Only 13 of 25 AI customer service products could be budgeted from public pricing and billing terms in an EPR Research audit on November 21, 2026. For those 13, the specified configuration could be priced at 1,000, 5,000 and 10,000 monthly support interactions. The count measures whether public terms allow a price, not product quality, and each estimate prices a different meter, so totals are not like-for-like.
Last updated: November 21, 2026
Summary: what did the audit find?
The EPR Research audit found that 13 of 25 AI customer service agent products (52%) could be priced at 1,000, 5,000 and 10,000 monthly interactions for the configuration specified in each row. Salesforce Agentforce and Front Autopilot supported usage-only estimates, CustomGPT.ai supported estimates at two volumes, and nine products lacked enough verified terms for any estimate.
Public documentation identified a clear charge trigger for 20 of the 25 products (80%). Seven of those 20 products still lacked full estimates because base subscription fees, tier thresholds, allowances or overage terms remained unresolved.
How did EPR Research audit the 25 products?
EPR Research designed and conducted this audit, fixing a purposive panel of 25 products, one per vendor, before collecting and scoring any vendor terms. The panel covers customer-facing text support agents and configurable products with a documented support use and an identifiable public price model or component.
EPR Research searched everything-pr.com on November 21, 2026 and found no earlier pricing audit of AI customer service agents. Ada's platform page and Sierra's product page were logged separately as quote-only checks. Ada describes conversation-based pricing and Sierra describes outcome-based pricing, and neither publishes a verified numeric tariff.
The test scenarios use steady-state web-text support with 1,000, 5,000 and 10,000 inbound interactions, one seat where the configuration allowed it, and a 50% base resolution rate for products that bill per outcome. The workbook also models 30% and 70% resolution cases.
Scenario assumptions and exclusions
Scenario assumptions applied published pack sizes as listed, stating message, turn and handoff parameters explicitly. Configurations stay consistent across volumes except where a specified tier changes. Estimates exclude tax, human labor, trial allowances, optional integrations and negotiated discounts, while one-time onboarding fees appear separately.
A charge trigger was scored as clear when public materials identified the charge or allowance event and at least one rule that changed the count, such as a handoff, a time window, a model multiplier, a minimum message count or the treatment of a failed attempt. A public historical tariff was not enough to model a current invoice, so EPR Research used each vendor's current billing documentation where it separated current and legacy terms.
The sample describes the products audited. It does not estimate the share of all vendors in the market, and the comparison does not rank product quality or establish the lowest possible quote.
The same approach of a fixed panel and dated public evidence runs through other EPR Research audits, including the corporate media kit audit and the AI launch availability study.
What counts as a billable interaction?
A billable interaction is defined by each vendor's meter, where a single support exchange can register as a resolution, a session, several replies or a bundle of credits. At 5,000 monthly interactions, raising the assumed resolution rate from 30% to 70% lifts the modeled Intercom Fin bill from $2,009 to $3,989, with separately billable procedure handoffs held at 10%, according to Intercom's plan terms.
Freshdesk Freddy stays at $2,479 across the same resolution rates because Freshdesk bills per chat session. The modeled Freshdesk Freddy bill rises to $2,969 if one in five cases needs a second billable chat session, according to the Freshdesk Freddy add-on rules.
Do handoffs to a human carry a charge?
Handoffs to a human carry a charge at some vendors. Front lists $0.39 when Autopilot performs work intended for human takeover and $0.89 for a resolution, plus a $0.05 triage tier, according to Front's billing documentation. Front applies the highest tier reached once per conversation.
Intercom can charge $0.99 when Fin completes a configured procedure that ends in a handoff. Re:amaze excludes human escalations from its AI resolution count, according to its resolution billing article.
Does a reopened conversation count twice?
A reopened conversation can count as a second billable interaction at some vendors. HubSpot can count a further resolution when an email conversation resumes after its 72-hour window and the AI resolves another issue, according to the HubSpot Customer Agent guide.
Zendesk tier documentation sets a 72-hour messaging and email window from the first message, distinguishing verified success from containment. Under the Zendesk resolution tier terms, containment without verification and assisted escalation do not draw down the current allowance.
Which costs sit outside the starting price?
Four kinds of cost commonly sit outside the advertised starting price: the helpdesk seat, a second fee on the same case, onboarding and commitment terms, and overage blocks or a second meter.
The helpdesk or operator seat
Helpdesk and operator seats frequently require an underlying subscription before an AI customer service agent can be deployed. Intercom's $0.99 Fin outcome price sits alongside the seat on an integrated plan, though Fin can also run on an existing helpdesk with a minimum monthly outcome commitment. Freshdesk's $49 session packs sit on top of the eligible Freshdesk Omni subscription for web chat.
Two fees on one case
Some vendors charge two fees on one case by billing both a helpdesk ticket fee and an automated resolution fee. Gorgias can charge for a helpdesk ticket containing an outbound reply and then add the AI fee if the AI fully automates the interaction. A human escalation can still generate the helpdesk ticket fee, according to Gorgias billing documentation.
Onboarding and commitment
Mandatory onboarding fees and annual commitments can substantially raise initial deployment expenses. HubSpot lists a $1,500 required one-time Professional onboarding fee, and Service Hub Professional carries an annual commitment when paid monthly, according to HubSpot Service Hub pricing. Front requires an onboarding package for contracts over $25,000, according to Front's pricing page.
Overage blocks and a second meter
Overage blocks and secondary meters can add unexpected costs when usage exceeds base plan limits. ChatBot.com pricing sells additional resolutions in 50-packs at $49.50. Botpress pricing publishes conversation pack rates alongside a separate model-usage allowance. Dixa's $0.60 Mim overage applies above an order-form limit, according to the Dixa overage price list, so the rate alone cannot price the committed volume.
What does each product cost at three volumes?
The table below shows recurring monthly estimates for 1,000, 5,000 and 10,000 interactions, based on public terms observed on November 21, 2026. Each row prices the configuration described in that vendor's public terms, so the totals are not like-for-like across rows. Annual prices appear as monthly equivalents where used, while usage-only rows exclude required platform or deployment costs. A status of Not priced indicates vendor evidence did not support an estimate at that volume.
| Product | Advertised starting price or public reference | 1,000 | 5,000 | 10,000 | Coverage |
| Intercom Fin | $0.99 per outcome; Essential seat $19 promotional, $29 reference (plan terms) | $623.00 | $2,999.00 | $5,969.00 | Recurring |
| Zendesk AI agents | Support $19 per seat per month; Suite Team $55 per seat per month shown (resolution tiers) | Not priced | Not priced | Not priced | Undisclosed |
| Freshdesk Freddy AI Agent | Omni Growth $29 per agent per month, billed annually (add-on rules) | $519.00 | $2,479.00 | $4,929.00 | Recurring |
| Gorgias AI Agent | Helpdesk from $10 per month in vendor guide; AI advertised from $0.90 on most plans (billing terms) | Not priced | Not priced | Not priced | Undisclosed |
| Tidio Lyro | 50 one-time free; paid Lyro starts at $63 per month as shown (Lyro limits) | Not priced | Not priced | Not priced | Undisclosed |
| Help Scout AI Answers / Envoy | $0.75 per resolution; Standard $25 per user per month displayed (resolution pricing) | $400.00 | $1,900.00 | $3,775.00 | Recurring |
| HubSpot Breeze Customer Agent | $0.50 per resolution; Service Pro $90 to $100 per month per seat (agent guide) | $320.00 | $1,320.00 | $2,570.00 | Recurring |
| Salesforce Agentforce | $2 per conversation or $500 per 100k Flex Credits | $2,000.00 | $10,000.00 | $20,000.00 | Usage only |
| Microsoft Copilot Studio | $0.01 per credit pay as you go; $200 per month per 25,000-credit pack (credit billing) | $40.00 | $200.00 | $400.00 | Recurring |
| Google Dialogflow CX / Conversational Agents | $0.007 per Flows chat turn; $0.012 per Playbooks chat turn | $24.00 | $120.00 | $240.00 | Recurring |
| Amazon Lex | $0.00075 per text request | $1.50 | $7.50 | $15.00 | Recurring, Lex text requests only |
| IBM watsonx Assistant | Historical Plus $140 per month; current U.S. public starting rate unverified (plan docs) | Not priced | Not priced | Not priced | Undisclosed |
| Voiceflow | Current public self-serve base unverified; historic $60 per month Pro example (cost reference) | Not priced | Not priced | Not priced | Undisclosed |
| Botpress | $0 for 100 conversations; Plus $150 per month annual equivalent | Not priced | Not priced | Not priced | Undisclosed |
| Chatbase | Basic $120 per month annually or $150 monthly | $150.00 | $390.00 | $790.00 | Recurring |
| DocsBot AI | $0 for personal or testing; $49 Personal; $149 Standard | $149.00 | $149.00 | $198.00 | Recurring |
| CustomGPT.ai | $99 Standard monthly or $89 annual equivalent | $99.00 | $499.00 | Not priced | Two volumes |
| SiteGPT | $39 per month annual equivalent; $59 monthly Starter | $129.00 | $129.00 | $247.00 | Recurring |
| ChatBot.com AI Agent | Essential $19 annual equivalent or $25 monthly; Growth $79 or $99 | $396.00 | $2,376.00 | $4,851.00 | Recurring |
| Crisp Hugo | Mini $45 per workspace per month; Essentials $95; Plus $295 (Hugo billing) | Not priced | Not priced | Not priced | Undisclosed |
| Re:amaze AI Agent | Basic $29 per user per month; $26.10 annual equivalent shown (resolution billing) | $449.75 | $2,149.75 | $4,274.75 | Recurring |
| Zoho SalesIQ Answer Bot | $20 per operator per month annually; $25 monthly Enterprise (pricing calculator) | $25.00 | $25.00 | $25.00 | Recurring |
| Kustomer AI Agents | Current base and contracted AI starting price undisclosed; related progressive KIQ starts at $0.75 (AI pricing help) | Not priced | Not priced | Not priced | Undisclosed |
| Front Autopilot | Autopilot from $0.39 per conversation on pricing page; docs also list $0.05 triage (billing docs) | $572.00 | $2,860.00 | $5,720.00 | Usage only |
| Dixa Mim | Core starts at €89 per agent per headline; Mim base rate undisclosed (overage list) | Not priced | Not priced | Not priced | Undisclosed |
Why do the cloud meters look so cheap?
Cloud meters appear cheap because tools like Amazon Lex and Google Dialogflow price isolated requests, turns or actions rather than a full helpdesk deployment. An Amazon Lex text-request charge covers the specified Lex calls only, according to Amazon's Lex pricing page. A paid helpdesk configuration can also include an inbox, agent seats and handoff workflows.
What does a flat total mean?
A flat total, such as Zoho SalesIQ at $25 across all three volumes, means the modeled plan price did not change between volumes. Raw totals should be evaluated alongside the deployment assumptions specified for each product.
How do five of the bills add up at 5,000 interactions?
Five worked examples at 5,000 monthly interactions show how each vendor's meter turns the same support volume into a bill.
- Intercom Fin, $2,999 recurring. Assume one Essential seat at the $29 reference rate, 2,500 qualifying resolutions and 500 completed procedure handoffs, as set out in Intercom's pricing. The 3,000 billable outcomes cost $2,970, and the seat brings the total to $2,999. With the same resolution rate and zero completed procedure handoffs, the estimate is $2,504.
- Freshdesk Freddy, $2,479 recurring. Assume all 5,000 interactions use one chat session, priced under Freshdesk Omni pricing. Fifty 100-session packs cost $2,450, and one Omni Growth seat adds $29 on a monthly-equivalent basis. The first 500 sessions are a one-time allowance for eligible new accounts, so the example uses steady-state cost. If 1,000 cases generate a second session, 60 packs bring the monthly-equivalent total to $2,969.
- HubSpot Breeze Customer Agent, $1,320 recurring. At a 50% resolution rate, 2,500 resolutions use 125,000 credits under HubSpot Service Hub pricing. One Service Hub Professional seat supplies 3,000 credits, and this example assumes all 3,000 included credits are available to Customer Agent. The 122,000 excess credits cost $1,220. The seat adds $100 with an annual commitment, for $1,320 recurring. Required onboarding brings the modeled first month to $2,820. At 30% and 70% resolution, recurring estimates are $820 and $1,820.
- Front Autopilot, $2,860 in usage charges. Assume 2,500 resolutions at $0.89 ($2,225), 1,500 handoffs at $0.39 ($585) and 1,000 triages at $0.05 ($50), per Front's billing documentation. The three outcomes are mutually exclusive in the model. At 30% resolutions and the same 30% handoff share, triage rises to 40% and usage costs $2,020. The total invoice also depends on the platform and contract terms.
- SiteGPT, $129 recurring. On the Growth plan, 5,000 interactions with two GPT-4.1-mini messages each use the 10,000-message allowance listed on SiteGPT pricing. At three mini messages per interaction, one additional 5,000-message pack brings the estimate to $188. The same plan lists capacity for only 1,000 messages on full GPT-4.1, a tenfold difference in allowance per reply.
Which products could not be priced, and why?
Nine products could not be priced at any tested volume because public vendor materials omitted verified base fees, tier thresholds, allowances or overage terms. Those nine unpriced products are Zendesk AI agents, Gorgias AI Agent, Tidio Lyro, IBM watsonx Assistant, Voiceflow, Botpress, Crisp Hugo, Kustomer AI Agents and Dixa Mim.
Five of those nine vendors, including Zendesk, Gorgias, Tidio, Voiceflow and Botpress, published a clear charge trigger without providing the numeric rates needed to calculate a total monthly bill.
Salesforce Agentforce and Front Autopilot were priced on usage only, while CustomGPT.ai yielded estimates at 1,000 and 5,000 interactions but lacked complete pricing at 10,000 interactions.
How should buyers use these numbers?
Buyers should use the audit data to evaluate billing mechanisms before comparing headline costs, then examine four pricing components before signing a vendor agreement. Per-message and per-outcome meters turn the same support volume into different bills, as the SiteGPT and Intercom Fin examples show, because each system meters a different unit and bundles different components.
- Ask what specific event triggers a billable charge, and verify whether human handoffs, reopened conversations or failed attempts consume allowances.
- Ask which helpdesk, operator seat or base platform licenses must be purchased to run the AI agent.
- Ask for exact overage rates and mandatory pack increments when interaction volume exceeds the plan threshold.
- Ask about mandatory onboarding packages and minimum contract commitment lengths.
Buyers should model their historical resolution rate against each vendor's structure. Per-outcome models such as Intercom Fin and HubSpot Breeze scale directly with resolution rates, whereas per-session models such as Freshdesk Freddy remain fixed regardless of resolution success.
How often will the study be refreshed?
EPR Research plans to refresh the 25-product panel quarterly in February, May, August and November. Each update will document verified pricing effective dates, prior and updated tariffs, modifications to charge triggers and changes in pricing coverage.
Assumptions for 1,000, 5,000 and 10,000 interaction volumes at 30%, 50% and 70% resolution rates will remain standardized across updates. Any panel additions or removals will be reported separately to maintain historical tracking, with updated vendor terms analyzed against a baseline month of 5,000 interactions.
Related audits, including the survey disclosure study, are collected on the EPR research hub.