A case system with AI attached, or the other way round?
In September 2026 Salesforce completed a $3.6bn purchase of Fin — the company formerly called Intercom — and now leads its service pages with Fin as the flagship agent rather than with case management. That is not a feature release; it is a change in what the product is. Service Cloud today is closer to an AI resolution layer with case management behind it than to a console with a chatbot bolted on.
Salesforce publishes an average of 76% of queries resolved by Fin. The section below is about what that number counts, because the contract definition and the common-sense one are not the same.
Salesforce bought its service agent rather than building it.
Worth knowing in detail, because it explains why the product’s centre of gravity moved so suddenly, and because “deeply integrated” means something different for an acquisition this recent.
Intercom renamed itself Fin in May 2026. Salesforce signed to buy it on 15 June 2026 for about $3.6bn, and completed on 10 September 2026, bringing more than 30,000 customers with it. Salesforce says it is folding Fin into Agentforce.
The practical question for a buyer is integration depth, and the honest answer today is that it varies. Salesforce’s announcements did not spell out the Service Cloud relationship, and independent analysts expect packaged Fin agents to appear inside Service Cloud and Slack rather than describing something already shipped.
So ask specifically. Which pieces are native today, which are roadmap, and what the licensing looks like for each. An acquisition closed weeks ago is not the same as a product built together over years, and no amount of keynote language changes that.
Intercom becomes Fin
The company renames itself after its AI agent. Intercom survives as the name of the customer service software it has sold for years.
Salesforce signs
An agreement to acquire for approximately $3.6bn, subject to customary purchase price adjustments, framed as making Agentforce stronger in service.
The deal completes
Closed on 10 September, bringing the platform, the technical team and more than 30,000 customers into Salesforce.
Three things get called resolution. Only one is.
This is the most important section on the page. Before any contract is signed, pin down which of these the published rate is measuring, and which one you will be billed against.
Deflection
The customer did not reach a person. That includes the ones who gave up, closed the window, and went to social media instead. Deflection counts an absence, not an outcome.
Containment
The conversation stayed with the agent and did not escalate. Better, but it still counts a conversation that ended in an unanswered question as a success.
Resolution
The customer’s problem was actually solved, and they did not come back about the same thing within a sensible window. This is the only one worth paying for.
The four questions we put to the vendor
Is the rate measured on conversations started or conversations completed? Does a customer who abandons count as resolved? Does a customer who returns within seven days about the same issue reverse the resolution? And is the figure you are quoting me an average across all customers, or one from a customer whose volume mix looks like mine?
None of these is hostile. A vendor confident in the product answers them in a sentence each. The published figures — 76% on average, Bloomerang at 84% of conversations with an 85% CX score, Formula 1 resolving race-day issues 50% faster, Forrester’s cited 396% ROI over three years — all become more useful once you know which definition is behind them.
Your metrics will get worse. That is the sign it worked.
Here is the thing nobody mentions in the business case. If agents absorb the repetitive end of the queue, then what reaches a person is the hard end: the angry, the complicated, the ones with no policy covering them.
Average handle time goes up, because the easy three-minute cases have gone. First-contact resolution goes down, because the cases that resolve on first contact were the ones the agent took. Satisfaction per case may dip while overall satisfaction rises.
If you have not told your leadership this in advance, those numbers will be read as the project failing at exactly the moment it is working. Change the measures before you change the queue — and plan for fewer, more skilled, better-paid people rather than for the same team doing less.
The knowledge base stops being documentation and becomes the thing answering most of your customers. Someone needs to be paid to own it.
The role most organisations forget to createIllustrative, and deliberately unlabelled on the axis — the direction is the point, not the size. The first two moving the “wrong” way is the expected consequence of the last two moving the right way. Agree which four numbers you are managing before you start.
Eleven things are listed. You need about four.
Salesforce’s service page names a lot. Most organisations should start with a small subset and let specific problems justify the rest.
| Part | What it is | Start with it? |
|---|---|---|
| Customer Service Management | The core: cases, incidents, knowledge and assets, managed centrally. | Yes. This is the product. |
| Digital Channels | Messaging, chat, email and social, handled in one place. | One or two channels. Not all of them at once. |
| Self-Service | A help centre answering around the clock, personalised to the customer. | Yes, and it is only as good as the knowledge behind it. |
| Fin and Agentforce for Service | The AI agents: resolving for customers, briefing reps, giving supervisors live visibility. | After the knowledge base exists, not before. |
| Contact Center | An omnichannel contact centre, including the agentic version that unifies voice, digital, CRM data and agents. | Only if you run a real contact centre. |
| CCaaS Integrations | Connecting existing cloud telephony to the CRM. | If you already have telephony you are keeping. |
| Field Service and Operations | Scheduling, dispatch and mobile work for visits to a customer site. | Separate project. See Field Service. |
| Asset Service Lifecycle Management | Customer and asset data together, aimed at keeping equipment running. | Only where you service installed equipment. |
| Visual Remote Assistant | Live remote assistance, aimed at fixing things without sending somebody. | Later. Strong return when it fits. |
| IT Service and HR Service | Conversational, agent-first support for employees rather than customers. | Different audience, different project, different sponsor. |
| Service Cloud in Slack | Reps, experts and agents collaborating on a case in a channel. | If your experts already live in Slack, yes. |
Our usual first release is the first four rows, with one channel and a knowledge base somebody is paid to own. Everything below that line should wait until a specific, named problem justifies it.
Five rungs, and a second meter underneath.
Per user per month, billed annually. The rung that matters is Advanced, where an AI agent for customer service first appears at all.
| Edition | Price | What it adds |
|---|---|---|
| Starter Suite | $25per user / month | Entry CRM across sales, service, marketing and commerce, with built-in AI and Slack conversations. |
| Pro Suite | $100per user / month | Enhanced real-time chat, more customisation and automation, quoting and forecasting, AgentExchange. |
| Core | $195per user / month | Case management, a self-service help centre and Slack Business+. No AI agent yet. |
| Advanced | $395per user / month | An AI agent for customer service, Premier Success, a full sandbox, Backup & Recover and Data Detect. This is the rung where the product becomes what the marketing describes. |
| Max | $550per user / month | The full AI suite with Agentforce Coworker unmetered, contact centre tools, IT service with ten service-desk licences, Tableau Next, Slack Enterprise+ and 2.75m Flex Credits per org per year. |
Billed per action, not per answer
Agent work is metered in Flex Credits, and the nuance that decides your bill is that an action is the unit, not a reply. A one-step answer is cheap. A task that looks up an order, checks a policy, processes a refund and writes a confirmation is four actions, and the meter climbs with it. Voice actions cost more than standard ones.
Salesforce does not publish a plain rate card beside the edition table, and the per-action figures in circulation come from third-party analyses rather than from Salesforce. Get the rate card in writing, then model it against your real conversation mix — the proportion of one-step queries to multi-step tasks — rather than against an average. This is the single most common cause of a second-year surprise in this product.
Knowledge first. Always knowledge first.
The single best predictor of whether a service agent performs is not the model, the vendor or the integration. It is whether the answers exist, in writing, in a form something can retrieve.
Look at your own escalation reasons and you will usually find a fifth to a quarter of them are “the answer was not written down anywhere”. That is not an AI problem. It is the same problem that made your help centre useless, and it has been costing you for years in agent training and in repeat contacts.
Fixing it is unglamorous, cheap relative to the licences, and it improves service with or without an agent on top. It is also the work that makes every subsequent number believable.
Count your top fifty questions
Not categories — actual questions, in customers’ words, ranked by volume. Most contact centres have never done this and are surprised by the list.
Write the answer to each one
Once, properly, with the policy and the exceptions. Name an owner per article and a review date.
Decide what must never be automated
Complaints, vulnerability, anything with a legal or safety dimension. Write the list before anyone configures anything.
Define resolution, in the contract
With repeat contact inside the window. Then set the bar you expect, before you see a result.
Run one channel, measure honestly
Including the escalations and the abandonments. Then expand to the second channel, not to the fifth.
Where this leads next.
The five are read in any order, but these two are the ones this page leans on most.
© 2026 Mawkish Technologies. This guide, its diagrams and illustrations are protected by copyright. You’re welcome to learn from it and to quote it with credit and a link. Reproducing it, by hand or through AI, and publishing it as your own is not allowed. Read our copyright notice
What service leaders ask us.
What is Fin, and why does it keep coming up?
Fin is the customer service AI company formerly called Intercom, which renamed itself Fin in May 2026 and which Salesforce agreed to buy on 15 June 2026 for about $3.6bn, completing on 10 September 2026 and bringing more than 30,000 customers with it.
Salesforce now leads its service pages with Fin as the flagship service agent and is folding it into Agentforce. That is a large bet, and it changes what Service Cloud is: less a case-management system with AI attached, more an AI resolution layer with case management behind it.
Does it really resolve 76% of queries?
That is Salesforce's published average for Fin, with Bloomerang cited at 84% of conversations at an 85% CX score and Formula 1 at race-day issues resolved 50% faster. A Forrester Total Economic Impact study is cited at 396% ROI over three years.
Before you plan around any of them, pin down what resolution means in your contract. Deflection — the customer gave up — is not resolution. Containment is not resolution. Ask whether the rate is measured on conversations started or conversations completed, and what happens to the ones that escalate.
What does Service Cloud cost?
Per user per month, billed annually: Starter Suite from $25, Pro Suite from $100, Core at $195, Advanced at $395 and Max at $550.
The meaningful line is Advanced, which is where an AI agent for customer service first appears, and Max, which includes the full AI suite with Agentforce Coworker unmetered, the contact centre tools, IT service with ten service-desk licences, Tableau Next, Slack Enterprise+ and 2.75 million Flex Credits per org per year.
How is the agent work actually metered?
Through Flex Credits, and the important nuance is that you are billed per action rather than per answer. A one-step FAQ reply is cheap; a multi-step task fans out into several actions and the meter climbs with it. Voice actions cost more than standard ones.
Salesforce does not publish a plain rate card on its pricing page, and the per-action figures circulating come from third-party analyses rather than from Salesforce. Get the rate card in writing, and model it on your real conversation mix rather than on an average.
Which parts of Service Cloud do we actually need?
Far fewer than are on the page. Salesforce lists self-service, customer service management, contact centre, CCaaS integrations, digital channels, field service, asset service lifecycle management, visual remote assistant, IT service, HR service and Service Cloud in Slack.
Most organisations need case management, one or two channels, a knowledge base somebody owns, and a clear escalation path. Everything else should wait for a specific problem to justify it.
What happens to our service team?
The work changes shape rather than disappearing. Volume handled by agents is overwhelmingly the repetitive end, which means what reaches a person is harder, longer and more emotionally demanding than the average case used to be.
Plan for that honestly: fewer, more skilled people; different metrics, because average handle time gets worse when the easy cases are gone; and a knowledge base that someone is paid to maintain, since it is now the thing answering most of your customers.
Checked against Salesforce's own pages, its pricing pages and press coverage of the Fin acquisition in November 2026. This product is changing quickly; confirm what is native today before you sign.
Send us your escalation reasons.
Not your requirements — the list of why conversations reached a person last month, with volumes. It tells us what an agent could genuinely resolve, what should never be automated, and how much of your escalation rate is simply an answer nobody has written down yet.
- Your top fifty questions, counted
- Resolution defined before the contract, not after
- A written list of what is never automated
- Flex Credits modelled on your conversation mix
- Metrics changed before the queue changes
Salesforce, Agentforce, Data 360, Slack, Tableau, MuleSoft and other Salesforce product names are trademarks of Salesforce, Inc. Other names are the property of their owners. Screens shown are illustrations with sample data. Product descriptions follow Salesforce's own published material and can change, so check salesforce.com for current details.
The Mirwick Ladder is our shorthand for looking after the details nobody sees in Salesforce Service Cloud.