Intercom alternatives: the subscription is not the expensive part

This page does two things most comparisons skip. It treats Intercom's usage meter as the main event rather than a footnote, and it states plainly where our own argument fails — including a published Intercom discount that makes switching a bad idea for a whole class of company.

All pricing was read from vendor sources on 2026-07-25, and project-health figures were verified on 2026-07-29.

1. What the seats actually cost

Intercom sells three plans, priced per seat per month. Verified from the vendor's own structured pricing data:

• Essential — $39/seat/month, so $1,404/year for three seats
• Advanced — $99/seat/month, so $3,564/year for three
• Expert — $139/seat/month, so $5,004/year for three

At ten seats those become $4,680, $11,880 and $16,680 respectively. There is no volume discount and no annual-prepay discount on seats — we looked specifically, and the only annual-versus-monthly split published anywhere on the pricing page applies to the Copilot add-on, not to seats.

How we got these numbers, because the visible page cannot be trusted here: Intercom's rendered pricing cards were running a live A/B experiment showing a reduced Essential price, and the Advanced and Expert cards are digit-roller animations that contain no actual number in the page markup. The reliable source is the structured ld+json product data embedded in the page payload, where each plan carries an explicit SEAT_PER_MONTH unit label. Anyone quoting Intercom prices from a screenshot may be quoting an experiment.

2. The meter is the real cost driver

Fin, Intercom's AI agent, costs $0.99 per outcome with no upper bound. Here is Intercom's own definition of what triggers a charge:

An outcome is counted when: a customer confirms their issue is resolved, or they don't ask for more help after Fin responds, or Fin completes a workflow (Procedure), including handoffs. You're only charged once per conversation, even if multiple questions are answered.

Read the second condition again. A customer who gives up and closes the tab is billed the same as one whose problem was solved. Silence counts as success. That is not necessarily unreasonable — measuring genuine resolution is genuinely hard, and no vendor has solved it — but it means your bill is not a clean measure of value delivered.

To be fair to Intercom, the once-per-conversation rule is a real restraint. A customer who asks six follow-up questions costs $0.99, not six times that. The uncapped part is the total, not the individual exchange.

The structural problem is what the meter is indexed to. Seat costs track your headcount, which you decide. Outcome costs track your inbound support volume, which your customers decide. A product launch, an outage, a viral post, a confusing billing change — each one raises your Intercom bill at exactly the moment you least want a surprise. Intercom also publishes a minimum monthly commitment, with 50 outcomes given as its example, so there is a floor as well as no ceiling.

For scale: 421 outcomes a month costs about the same as a three-seat Expert subscription for a year. Support teams handling a few hundred conversations a week can reach that without anything unusual happening.

Seats and Fin are not the only meters, either. WhatsApp is charged per conversation, SMS per message with regional variation, bulk email campaigns per send, and phone by usage. Inbound support — replying to customer emails — is included, which is the part most teams assume they are paying for.

3. What actually transfers when you self-host

This is where the comparison has to be honest about scope. A shared inbox is a transferable capability: it is message routing, conversation history, canned replies, a help centre and a set of channel integrations. There is no proprietary dataset behind it. Nothing about it requires a vendor.

An AI agent that resolves conversations autonomously is a different thing. Self-hosting does not make that free. It converts a per-outcome charge into token spend with an LLM provider, plus the work of building and tuning the retrieval layer that makes answers accurate. Sometimes that is much cheaper. Sometimes, once you account for your own time, it is not.

So the honest framing is: self-hosting reliably eliminates the seat bill and only partially addresses the meter. Any page telling you a free tool replaces Fin outright is selling you something.

There is a second thing that does not transfer, and it is easy to overlook while staring at a price table: the channel integrations. Intercom maintains working connections to email, live chat, WhatsApp, SMS and phone, and absorbs the maintenance when a provider changes its API or tightens its approval process. WhatsApp in particular is not a simple API key — it runs through Meta's Business Platform with its own onboarding, template approval and per-conversation charges that you pay regardless of who hosts your inbox. Self-hosting moves that administrative work to you. It does not remove it.

What does transfer cleanly is worth naming precisely, because it is most of the product: conversation routing and assignment, a shared team inbox with internal notes, canned responses, tagging and search across history, contact records, a public help centre, SLA and business-hours rules, and a live chat widget. For a large share of support teams, that list is the job. The parts that do not transfer are the AI resolution layer and the convenience of someone else owning the integrations.

4. Chatwoot, assessed properly

Chatwoot is the closest functional replacement — a shared inbox positioned directly at Intercom's use case. As of 2026-07-29 it carries roughly 34.8k GitHub stars with 148+ commits in the last four weeks and same-day pushes, so it is unambiguously active.

On licensing, be precise: Chatwoot is open core. Everything outside the repository's enterprise/ directory is MIT licensed, which we confirmed by reading the licence file rather than trusting a badge — GitHub's API reports NOASSERTION because the file combines two licences, so automated checks can mislabel it. The enterprise/ directory is separately, commercially licensed. The core is genuinely permissive; some features are not part of it.

On security, the record deserves stating rather than burying. NVD lists 21 CVEs matching Chatwoot, nine of them from 2025 or later, with a maximum score of 9.1. The most recent, CVE-2026-63765 at 8.2, was published on 2026-07-23 — two days before we checked. That cuts both ways, and both ways are true. Disclosed-and-patched vulnerabilities are what a maintained, widely deployed project looks like; software nobody examines has no CVE history. But a customer support inbox is internet-facing by design and holds your customers' conversations, so an instance you stand up and never update is a genuine liability rather than a theoretical one.

There is also a result that undercuts the lazy version of this article. Chatwoot sells its own hosted plans, and its Business tier is priced at $39 per agent per month — identical to Intercom Essential. Switching vendors while staying on hosted software saves nothing. The arbitrage here exists in exactly one direction: self-hosting. If you were not going to run it yourself, this comparison does not apply to you.

5. The real self-hosted cost

Chatwoot's documentation specifies a floor of 2 cores and 4 GB RAM, and recommends 4 cores with 4 GB for workloads up to 10,000 conversations per day. Worth noting in the reader's favour: a three-to-ten seat team is nowhere near 10,000 daily conversations, so that guidance is comfortable rather than restrictive.

The catch is not RAM, it is cores. Both major vendors' 4 GB tiers ship only 2 vCPU, so satisfying the 4-core recommendation moves you up a tier:

• Vultr Cloud Compute Regular Performance, 8 GB / 4 vCPU — $480/year
• DigitalOcean Basic, 8 GB / 4 vCPU — $576/year
• Vultr 4 GB / 2 vCPU — $288/year, meets the RAM guidance but not the core count

Buying on memory alone is the easy mistake here: it silently under-provisions CPU. Beyond the instance, Chatwoot requires at least 1 GB of swap so that upgrades do not run the machine out of memory, wants 5–10 GB of Postgres storage, and runs Redis plus a Sidekiq worker. Sidekiq boots the entire Rails stack and the vendor notes it can exceed 1 GB of memory on a busy server, recommending you separate workers from the web process on active installations.

Against three Advanced seats at $3,564/year, a $480/year server leaves roughly $3,084/year — before any Fin usage, and before your own time. Server pricing snapshot: 2026-07-25.

If that patch cadence is the part you know you will not keep up with, the fix is managed hosting rather than a different piece of software. Cloudways runs the same DigitalOcean or Vultr hardware but takes OS patching, SSL renewal and staging off your plate, which is the specific risk this page keeps flagging. It costs more per month than a bare droplet, and for an internet-facing inbox holding customer conversations that is usually the right trade. If you would rather own the box outright, DigitalOcean gives new accounts $200 of credit, which covers a 4-vCPU instance for roughly the first four months. Both links are affiliate links.

The server invoice is the honest floor, not the honest total. Three costs sit on top of it and none appear on a vendor pricing page. Migration comes first: exporting conversation history, re-pointing your support email's MX or forwarding rules, re-authorising every channel, rebuilding macros and routing rules, and re-embedding the chat widget. None of it is hard in isolation, and all of it lands on someone during a week they had planned differently.

Backups come second, and they are not optional for a system holding customer conversations. Both DigitalOcean and Vultr sell automated backups as a paid add-on; we have deliberately left that figure out of the table above rather than estimate it, so treat the annual numbers as excluding backup. Budget for it explicitly, and test a restore rather than assuming one works.

Deliverability is third and catches people out. A self-hosted inbox still needs email that reliably reaches customers, which means a transactional provider plus correct SPF, DKIM and DMARC records. Intercom handles that reputation for you today. It is a solved problem, not a free one.

6. The case against switching

Intercom publishes a discount that defeats this entire analysis for some readers. From its own pricing FAQ: startups can get up to 93% off plus one year of Fin at no cost through its Early Stage programme. If you qualify, you pay almost nothing in year one — including the meter this whole page is built around. There is no arbitrage to capture. Take the discount.

That is not a small caveat, and it is the reason this page is aimed at teams past that window or ineligible for it — which, conveniently or not, is also when seat counts and support volume have grown enough for the numbers to bite.

Three more honest reasons to stay:

• Fin may be replacing salary, not software. If it genuinely resolves a large share of your volume, $0.99 per outcome can be far cheaper than the support hours it displaces. Compare it to labour, not to zero.
• At small scale the maths simply fails. One Essential seat is $468/year against $480/year for a server. You would pay more to run it yourself, and do the work.
• Someone has to own the patch cadence. Given a CVE published two days before this page was written, "we'll update it eventually" is not a plan for an internet-facing inbox.

7. Verdict

Self-hosting Chatwoot is worth doing when you are paying for three or more seats, you are past the startup-discount window, and someone on the team will genuinely own the server. At three Advanced seats the gap is around $3,084/year, and it widens with every hire.

Model the meter separately from the seats. Seats are the predictable part; Fin is the part that surprises you. And if the honest answer to "who patches this?" is nobody, the correct decision is to keep paying a vendor — that is a real service, not a failure of nerve.

Disclosure

This page contains affiliate links. The Cloudways and DigitalOcean links in section 5, and the hosting providers listed at the foot of this page, pay us a commission if you sign up through them. The prices quoted are the vendors' published list prices and clicking through does not change what you pay. Chatwoot itself has no affiliate programme. The instance tier we compare against is set by Chatwoot's documented 4-core recommendation, not by which host pays us more.