Best AI Customer Service Software in 2026: The Two-Architecture Shortlist

The short answer

There is no single best AI customer service software. There is a best one for what you want to keep, your ticket mix, and how you want to pay for AI. As of June 2026: pick Decagon or Sierra if you want a heavily-funded standalone agent layer with the largest named enterprise references and you intend to keep your existing helpdesk. Pick Intercom Fin if you already run Intercom for in-app messaging. Pick Zendesk AI if you are standardizing one mature vendor across many departments and want the safe committee choice. Pick Gorgias AI for the widest native Shopify app marketplace. Pick Ada for very high-volume enterprise deflection. Pick Richpanel if you want the helpdesk and the agents rebuilt as one platform on one bill, around $0.30 per conversation, the AI set up by a CX Manager AI rather than a vendor services team, evals your own CX team authors, and a 50% resolution guarantee in 30 days or your money back.

Where Richpanel is the weaker pick, stated up front: Decagon and Sierra out-fund us and carry larger named enterprise logos, and above roughly $30K a year that maturity signal genuinely moves buyers. We also do not host native voice (we integrate with Aircall, Dialpad, and JustCall). The full reasoning, the seven criteria, the matrix, and a decision tree are below.

Why these lists mislead

The author always wins. That is the tell.

Most of the highest-ranking "best AI customer service software" pages share one of two flaws: they are written by a vendor that appears in its own ranking, or by an affiliate that scores on a feature checklist and links the highest-paying tool first. In both cases the scorer has a stake in the result.

The pattern is consistent enough to be almost a law. Fin (Intercom) ranks Fin number one on its own "best AI chatbots" page with the highest resolution claims.^[1] Gorgias publishes a Gorgias-versus-Richpanel comparison in which Gorgias wins.^[2] Affiliate roundups carry the opposite bias: the tool with the most checkmarks and the best affiliate payout floats to the top, whether or not those features change a single resolution. Neither tells you which platform will actually resolve your tickets.

None of this is dishonest, exactly. The products are real and several are excellent. But a ranking where the scorer is also a contestant is not a ranking. It is a pitch with a comparison table bolted on, and the criteria are almost always chosen after the conclusion so the home team wins every column. The giveaway is an asymmetric scorecard: when one vendor leads on all seven axes, either the author did not pick a stiff enough competitor, or the criteria were reverse-engineered from the desired result.

So here is the deal for this article. Richpanel is one of the seven platforms below, and I am not going to pretend otherwise or claim we win everything. We do not, and I will be specific about where we lose. I name a situation where each of the other six is the better choice, define every criterion before the matrix appears, and link each vendor's own page so you can check my reading. If a cell is wrong, the correction address is at the bottom and we will fix it in public.

The split that reorders every list

Two architectures wear the same three words.

Before you compare any vendor, sort it into one of two buckets, because the buckets answer different questions and have different failure modes. "AI customer service software" and "AI customer service agents" are used interchangeably, but the products behind the phrases are not the same shape.

Pattern A: AI bolted onto a helpdesk. Zendesk, Gorgias, Intercom Fin, and Freshdesk began as ticketing systems, some of them in the 2010s, then added an AI layer as a separately-priced feature. You keep the helpdesk you know, and the AI rides on top of it. The strength is continuity: same inbox, same workflows, same UI your team already learned. The structural catch is the billing. You pay for the seat or the ticket, and then you pay again per AI resolution, which we break down in the metering section below.

Pattern B: a standalone AI agent layer. Decagon, Sierra, and Ada are the agent on its own. They do not replace your helpdesk; they sit on top of it (Zendesk, Salesforce, or whatever you run) and resolve conversations, handing off to that helpdesk for the human tickets. The strength is focus and, for the well-funded players, real enterprise pedigree. The catch is that you are now running and paying for two systems, and the agent is bounded by whatever data and actions the host helpdesk exposes.

The third option is to collapse the two. An AI-native platform rebuilds the helpdesk around the agent, so one product resolves the repeat work and your people own the exceptions, on one bill. That is the bucket Richpanel is in. For the agent-first view of this category, see our pillar on the best AI agents for customer support. None of the three buckets is automatically right. Which one fits depends on a question the published lists never ask: what are you trying to keep? If you are wedded to your current helpdesk and only want autonomous resolution added, a Pattern-B layer is the cleaner fit. If your incumbent AI has not earned its place and you would rather consolidate, the AI-native cut wins. The buckets, not the brand names, are where the decision starts.

The two questions that decide the ranking

Does it resolve the ticket, and what does the AI cost on top?

Within and across those buckets, two fault lines split the field, and applying them reorders every published ranking. The first is resolution versus deflection. The second is whether AI is metered on top of the platform you already pay for.

Fault line one: resolution, not deflection. Deflection counts a ticket as handled when the customer stops asking. That includes the customer getting a useful answer. It also includes the customer giving up, rage-quitting to a competitor, or re-opening the same issue angrier two days later. Resolution counts a ticket as handled only when the customer's actual problem is solved, the action it required is taken, and the outcome is confirmed. A platform that "deflects" 80% of tickets can be operationally worse than one that resolves 60% and escalates the rest cleanly, because the deflected-but-unresolved volume is your most expensive customers: the ones who churn quietly or escalate loudly. This is also where the legacy helpdesk AI separates from the agent layers: native Zendesk AI leans toward deflecting to help-center content, and real-world autonomous resolution there tends to land around 10 to 20 percent,^[3] while the dedicated agents (Decagon, Sierra, Ada) and AI-native platforms are built to resolve and act end to end. We wrote the full breakdown in AI chatbot vs. AI agent.

Fault line two: the AI double-meter, and where it does not apply. The Pattern-A helpdesks meter AI on top of the platform: as of June 2026 that second meter runs roughly $0.90 per AI resolution on Gorgias, $0.99 per resolution on Intercom Fin on top of $29 to $132 seats, and a stacked add-on bill on Zendesk that reaches roughly $215 per seat before per-resolution AI is added on.^[4] At volume, the AI line can equal or exceed the platform line. The standalone agent layers price differently: Decagon and Sierra use outcome-based pricing, so the "double-meter" critique does not apply to them. It is a real difference between billing models, not a knock on every vendor. An AI-native platform charges once: Richpanel works out to roughly $0.30 per conversation with a model and token budget you choose, with no second AI meter. Rank these vendors on feature count and deflection and they blur together. Rank them on confirmed resolution, action execution, and total metered cost, and the field separates fast.

Defined before the comparison

Seven criteria, defined so two evaluators score the same.

These are operational, not vibes. Each is defined so two people scoring the same vendor would land in the same place. The weights reflect a broad evaluation: a CX or support leader, usually already burned by an incumbent AI or facing a renewal, who wants a list to trust rather than a sales page.

Resolution vs. deflection (weight: high)

Does the platform resolve end to end autonomously, draft replies for a human to approve, or only deflect to help-center content? Operational test: of 100 inbound tickets, how many are closed with no human touch and a confirmed outcome, not just "the customer stopped replying"? Score on actions executed, not on "AI-powered" claims.

Action execution depth (weight: high)

Can it execute real operations (refunds, cancellations, order or address edits, subscription changes) as validated, policy-bounded tool calls, or does it only generate text? This is the line between a reply and a resolution. Count the write-actions a vendor can prove on your stack, not the ones the marketing page implies.

Eval ownership and QA governance (weight: high)

Who authors the test cases the AI is held to, and who can see a regression? Will the vendor run the AI against a sample of your historical tickets and show per-response accuracy before go-live? And once live, who reviews quality: a sampled QA process, or every conversation? CX-team-authored evals with every regression visible beat a vendor dashboard score you cannot inspect. We cover the failure-mode side of this in AI hallucination defense.

Pricing model (weight: high)

One bill, a per-resolution double-meter on top of seats or tickets, or outcome-based? The question is not "cheapest" but "do the vendor's incentives match mine, and is AI a second meter?" Flag the double-meter where it exists (Pattern A), recognize outcome pricing as a different model (Pattern B), and model your real monthly volume against each before deciding.

Time to value (weight: medium)

How long from signature to the AI resolving real tickets? Look for proof during the evaluation (an agent built live on your data) and a deployment measured in weeks, not a multi-quarter services engagement. Anchor against published numbers: Decagon cites roughly six weeks, Sierra four to ten weeks in its own case studies.^[5]

Maturity and references at scale (weight: medium, rising with deal size)

Above roughly $30K a year, a buyer is managing career risk, and "is this proven?" outweighs "does it work?" Named references at your scale, audit reports (SOC 2, HIPAA-audited, GDPR), uptime, and funding all read as durability. This is the axis where the well-funded agent layers are strongest, and where a younger vendor has to compensate with proof. Weigh it honestly against your own risk tolerance.

Channel coverage and voice (weight: medium)

Email, chat, social, SMS, and voice in one inbox with shared context, or a chat widget with bolt-ons? Note whether voice is native or integrated: several platforms here, including Richpanel, integrate voice through Aircall, Dialpad, or JustCall rather than hosting it. If voice must be a single native pane, that narrows the field hard.

Criteria 1 through 4 carry the most weight because they separate a platform that resolves and acts from one that deflects and meters. Criteria 5 through 7 decide fit, risk, and economics once a platform can actually resolve. Maturity is weighted medium in general but climbs to decisive past roughly $30K a year, which is exactly why it gets its own row rather than a footnote.

The comparison, as of June 2026

Seven platforms, five comparable axes.

Cells reflect each vendor's public product, pricing, and funding pages as of June 2026. Each platform name links to the page used to source its row. Pricing and valuations are volatile in this category, so verify the cells that decide your choice directly with the vendor. Where a capability is real but not separately documented, the cell says so rather than guessing.

Platform	Architecture	Resolution and actions	AI pricing model	Maturity signal	Channels and voice
Richpanel	AI-native helpdesk (agent + helpdesk in one)	Autonomous resolution or collaborative draft; typed, policy-bounded actions	One bill, ~$0.30/conversation, no separate AI meter	2,000+ brands, founded 2020; SOC 2, HIPAA-audited, GDPR	Email, chat, social, SMS; voice via Aircall/Dialpad/JustCall
Intercom Fin	Pattern A: AI on Intercom (rebranded Fin, 2026)	Autonomous resolution over knowledge plus structured Actions	$0.99/resolution on top of $29–132 seats	Large install base; widely-cited resolution benchmark	Chat, email, in-app strong; social via add-ons
Decagon	Pattern B: standalone agent layer	Autonomous resolution with actions; self-serve eval tooling	Outcome-based, contact sales (reported ~$95K–$590K+)	$4.5B-funded with large named enterprise logos	Rides the host helpdesk's channels
Sierra	Pattern B: standalone agent layer	Autonomous resolution; evals vendor-authored, customer-reviewed	Outcome-based (~$150K floor reported)	$15.8B valuation; mid-market access	Rides the host helpdesk's channels
Zendesk AI	Pattern A: AI on a mature enterprise helpdesk	Deflection-leaning; real-world ~10–20% autonomous	Add-on stack ~$215/seat plus per-resolution AI on top	Broadest, most established platform (since 2007)	Broadest channels; Talk voice is a resold Aircall partner
Gorgias AI	Pattern A: ecom helpdesk + AI Agent	AI Agent answers and some actions; deep native Shopify	Double-meter: ticket fee + ~$0.90/AI resolution	Category leader in DTC ecommerce	Email, chat, social with deep ecom app ecosystem
Ada	Pattern B: high-volume agent layer	Autonomous resolution tuned for very high volume	Outcome/volume-based, contact sales	Enterprise scale; ~300K conversations/yr floor	Chat-first across digital channels

Pricing, funding, and product facts verified against each vendor's own pages and public reporting as of June 2026. If your reading of any cell differs from current reality, email amit@richpanel.com and we will update it. The goal is to be accurate, not to win a column we have not earned.

The honest read of this table: autonomous resolution and action execution are no longer rare; the agent layers and the AI-native cut all do both. The real separation is in three columns. The pricing column splits the bolt-on double-meter from outcome pricing from a single per-conversation bill. The maturity column is where Decagon and Sierra lead outright, on funding and named logos. And the architecture column decides whether you are adding a second system or replacing one. Richpanel does not top the maturity column, and on this list that is the column that most often decides a large deal.

Where each one wins

The strength I would actually send a buyer toward.

For each of the other six platforms, here is a specific situation where it is the better choice than Richpanel. If your situation matches, take it seriously.

Intercom Fin

Fin has one of the largest install bases of any agent on this list, inherits Intercom's enterprise credibility, and resolves over knowledge plus structured actions. Choose Intercom Fin over Richpanel if you already run Intercom for chat and in-app product messaging. Adding Fin is then the lowest-friction path to autonomous resolution, with no platform switch and a team that already knows the UI. If your buying committee includes a CTO who wants the most-deployed, analyst-recognized option (a real concern we have lost deals to), Fin's maturity is a legitimate advantage. The cost to weigh: Fin is $0.99 per resolution on top of $29 to $132 seats as of June 2026, the per-resolution meter we describe above. For the switch-off cut, see best Intercom Fin alternatives.

Decagon

Decagon is one of the most heavily funded agent layers in the category and carries large named enterprise references, which matters more than any feature once a deal crosses roughly $30K a year. Choose Decagon over Richpanel if your single most important criterion is a proven, big-logo agent and you intend to keep your existing helpdesk rather than replace it. This is the honest one for us: a buyer managing career risk on a customer-facing system often weighs "who else my size runs this" above resolution mechanics, and on funding and marquee logos Decagon is ahead of Richpanel today. Decagon also ships self-serve eval tooling, so this is not a case where evals are hidden. The trade-offs to weigh are the two-system footprint (the agent rides your incumbent helpdesk) and contact-sales, outcome-based pricing reported in the $95K-plus range. If Decagon is on your shortlist, our Richpanel vs. Decagon comparison lays out the differences side by side.

Sierra

Sierra is a similarly well-capitalized agent layer with a high valuation and genuine mid-market access, and its outcome-based pricing aligns the vendor's incentive with resolutions rather than seats. Choose Sierra over Richpanel if you want a pure agent layer from an enterprise-credible vendor, you are keeping your helpdesk, and outcome pricing fits how your finance team wants to buy AI. Like Decagon, Sierra leads Richpanel on funding and brand maturity, the signal that closes the largest deals. One nuance worth getting right in your own diligence: Sierra's evals are vendor-authored and customer-reviewed, which is collaborative but not the same as your team authoring every test case, so ask exactly who owns the eval set. Side-by-side detail in our Richpanel vs. Sierra comparison.

Zendesk AI

Zendesk is the broadest, most established platform on this list, the default since 2007, and the safe choice when a committee is managing risk across many departments. Choose Zendesk over Richpanel if you are a large organization standardizing one vendor across CX, IT service, and internal help desks, and you value that breadth and governance maturity over AI-native resolution depth. Autonomous AI is now bundled into Zendesk plans, so the old "Zendesk has no AI" jab is stale; the honest read is that its native AI leans toward deflection and real-world autonomous resolution tends to land around 10 to 20 percent. The other trade-off is cost: the add-on stack runs roughly $215 per seat once you add Copilot and QA, plus per-resolution AI on top, the cost-fatigue story we hear most from Zendesk switchers. If you are on Zendesk, see our Richpanel vs. Zendesk comparison and the broader best customer service software shortlist.

Gorgias AI

Gorgias has the widest ecommerce app marketplace of anyone here, built specifically for DTC operations. Choose Gorgias over Richpanel if you want the maximum number of native Shopify-ecosystem integrations in one place and you are comfortable with AI that leans toward assist today. The flip side, and the reason it shows up so often in our switch conversations, is the AI maturity plus the double-meter: across our 69 demos, Gorgias was the single most-cited incumbent prospects were leaving, usually citing AI answer quality and a ticket-fee-plus-roughly-$0.90-per-AI-resolution bill as of June 2026. On raw ecom app-marketplace breadth, though, it is the leader. If Gorgias is your incumbent, see our Richpanel vs. Gorgias comparison and the Shopify-specific cut in best AI customer service software for Shopify.

Ada

Ada is built for very high-volume deflection and resolution at enterprise scale, with a mature automation engine. Choose Ada over Richpanel if you are a large enterprise pushing hundreds of thousands of conversations a year through digital channels and your priority is automation throughput at that tier. Ada's reported floor of roughly 300K conversations a year puts it out of range for most SMB and mid-market teams, but at the top of the volume curve that scale focus is a genuine fit Richpanel does not target.

Richpanel

For completeness, here is where we are the right answer, stated as plainly as the others. Choose Richpanel if you want the repeat work resolved autonomously on one bill, flat per-conversation economics (around $0.30 per conversation, with the model and token budget you choose, detailed on our pricing page) instead of seats plus a per-resolution meter, a CX Manager AI that reads your business and sets up the agents rather than a vendor services team doing it for you, evals your own CX team authors with every regression visible, a QA AI that reviews every closed conversation, the AI built live on your own data and proven before go-live, multi-brand support in one workspace, and a 50% resolution guarantee in 30 days with your money back if it misses. The CX leader keeps their team and scales output without scaling headcount; the AI absorbs the boring volume. In production that has looked like a wellness brand whose AI sends 60% of every customer message at a higher CSAT than its human team (4.43 versus 4.25).^[6] Where we are weaker than the field: Decagon and Sierra out-fund us and carry larger named enterprise logos, so if "most-established, biggest-reference agent" is your top criterion, weigh that seriously; we do not host native voice (we integrate with Aircall, Dialpad, and JustCall); and if you want to keep your existing helpdesk and only bolt on an agent, a Pattern-B layer fits that shape better than an all-in-one.

A decision tree, not a verdict

Match your situation to the shortlist.

There is no single best AI customer service software. There is a best one for what you want to keep, your scale, your risk tolerance, and how you want to pay for AI. Map yourself to a line below.

Committed to your current helpdesk, want a proven big-logo agent bolted on, budget is enterprise: Decagon or Sierra, then Intercom Fin if you are already on Intercom.
Already deep in Intercom for in-app messaging, lowest-friction add: Intercom Fin.
Replacing an incumbent AI that underdelivered, want the helpdesk and agent on one bill with proof on your own data and a guarantee: Richpanel, then a Pattern-B layer as the comparison.
Shopify or DTC ecommerce, want the widest native ecom app marketplace, comfortable with assist-leaning AI: Gorgias AI, then Richpanel if AI resolution and one-bill economics matter more than app-store breadth.
Enterprise standardizing one platform across many departments, breadth and governance over AI depth: Zendesk AI.
Very high-volume enterprise deflection, hundreds of thousands of conversations a year: Ada.
Above roughly $30K a year and "is this proven at my scale?" is your gating question: lead the eval with references and audit reports from every vendor, not feature demos, and weigh the funded agent layers accordingly.

Notice that the right answer flips on facts about you, not on which vendor wrote the article. That is what a real comparison is supposed to do. If a single name appeared on every line, you would be reading marketing again.

Run these on every demo

Six tests that cut through the pitch.

Whichever shortlist you land on, these six tests separate platforms that resolve from platforms that demo well. For the full version, see our 40-question vendor RFP template.

1. Run the AI on 100 of my historical tickets.

Ask for per-response accuracy and a walk-through of the failures. A vendor that will not do this is selling a demo, not production.

2. Is your headline rate deflection or confirmed resolution?

If they conflate the two, or cannot define the difference, the number is marketing.

3. Show me the total bill at my real volume, AI included.

For a bolt-on, make them add the per-resolution AI meter to the seat or ticket price. For an outcome-based agent, pin down the price per resolution and the volume tiers.

4. Show me an action executed as a tool call, not free text.

Ask to see the typed parameters and the policy constraints on a refund or cancellation. Free-text "I'll refund you" with no validation layer is how a bot sells a car for one dollar.

5. Who authors the evals, and who reviews quality once live?

Can your team write the test cases and see every regression, or is it a vendor score you cannot inspect? Sampled QA, or every conversation reviewed?

6. Connect me with three customers my size, live within your timeframe.

Above roughly $30K a year, named references at your scale and audit reports (SOC 2, and HIPAA if you handle PHI) are a more reliable signal than any ranking, including this one.

How this comparison is limited

What this guide cannot tell you.

An honest comparison names its own blind spots. Three apply here.

It is written by a participant. Richpanel is in the matrix, and no amount of even-handedness fully removes that. The mitigation is structural: criteria defined before scoring, a named competitor strength on every row, an architecture and maturity column where we do not lead, linked source pages, and a public correction address. Read it as a more useful starting point than a vendor listicle, not as a neutral analyst report.
Pricing and funding move faster than any page. Every cell is a snapshot of public documentation as of June 2026. This category re-prices, re-bundles, and re-raises constantly. A per-resolution number or a valuation quoted here may have changed by the time you read it, so verify the cells that decide your choice directly with the vendor.
The matrix omits valid platforms. Freshdesk, Help Scout, Gladly, Kustomer, Salesforce Agentforce, Lorikeet, and others are real options we did not score here, chosen to keep a seven-row comparison legible and to cover both architectures cleanly. Several appear in our broader best customer service software shortlist and, for the helpdesk-led cut, our best AI helpdesk software shortlist. Their absence here is editorial, not a judgment.

The claim this guide is willing to stand behind is narrow and defensible: in 2026 the deciding questions are which architecture you want, whether a platform resolves rather than deflects, and what AI actually costs under its billing model. A vendor's willingness to prove accuracy on your own data before go-live, and to quote the all-in cost at your real volume, are the two most predictive signals you can test before signing.

Frequently asked

Choosing an AI platform, in plain English.

What is the difference between AI customer service software and an AI customer service agent?

The phrases are used interchangeably, but the products split two ways. AI bolted onto a helpdesk (Zendesk, Gorgias, Intercom Fin, Freshdesk) adds an AI layer to a ticketing system you already run, usually metered separately. A standalone AI agent layer (Decagon, Sierra, Ada) is the agent on its own, sitting on someone else's helpdesk you keep paying for. An AI-native platform like Richpanel rebuilds the helpdesk around the agent, so one platform resolves the repeat work and your people own the exceptions on one bill. Search "ai customer service software" and "ai customer service agents" and you reach the same shortlist, so score on resolution depth and metering, not on which phrase you typed.

Which AI customer service software actually resolves tickets instead of just deflecting?

Resolution means the customer's problem is solved, the action it required was taken (a refund, a cancellation, an address change), and the outcome is confirmed. Deflection only means the customer stopped asking, which can mean they gave up. As of June 2026, the agent layers (Decagon, Sierra, Ada) and the AI-native platforms (Richpanel) are built to resolve and take actions end to end; legacy helpdesk AI like native Zendesk AI leans toward deflection, and real-world autonomous resolution there tends to land around 10 to 20 percent. The test that cuts through it: ask any vendor to run its AI on 100 of your historical tickets and report confirmed resolution, not deflection.

What is the AI double-meter, and which platforms charge it?

A double-meter is when a helpdesk charges you for the seat or the ticket, and then charges again per AI resolution on top. As of June 2026 it applies to the bolt-on helpdesks: Gorgias adds roughly $0.90 per AI resolution on top of its ticket plan, Zendesk stacks add-ons to roughly $215 per seat plus per-resolution AI, and Intercom Fin adds $0.99 per resolution on top of $29 to $132 seats. Standalone agent layers like Decagon and Sierra use outcome-based pricing instead, which is a different model, not a double-meter. Richpanel charges once, roughly $0.30 per conversation with a model and token budget you choose.

Is Decagon or Sierra better than Richpanel?

Decagon and Sierra are heavily funded standalone agent layers with large named enterprise references, and above roughly $30K a year that maturity signal is real: a buyer managing career risk often weighs proven references at scale over feature depth. If your top criterion is the most-established, biggest-logo agent and you intend to keep your existing helpdesk, a Pattern-B layer fits that shape. Richpanel wins when you want the helpdesk and the agent rebuilt as one platform on one bill, the AI set up by a CX Manager AI rather than a vendor services team, evals authored by your own CX team with every regression visible, and a 50% resolution guarantee in 30 days or your money back. Different shapes for different buyers.

Should I buy AI-native customer service software or bolt an AI agent onto my existing helpdesk?

If you are committed to your current helpdesk and only want to add autonomous resolution, a standalone agent layer (Decagon, Sierra, Ada) or your incumbent's own AI add-on is the lower-friction path, with the trade-off of a second bill and an agent constrained by the host platform's data model. If your incumbent AI has not actually moved your resolution rate, or you are consolidating helpdesk plus a bolt-on AI plus returns into one tool, an AI-native platform like Richpanel resolves on one bill. The deciding question is usually whether the AI you already pay for has earned its place.

Sources & references

Where the claims come from.

Inline citations [1]–[6] map to the entries below. Vendor product, pricing, and funding pages used to source matrix rows are linked inline in the table.

Fin (Intercom), "I Reviewed the Best AI Chatbots for Customer Support." Vendor-authored ranking placing Fin at number one. Cited as an example of the self-ranking pattern. fin.ai/learn/best-ai-chatbots-customer-support
Gorgias, "Gorgias vs Richpanel." Vendor-authored comparison in which Gorgias wins. Cited as an example of a competitor publishing its own scorecard. gorgias.com/comparison/richpanel
Real-world autonomous-resolution range for legacy helpdesk AI. Native Zendesk AI leans toward deflection to help-center content; observed real-world autonomous resolution tends to land around 10 to 20 percent, well below headline market claims of 50 to 80 percent. Drawn from the Richpanel buyer demo dataset (entry 5) and Zendesk's own product positioning (zendesk.com).
Competitor AI pricing, verified June 2026. Per-resolution AI charges referenced in the pricing column and fault-line section: Gorgias ~$0.90/AI resolution on top of ticket fees (gorgias.com/pricing); Intercom Fin $0.99/resolution on top of $29–132 seats (intercom.com/fin); Zendesk add-on stack ~$215/seat plus per-resolution AI (zendesk.com/pricing). Decagon and Sierra use outcome-based pricing, a different model. Figures are point-in-time and change frequently.
Time-to-value and funding anchors. Decagon publishes roughly six-week implementations and is reported at $4.5B in funding; Sierra cites four-to-ten-week deployments in its own case studies and is reported at a $15.8B valuation. Used in the time-to-value and maturity criteria. decagon.ai / sierra.ai
Richpanel production case study (wellness brand) and buyer demo dataset. The brand's AI sends 60% of every customer message at a higher CSAT than its human team (4.43 versus 4.25), fully autonomous, with a median first response of 28 seconds (richpanel.com/case-studies/wellness). The "Gorgias most-cited incumbent" and "25-plus prospects said their incumbent AI did not work" observations are drawn from 69 recorded inbound demo calls, April to May 2026; underlying call data is confidential, aggregate counts are publishable. Methodology available on request via amit@richpanel.com.

Version history, v1.0 (2026-06-09): initial publication. Matrix cells are a snapshot of public vendor documentation, pricing, and funding as of this date.

"AI customer service software" now means two different architectures. The right one depends on what you keep.