic guardrails sit over natural-language workflows, so decisions stay explainable when your compliance team asks.
Voice agents execute actions mid-call, and WhatsApp reaches your customers in 180+ countries. The piece your QA lead will want is the Coach Agent, which reviews 100% of tickets against the 2-5% a CSAT survey typically samples.
Lorikeet reports 97% faster resolution times and 99% accuracy, and names Linktree, Magic Eden and Eucalyptus among its customers. There's no free tier and no published trial.
Lorikeet aims at fintech and healthtech. Gorgias aims at your storefront, and if you run Shopify it's hard to ignore. It says it powers 17,000+ ecommerce brands and 40% of Shopify brands, with 100+ integrations across 17 categories and native Shopify, BigCommerce, Magento and WooCommerce support.
Gorgias cites Bare Minerals at an 8.83x return on AI-driven interactions. The ladder is published in full, and the AI Agent is a separate line on it, not a feature inside a tier.
Trustpilot reviewers rate Gorgias 2.2/5 across 151 reviews. They credit the consolidated helpdesk across multiple stores and the time automation saves them. Then they describe unresponsive or "nonexistent support" when something breaks, a new interface that launched "completely broken", integration problems, and unsolicited marketing email from multiple Gorgias domains with no effective unsubscribe.
That's a hard signal to wave off. So what do you do with it? Weigh it against the Shopify fit, which is genuine, and remember what you're buying here. Customer support.
Gorgias puts its AI on one add-on line. Freshdesk splits the AI you're paying for three ways. Freddy AI Agent resolves autonomously. Freddy AI Copilot sits with your reps, suggesting replies and translating live across 60+ languages, while Freddy AI Insights handles analytics and root-cause alerting.
Freshworks says the Agent resolves up to 80% of queries autonomously, that Copilot contributes to 60% improved agent productivity, and that 50,000+ businesses use Freshdesk.
Growth includes 500 Freddy AI Agent sessions before the meter starts. Run 5,000 sessions in a month and you bill 4,500 of them, so $2,205 before a single seat. Here's how the rest of the ladder stacks up.
Trustpilot shows 1.5/5 across 407 reviews, and the distribution is genuinely split rather than uniformly bad, with 38% five-star and 15% four-star. Positive reviewers cite solid core ticketing and value for the price. Negative ones report scripted, unresponsive support, unexpected renewal charges and difficult cancellation, plus bugs and fragmented third-party integrations.
Three AI lines is a lot to track. Kustomer's invoice gets to four. Start with what works, though: Kustomer's angle is the timeline, and it's a good one. Chat, email, voice and social land on one continuous customer record, and the AI modules live inside that record: Concierge, Envoy, Architect and Data Explorer, plus native Model Context Protocol support so outside agents can act on your live Kustomer data.
Kustomer's own homepage lists more than 600 named customers, among them Sweetgreen, Turo, SKIMS, Hopper and Rappi, so you can ask for peers. The same page credits Hopper with 50% faster first response times.
Pricing is where it gets uncomfortable. The Enterprise and Ultimate tiers are quoted after a sales assessment, and the meters stack above them: $0.60 per engaged AI conversation for customer-facing agents and $40 per user per month for rep-facing agents on the AI side, plus $25 per user per month for the HIPAA-enabled subscription and $50 per GB per month for storage.
How many lines are on that invoice? Four, and you can't total them before a call.
The independent base is six Trustpilot reviews, and it skews negative. Reviewers call the bot responses irrelevant rather than intelligent, report chat threads auto-closing mid-conversation, and describe escalation to a human as impossible.
Pylon starts from a different customer. Not a consumer with a browser, but another company with a shared Slack channel, and it builds accordingly.
Slack, Teams, email, SMS, WhatsApp, phone, Discord and Telegram, plus a customer portal, chat widget, account and product intelligence, triggers, reporting and broadcasts. Five named agent types (Assist, Background, Slack, Support and Skills) investigate and resolve issues.
The 25+ connectors lean B2B too, including Salesforce, HubSpot, Jira and Linear, with a public API and webhooks behind them.
The compliance file is strong, and the ISO 42001 line puts Pylon in a group of six on this page. Pylon reports 1,500+ B2B customers, names Cognition AI as one of them, and is backed by Andreessen Horowitz and Y Combinator. No funding total or founding date appears on the official pages, so company scale is one more thing your diligence has to establish. Pricing is a demo booking.
Pylon points you at its investors. Kore.ai points you at analysts, and their reports do half the work of your internal business case. Leader in the Gartner Magic Quadrant for Conversational AI Platforms (2026). Leader in the Forrester Wave for Conversational AI in Customer Service (Q2 2026). Morgan Stanley, Deutsche Bank, Johnson & Johnson and Pfizer are customers.
The AI for Service module gives your floor real-time agent assist, automated quality scoring across 100% of interactions, sentiment and conversation intelligence, and proactive outbound campaigns across chat, voice and web. Kore.ai's own service page puts one retail-bank deployment at 95% AI agent accuracy, 75% automated interactions and a $97M annual cost reduction, none of it audited.
PeerSpot reviewers rate it 3.9/5 across 16 reviews, which is the thinnest kind of comfort. They like a low-code builder usable without deep AI expertise, and report 30-45% contact-center budget reductions.
They also describe incomplete, disorganized documentation that pushes them into support tickets, and support quality after implementation that varies by region. One reviewer says the platform lacks modern agentic features some competitors now offer. Pricing runs entirely through a sales form, so your timeline starts with a call.
This one wants the whole stack. Crescendo's own comparison pages line it up against Zendesk, Genesys and Agentforce, not alongside them.
Seven purpose-built agents cover Concierge, Agent Assist, Quality, Applied Insights, Optimization, Integration and Knowledge, wrapped around AI live chat, AI voice, email autoresponders, a unified helpdesk and CSAT dashboards.
Crescendo claims up to 99.8% accuracy in AI-generated resolutions, and automatic resolution of over 90% of customer emails. The case studies on its own site put EVPassport's response time down from 7 hours to 18 seconds with a 75% CSAT increase, and Rachio at 95% overall satisfaction and 65% instant resolutions against 1M ticket sales annually.
What are you actually buying here? Software and staffing. The features page markets access to 3,000+ trained support agents, and onboarding is white-glove, with Crescendo's own material putting go-live at roughly 30 days.
Compliance is documented for your reviewer, with dedicated trust, security, SLA and SLO pages behind it. Integrations are described as governed tools that connect backend systems, with no named connectors on the page.
Crescendo wants to replace your stack. Forethought became part of one: Zendesk has owned it since March 2026. The acquisition was announced March 11 and completed on March 26, reportedly north of $200M, Zendesk's largest in two decades, and the agents became purchasable under Zendesk on June 4, 2026.
The platform itself is substantial. Discover mines insight from your tickets. Solve resolves across chat, email, voice, SMS, Slack, mobile and a headless API. Triage routes your tickets, QA scores your agents, an Agentic AI Copilot assists the humans, and a Browser Agent reaches legacy systems that have no API at all.
Forethought markets up to 98% resolution, 15x average return and a 55% cut in first response time, with Grammarly, Thumbtack, Upwork and UPS among its customers.
Then read the other side. Trustpilot shows 2.5/5 from 8 reviews, 88% of them one-star, and the complaint is remarkably consistent. The bot goes in circles and won't hand off to a human. Reviewers describe answers that repeat what is already on the site, and call the experience "useless" and "garbage technology".
Small sample, and those eight read like end customers rather than buyers. Ask about the escalation path in writing before anything else, because all three tiers are quoted and a sales-run Proof of Value stands in for a free trial.
Forethought at least has a parent company behind it now. Wonderful is early stage, and the homepage says so, describing a company actively hiring. That's why it lands last: there's the least here for you to underwrite.
The engineering ideas are good. Agent Studio gives you reusable skills, versioning with rollback, A/B testing between agent variants, a reuse catalog and resource-level permissions, and an A2A protocol hands work between agents across platforms.
Agents work across voice, chat, email and documents, connected to the systems the work runs on, so they read data, trigger actions and update records. Wonderful cites a telecom deployment at 91.5% containment, and bank collections automation across 2.5M+ clients.
Deployment runs through vendor-supplied local teams. Pricing runs through a contact form. If your procurement process has a floor, this vendor sits under it today. If you're running an innovation pilot with a friendly legal team, the feature set is worth a conversation.
A young vendor is one risk you can see coming. Ownership is the one that arrives without warning, and two acquisitions on this page have already landed. Zendesk bought Forethought in March 2026, so two entries here now answer to one owner. NiCE closed its $955M acquisition of Cognigy in September 2025.
The ownership structures are shifting too. Zendesk itself is private-equity owned, taken out for approximately $10.2 billion by a consortium led by Hellman and Friedman and Permira, with ADIA and GIC, in a deal completed November 22, 2022. Yellow.ai is heading for the public markets through a $550 million SPAC merger.
So what happens to your roadmap when the cap table changes? Get that answer in the contract, not on a call.
Usually not. Zendesk's $19/agent Support Team plan is ticketing only, and AI Agents begin at $55/agent on Suite Team, then bill per automated resolution on top of that. Treat the headline seat price as your floor, never as your number.
The unit changes at every vendor, so compare definitions before you compare rates. A conversation, a task, a session and a successfully resolved ticket are four different products. Lorikeet states plainly that failed attempts do not bill, and the meters at Sierra, Intercom, Yellow.ai and Zendesk are resolution-gated on their face too. The rates are in the cost table above. The spread between the cheapest and priciest published unit runs at least 40 times over, from Salesforce's $0.10 Flex Credit action to eesel's $4.00 Heavy task, with eesel's Light tasks free underneath both.
Two of these give you a real free plan. Tidio starts at $0/mo with 50 billable conversations, 10 seats and 50 one-off Lyro AI conversations. Yellow.ai hands you 1 AI agent and 500 chat sessions a month before the $0.99 meter starts. eesel sits in between, since Light tasks cost nothing and the $50 usage credit needs no credit card.
Four more give you a clock instead. Zendesk and Intercom run 14 days, and so does Freshdesk on its Enterprise trial. Gorgias gives you 7 days capped at 10 email messages. Forethought substitutes a sales-led Proof of Value, with no clock published. The rest publish no free tier and no self-serve trial at all, so evaluation there starts with a sales call.
Thirteen of the 22 route their plan pricing through sales. That puts a demo cycle in front of your comparison instead of after it, so budget the calendar time into your plan now. Crescendo's own material says a deployment goes live in roughly 30 days. Lorikeet's Enterprise tier includes dedicated engineering support throughout the contract. Wonderful implements through its own deployment teams. None of that is a quick swap. If you need a defensible number this quarter, start with the eight vendors that publish a rate.
Six hold ISO 42001: Zendesk, Intercom, Sierra, Cresta, Maven AGI and Pylon. Only Sierra names FedRAMP, and only Sierra and Decagon name the EU AI Act. Set your bar at ISO 42001 certified today, and that one line cuts the field by roughly three quarters.
That bar gives you a shortlist. It doesn't give you a commitment. Certification status is exactly the kind of thing that shifts between a website and a contract, and everything above stops where the public record stops. The rest you get in writing. Your RFP, the request for proposal you send to the shortlist, is where these gaps close.
Here is the shortlist that falls out of the tables above. Start with your incumbent's meter, because the paper is already signed. Add one independent layer you can price tonight, which means eesel AI, Lorikeet or standalone Fin. Then, only if security sets the bar at ISO 42001, add Sierra, Cresta, Maven AGI or Pylon, and budget a demo cycle for whichever you pick. That is three names and one calendar, and the eight questions below are what you send them.
Vendors that answer plainly usually have something to show you. Vendors that route every question back to a demo have told you something too.