Airkit.ai implementation cost: Agentforce credits, editions and services

Most teams that search for an Airkit.ai implementation cost are holding one of two documents. One is a renewal or migration note from the days when Airkit ran their self-service journeys. The other is a board paper that says "we want what Airkit did, inside Salesforce". Both lead to the same purchase today, and it is not called Airkit.

Salesforce's completion notice is short: "Salesforce completed its acquisition of Airkit.ai on October 16, 2023." The signing announcement said where the product would live: "Upon close of the acquisition, Airkit.ai will become a part of Service Cloud and continue to be led by Adam Evans, Co-Founder and CTO of Airkit.ai." No Salesforce price list we read in October 2026 carries a line item with the Airkit name. What a buyer licenses now is Agentforce, deployed as a customer-facing service agent on Service Cloud. So this guide prices that, and treats the Airkit journeys you already run as the scope that the build has to reproduce.

#What you actually buy when the brief says Airkit

Airkit's selling point was low-code customer journeys: a form, an identity check, a lookup against an order or policy system, a branching conversation, and a handoff. In Salesforce today each of those pieces has a named home. The conversation runs as an Agentforce agent. The branches become subagents (the name Salesforce now uses for what it used to call topics). The lookups become actions built on Flow, Apex or an external service through a named credential. The channel is Messaging for In-App and Web, email, or voice. The handoff lands on an Omni-Channel queue as a case.

That mapping matters for cost because each part is paid for in a different way. Agent usage is metered. Service Cloud seats are licensed per user. The actions are build effort. Knowledge articles are content effort. A quote that rolls these into one "Airkit migration" figure hides the part that keeps costing money after go-live, which is consumption.

#The Salesforce price list, as at October 2026

These are list prices from Salesforce's own pages, read on 11 October 2026, in US dollars unless stated. Salesforce marks every one of these pages as information only and subject to change, and a negotiated order form will differ. The figures come from the Agentforce pricing page and the Service Cloud pricing page.

Two lines in that table need care. The per-conversation price still appears on the Agentforce page, but Salesforce labels it for existing Agentforce Conversations customers and tells everyone else to contact their account executive. A new buyer replacing Airkit should therefore model on Flex Credits. Second, the US$125 add-on is described as unmetered Agentforce usage for employees. It covers your service reps using Agentforce inside the console. It does not cover the customer-facing agent that replaces an Airkit journey, which draws on credits.

The Service Cloud page adds a footnote worth reading before anyone promises a fixed bill: "Agentforce for Service and Max Edition: Limited credits included." The next sentence says additional credits are available for purchase. Treat any included credits as a starting balance, not a ceiling.

#Turning credits into a monthly figure

The pricing page states the unit plainly: "Agentforce actions are 20 Flex Credits, while Agentforce Voice actions are 30 Flex Credits." At US$500 per 100,000 credits, one credit is half a US cent, so a standard action costs ten US cents at list.

The important word is action, not conversation. Salesforce's own worked example on the same page is a self-service "where is my order" agent. It authenticates the customer from an email address and retrieves the latest order: two actions, 40 credits. At 20 requests a day for 30 days, Salesforce shows a cost of US$120 a month. That example is close to a classic Airkit journey, which makes it a useful sanity check.

Do the same arithmetic for each journey you intend to rebuild. A returns journey that verifies identity, reads the order, checks the returns policy article, creates a case and books a courier is five actions, or 100 credits, each time it completes. If the agent retries an action because an order system timed out, that is another 20 credits. Volume forecasts should come from your current Airkit or web analytics, by journey, not from a vendor's benchmark.

Salesforce is also clear about what its examples leave out: "This example does not include other costs like Data 360 credits or other consumption services." If the agent grounds answers in Data 360, that consumption is a separate meter, priced on its own page.

#What the implementation is made of

A credible quote for replacing Airkit with Agentforce is built from work packages you can name and accept one at a time.

Journey inventory and discovery. List every live Airkit journey, its monthly volume, the systems it reads, and what happens when it fails. Decide which ones move, which ones retire, and which ones become a plain web form. This is where scope is won or lost.

Service Cloud foundations. If you already run Service Cloud, this is a check of case record types, queues, Omni-Channel routing and entitlement rules. If you do not, it is a full Service Cloud implementation and should be quoted as such, separately.

Agent design. For each journey, write the subagent, its instructions, its stop rules and its handoff. Keep one job per subagent and one effect per action.

Actions and integrations. Each lookup or update becomes a Flow, an Apex method or an external call through a named credential. Order, policy and billing systems are usually the largest line here, because each needs error handling and a test double for the sandbox. Where the system sits behind an API gateway or MuleSoft, the quote should say who owns that layer.

Identity verification. Airkit journeys often did their own verification. In Agentforce you design it explicitly: what the agent asks, what it checks, and what the case shows when the check fails.

Channels. Messaging for In-App and Web, email, and voice are separate deployments with separate consent and routing. Start with one.

Knowledge. If the agent answers policy questions, it needs current articles with named owners. Content work is often underquoted because nobody counts articles until testing starts.

Testing. Test each subagent against real utterances, including the ones that should hand off. Ask whether test runs draw on your credit balance under your contract, and budget for it if they do.

Cutover and hypercare. Run Airkit and Agentforce side by side on one channel, switch traffic, keep the way back for an agreed period, and read handoff notes daily for the first weeks.

#What moves the effort up or down

The biggest driver is how many distinct systems your journeys touch. A journey that reads one order system through an existing API is days of work. One that reads three systems, two without APIs, is a project in itself.

The second driver is how much of the Airkit logic lived in the journey rather than in a back-end system. Branching rules that were hard-coded in Airkit have to be rewritten as Flow or Apex, then tested again. If the same rules already sit in your ERP, the agent can call them and the work shrinks.

Third is channel count. One web messaging deployment is a modest build. Adding voice brings telephony, recording consent and latency testing, which is a different cost profile altogether.

Effort goes down when you already hold Service Cloud Enterprise or above, already have clean case data, and already maintain a knowledge base with owners. It goes up when the migration is also the moment someone decides to redesign the service model.

#Choosing the edition for the people behind the agent

An Airkit replacement changes the licence conversation for your human team as well as for the agent. When a journey stops, the case lands with a rep, and that rep needs a Service Cloud seat. The edition you pick decides what comes bundled with it.

Core is the case management edition. Advanced adds what Salesforce lists as Casey, the help agent, plus a Premier Success Plan, a full sandbox and Backup and Recover. Max adds the full Agentforce for Service suite for reps, digital channels and the 2.75M Flex Credits per org per year. Those included credits are the part most relevant to an Airkit replacement, because they can fund the customer-facing agent's first months of actions.

The arithmetic is worth doing honestly. If you have forty reps, the step from Advanced to Max is US$155 per user per month at list, across every seat, for a year. Compare that with the credits you would otherwise buy at US$500 per 100,000. For a small team with high self-service volume, buying credits on Advanced may cost less. For a large team that also wants the rep-facing AI features, Max may win. Run both cases with your own volumes before the account executive runs them for you.

#Getting quotes you can compare

Send every bidder the same pack: the journey inventory with volumes, the systems list with API documentation, your current Service Cloud edition and seat count, and the channels in scope for phase one. Ask each to return:

• a price per work package, not one total, with the assumptions written against each
• the number of actions per journey they intend to build, so you can check the credit forecast yourself
• what they will test, how, and whether those test runs draw on your credits
• what is out of scope, in writing, including telephony, data migration of historic transcripts, and knowledge authoring
• who on your side they need, for how many days, and when

Historic Airkit transcripts are a separate decision. Most organisations keep them as an archive under their retention policy rather than loading them as live cases. A quote that assumes full migration of conversation history should say why.

#Costs that are usually left out

Several costs sit outside both the licence and the services quote, and they are the ones that surprise finance in month three.

• Credit overrun. Retries and long multi-step journeys draw more credits than the demo did. Ask for a monthly credit report through Digital Wallet from day one.
• Data 360 consumption. If retrieval runs through Data 360, its credits are metered separately from Agentforce actions.
• Sandboxes. A Full Copy sandbox for realistic testing may not be included in your edition. Check before the test plan assumes one.
• Seat changes. Moving from Core to Advanced or Max for the AI features changes the per-user price for every licensed rep, not just the ones who use the agent.
• Licence overlap. If Airkit is still under contract, you will pay for both while the journeys move.
• Your own people. A service lead, an admin and a knowledge owner will spend real time on design reviews and testing. Put those days in the business case.
• Change management. Reps need to know what the agent now handles and how its handoffs look in the console.

#Where the other Airkit questions live

How Airkit's journeys map onto channels, cases and queues is on Airkit.ai and Service Cloud. The acquisition timeline is on what happened to Airkit.ai. The licence side, including how existing Airkit contracts sit beside Agentforce, is on Airkit.ai licence cost. Integration patterns for order and identity systems are on integrating Airkit.ai with Salesforce. Whether to move at all is on should I migrate from Airkit.ai, and how to test a partner before you sign is on hire an Airkit.ai Salesforce partner.

If you want help turning a journey inventory into a credit forecast and a phased build plan, our Agentforce consulting team does that work first, before anyone prices the build.