AI Now Costs More Than the Helpdesk in 4 Out of 5 Ecommerce Support Scenarios

What does AI customer service pricing actually look like when you model it across real ecommerce scenarios? We checked seven vendors, forty scenarios, and found the answer surprised us.

↓ Download the full scenario grid spreadsheet with all 40 scenarios, raw data, and methodology.

Something quiet happened in AI customer support pricing last year. Three major platforms independently arrived at almost the same price point for AI resolutions. Gorgias: $1.00 per resolved conversation. Intercom: $0.99 per resolution. HubSpot, once you decode their credit system: $1.00 per conversation.

A dollar. Give or take a penny.

Three different companies, three different architectures, three different customer bases. Same number. That’s not a coincidence. That’s the market settling on what it thinks AI support is worth. And a recent Gartner report predicts that number will exceed $3 by 2030 as vendor subsidies give way to real economics. Whether that prediction is right depends on something most people in this industry aren’t talking about yet. We’ll come back to it.

What we wanted to know was simpler. What do ecommerce brands actually pay, right now, when they turn on AI support?

So we built a spreadsheet. 40 scenarios. Seven vendors. Five ticket volumes, four team sizes, AI on and AI off for each one. Every price pulled from public pricing pages on 24 February 2026, with screenshots. Then we split every vendor’s cost into two columns: what you pay for the helpdesk platform, and what you pay for the AI.

That split is where it got interesting.

In 79% of scenarios we modelled, the AI automation line item was larger than the helpdesk base cost. Not by a little. At 2,000 tickets a month with a single agent, AI represents 77% to 95% of the total bill depending on vendor. At 5,000 tickets, it’s 77% to 98%.

Most ecommerce teams think they’re paying for a helpdesk that happens to have AI bolted on. The data says the opposite. They’re paying for AI that happens to need a helpdesk underneath.

How We Calculated This #

Every figure in this piece comes from publicly listed pricing as of 24 February 2026. Gorgias, Intercom, Zendesk, Tidio/Lyro, Freshdesk, HubSpot, and Hay. All collected on the same day, monthly billing rates, no negotiated discounts or annual deals. Timestamped screenshots archived for all of them. If a vendor changes their pricing tomorrow, the screenshots show what it was today.

Salesforce was excluded because they don’t publish AI agent pricing publicly. You have to get on a call. That’s a data point in itself. We applied a default 50% AI resolution rate across all scenarios, which is conservative for mid-market ecommerce. Your actual rate will vary. A fresh deployment might resolve 20% of tickets in month one. A mature, well-trained one could hit 70%. And in practice, the rate depends less on which platform you pick and more on how well it’s been set up, the quality of your instructions, playbooks, and knowledge base coverage. The accompanying spreadsheet lets you change the resolution rate assumption and watch every vendor’s costs recalculate in real time, from 40% to 80%. This matters more than most people realise, because on per-resolution pricing, your bill scales directly with how good your AI gets. Better AI, higher cost. We’ll come back to that.

Freshdesk, Tidio, and Hay price natively in EUR. All EUR amounts in this piece have been converted to USD at the 24 February 2026 mid-market rate of €1 = $1.18. Cells marked with an asterisk (*) contain converted figures. No FX margin has been applied.

They Don’t Even Measure the Same Thing #

Before the numbers though, here’s something that surprised us. These platforms don’t just charge different amounts. They charge for different things. A Freshdesk “session” isn’t an Intercom “resolution” isn’t a Zendesk “automated resolution.” Comparing them without understanding this is like comparing rent prices across cities without checking whether they include utilities.

Source: Vendor pricing pages, 24 February 2026.

The session-versus-resolution distinction deserves a paragraph of its own. Freshdesk charges $0.53 per session.* A session is any customer interaction in a 24-hour window, regardless of outcome. If the AI tries and fails, you still pay. At a 60% resolution rate, your effective cost per actual resolution is $0.88. At 30%, it’s $1.77. I’ve talked to operators who didn’t discover this until they reconciled their first invoice.

What the Numbers Actually Show #

Enough context. Here are the tables. We’ve broken them out by ticket volume so you can find your approximate situation without wading through forty rows. Each cell shows helpdesk base + AI cost = total monthly spend. Hay is AI-only, so there’s no base component.

One thing to keep in mind as you scan these: the AI resolution rate is set at 50% as a conservative baseline. Vendors quote rates anywhere from 40% to 80%, and the real number depends on implementation quality more than platform choice. If you want to see how these tables shift at higher resolution rates, the full spreadsheet lets you change that assumption and every cost recalculates automatically. The base stays the same either way. That’s the whole point of splitting them.

300 Tickets/Month #

1,000 Tickets/Month #

2,000 Tickets/Month #

5,000 Tickets/Month #

10,000 Tickets/Month #

TTS = Talk to Sales. Gorgias requires a custom plan above 5,000 tickets. Freshdesk includes 500 free AI sessions on Pro/Enterprise plans, which is why the AI cost is zero at 300 tickets. Hay is AI-only and requires an existing helpdesk underneath.

* Freshdesk, Tidio, and Hay price natively in EUR. Converted at €1 = $1.18 (mid-market rate, 24 February 2026). Native EUR prices: Freshdesk Freddy AI sessions €0.45/session. Hay Solo €50, Starter €135, Growth €725. Tidio Core €32.50, Plus €749. Your actual USD cost will vary with the exchange rate at time of billing.

Find Yourself in the Data #

If you’re a 5-agent Shopify team at 2,000 tickets a month #

Probably the most common profile we see in growing DTC brands. Your options range from $1,254 to $2,325 for the same workload. That’s over a $1,000 monthly gap. And here’s the part most teams miss: on Intercom, 67% of your bill is AI. On Zendesk, 75%. On Freshdesk, 80%. The helpdesk is the smaller number on every single invoice.

If you’re running lean: 1 agent, 1,000 tickets #

Solo founder or one dedicated support person. Efficient. Needs AI to stay that way. On Intercom, your $99 seat is a rounding error next to $495 in AI costs. That’s 83% of your bill. On Zendesk it’s 87%. I genuinely didn’t expect the ratio to be that stark. We double-checked it.

If you’re scaling: 3 agents, 5,000 tickets #

This is where the numbers get serious. Monthly cost ranges from $2,772 to $4,245 depending on platform. Every vendor’s AI cost exceeds $2,400. The helpdesk base across all of them is between $147 and $345. Think about that for a second. The platform that routes your tickets, manages your agents, runs your reporting? It’s 6% to 13% of the bill. Everything else is AI.

If you just want a chatbot at 300 tickets #

Freshdesk at $49 total. Not close. Their 500 free sessions cover your entire volume with room to spare. We’re being straightforward about this because there’s no point pretending otherwise. At low volume, Freshdesk’s free tier is the rational choice. But keep your eye on what happens at 501 sessions, because the free tier has a hard ceiling and there’s no gradual ramp after it.

Where the Hidden Costs Are #

The tables give you the totals. This section is about the details that don’t show up until you’re already committed.

Gorgias: The Double-Billing Detail #

Most people assume that when AI resolves a ticket on Gorgias, you pay the AI fee instead of the ticket fee. Reasonable assumption. Wrong. You pay both. An AI-resolved conversation is still a billable helpdesk ticket AND incurs the $1.00 automation fee. Gorgias confirmed this in their billing documentation. Their pricing page is clear about the per-resolution rate, which is more than most vendors offer. But this specific interaction between the two billing layers is the kind of thing you only discover when you reconcile your first month.

Zendesk: Three Pricing Dimensions at Once #

Zendesk has been the industry standard for nearly two decades. There’s a reason for that. But their AI pricing requires you to think in three dimensions simultaneously: agent seats ($115/month on Suite Professional), the Advanced AI add-on ($50/agent/month), and then per-resolution charges ($1.50 committed, $2.00 pay-as-you-go). A five-agent team pays $250/month just for the AI add-on before a single resolution. And that’s before Copilot ($50/agent) or Workforce Management ($50/agent) if you need those too. We’ve talked to operators who were surprised when they modelled this out. The base plan that looked like $115/agent turned out to be closer to $265/agent in practice.

Freshdesk: Sessions Aren’t Resolutions #

Worth repeating because it changes the maths entirely. Freshdesk charges $0.53 per session.* A session is any customer interaction, resolved or not. If a customer opens a chat, the AI attempts to help, the customer abandons, and opens a new chat an hour later, that could be two sessions for zero resolutions. At $0.53 per session, your effective cost per successful resolution depends entirely on your AI’s success rate. Their free tier at low volume is generous, no question. But the unit economics change once you’re past 500 sessions, and the session-versus-resolution distinction matters more than any headline rate.

HubSpot: The Credit System #

HubSpot’s AI pricing isn’t on the Service Hub pricing page. That surprised us. You find it in a separate knowledge base article about their credit system. Each AI conversation costs 100 credits. Credits cost $0.01 each. So: $1.00 per conversation. Straightforward once you find it. The wrinkle: those credits are shared across every Breeze AI feature. Marketing, sales, service, content. If your marketing team is burning through credits on AI-generated campaigns, your support AI competes for the same pool. It’s an ambitious system that lets HubSpot price AI usage across their entire platform. The trade-off is that modelling your support costs in isolation requires knowing what the rest of your organisation is doing with credits.

Tidio: The Cliff #

Tidio’s Core plan is $38/month.* Fifty AI conversations included. Good entry point for a small shop testing the waters.

Need 51 conversations? The next tier is Plus. $884/month.*

That’s a 23x increase. For one extra conversation. There is no middle tier. We checked. For a seasonal ecommerce brand that does 40 AI conversations in January and 80 in November, this creates a genuine planning problem. You’re either well under the Core ceiling or you’re jumping to a plan that costs more than most competitors charge for thousands of conversations.

A Note on Transparency #

We didn’t set out to rank these vendors. We just wanted to collect the prices. But the experience of collecting them told its own story.

Some vendors took five minutes. Price on the pricing page. Calculator in your head. Done. Others required reading knowledge base articles, decoding credit systems, adding up add-ons, and in one case, realising we couldn’t get a price at all without booking a sales call.

When three platforms independently converge on $1/resolution, and Gartner predicts that figure will triple within four years, you’d assume the underlying technology must be expensive. It isn’t. Not any more.

Epoch AI found that LLM inference prices have been falling at a median rate of 50x per year, accelerating to 200x per year since January 2024. The Stanford AI Index Report documented a 280-fold cost reduction for GPT-3.5-level performance in roughly eighteen months. A customer service conversation that cost a vendor several dollars in raw compute two years ago now costs pennies. The underlying technology is getting radically cheaper every quarter. But the price you pay stays at a dollar. Or goes up.

That’s because these platforms don’t price against compute. They price against the cost of a human agent. If a human resolution costs $5–10, then $0.99 feels like a bargain regardless of what it costs the vendor to deliver it. Intercom’s VP of Engineering said exactly this in a Stripe case study: they chose outcome-based pricing specifically so the conversation wouldn’t be about how low the price could go. That’s not a secret. It’s their stated strategy.

So Gartner’s $3 prediction starts to make more sense. Read the fine print and the analyst behind it, Patrick Quinlan, told CX Today that the per-unit consumption cost “is often very low.” The $3 figure bundles everything around the AI: implementation, data preparation, specialist talent, maintenance, and vendor margins. Sabio Group’s Head of R&D called it “not a technology failure” but “a partnership failure,” pointing to consumption-based pricing that sounds cheap until the invoice arrives.

None of this means these vendors are doing anything wrong, exactly. Pricing against the value you deliver rather than the cost of delivery is standard practice. But it does mean the gap between what AI costs to run and what you pay for it is widening, not narrowing. And when 79% of the scenarios above show AI costing more than the helpdesk itself, the question stops being “which vendor is cheapest” and becomes “which vendor lets you see what you’re actually paying for.”

Where Hay Fits #

Disclosure, obviously: we built Hay. It’s in this study at the same level of scrutiny as every other vendor. Hay is an AI automation layer, not a helpdesk. It requires an existing platform underneath. In the AI-only comparison, Hay is the cheapest option in 16 of 20 scenarios. Freshdesk wins the other four, all at 300 tickets where their free sessions cover everything. At 1,000+ tickets, Hay’s bundled pricing saves 57–85% versus the next cheapest alternative at the default 50% resolution rate. The exact savings shift depending on the rate you model – the spreadsheet lets you test this yourself. The numbers are in the tables. Draw your own conclusions.

The Question This Data Leaves Open #

Split your bill. That’s the practical takeaway. Separate your helpdesk base from your AI cost. Most vendors don’t make this easy, which is why we did it across 40 scenarios.

But the bigger question is one this data can’t answer yet.

Today’s $1/resolution consensus price has three possible futures. It stays roughly where it is because competitive pressure holds it down. It drops, because the cost of running AI keeps falling and new entrants price closer to compute. Or it climbs toward Gartner’s predicted $3+. Not because AI gets more expensive to run, but because the pricing model stays anchored to what a human agent costs, and the margin quietly widens.

If you’re building your support budget around $1/resolution, that’s fine for 2026. But the teams that come out ahead in 2027 and 2028 will be the ones who modelled all three scenarios before they had to.

The full scenario grid, methodology statement, and pricing screenshots are published alongside this piece. Every number is verifiable.

Methodology Statement #

All pricing collected 24 February 2026 from public vendor pricing pages. Screenshots archived. Monthly billing rates throughout. Freshdesk, Tidio, and Hay price in EUR; all EUR figures converted to USD at €1 = $1.18 (mid-market rate, 24 February 2026) and marked with asterisk (*). AI resolution rate: 50% of total ticket volume (default; the accompanying spreadsheet allows this to be adjusted from 40–80% to model different scenarios). Freshdesk sessions = tickets × 1.2. Zendesk Suite Professional ($115/agent) as base; Advanced AI add-on ($50/agent) + $1.50/AR committed rate. Intercom Advanced plan ($99/seat). HubSpot Service Hub Professional ($100/seat). Gorgias plan tiers per published pricing. Tidio Core ($38*/mo) and Plus ($884*/mo) as listed. Hay pricing per hay.chat/pricing (Solo $59*, Starter $159*, Growth $856*). Salesforce excluded: does not publish AI agent pricing publicly. Amazon Connect included as sidebar in full dataset only (infrastructure service, not a finished helpdesk). Full methodology, assumptions, and raw data published in the accompanying spreadsheet.