Shopify
Stripe
Slack
Notion
Google Workspace
Confluence
Intercom Fin Implementation Plan (2026): Week by Week, Who Owns Each Task, and What It Costs
A Fin implementation plan follows Intercom's own four steps, Train, Test, Deploy and Analyze, and for a mid-size support team it fits in about 8 weeks: one week auditing past conversations, three weeks of content, Guidance and Procedures, one week of testing, two weeks of staged rollout with daily review, then a weekly review. Fin's Deployment Services page says "pilot in weeks, enterprise rollout in under 90 days". The software runs $0.99 per outcome plus seats; the bigger cost is about 157 hours of your team's time.
Key takeaways
- A Fin implementation plan for a mid-size team runs about 8 weeks on Intercom's Train, Test, Deploy and Analyze steps, plus a weekly review that continues after launch.
- Fin's batch tests take up to 50 questions per test group, and Intercom strongly recommends running Simulations before every Procedure launch.
- Intercom's own chat deployment article says to set Fin live to a small test or internal audience first, such as yourself and your teammates.
- Faye, an Intercom partner, calls 60% to 70% involvement and 40% to 50% resolution a healthy Fin deployment; that is its claim, updated September 22, 2026.
- In our worked example, 2,000 conversations a month on 5 Advanced seats costs about $1,163.55 a month in software and about 157 hours of team time to roll out.
This is the third of our Intercom pages. If you're still choosing between doing it yourself, hiring a partner or handing it to a vendor, start with Intercom AI implementation; for named agencies, see the best Intercom implementation partners. This page assumes you've picked Fin and want the plan; the AI customer support implementation guide covers the same phases on any help desk. Disclosure: Macha publishes this blog and sells an AI agent that runs inside Intercom and competes with Fin. We checked every Intercom and Fin page linked below on October 14, 2026.
What is Intercom's own implementation process for Fin?
Intercom organizes Fin into four areas in the product, and its Deployment Services page uses the same four as its method, in a different order:
| Step | Intercom's description | What you configure (per Fin's overview) |
|---|---|---|
| Analyze | "Identify high-impact opportunities and define success criteria" | Performance dashboard, plus Topics Explorer, Recommendations and CX Score with the Pro add-on |
| Train | "Connect systems, configure workflows, optimize knowledge, and shape Fin's behavior" | Content, Guidance, Procedures and data connectors |
| Test | "Validate responses, refine configurations, and run end-to-end testing" | Previews, batch tests, Simulations |
| Deploy | "Roll out incrementally, measure results, and drive continuous improvement" | Channels, audiences, Simple deploy or Workflows, usage limits |
The product menu reads Train, Test, Deploy, Analyze. Deployment Services puts Analyze first, and that's the order a plan should follow: you can't pick what to train Fin on until you know what your customers ask. Analyze then comes back at the end as the weekly habit.
Intercom doesn't publish a week-level schedule. Deployment Services says "sandbox on day 1, pilot in weeks, enterprise rollout in under 90 days". Faye's kickoff-to-hypercare post says a focused jumpstart "typically runs a few weeks". The plan below is our synthesis of those sources for a team of 5 to 20 agents with one or two topics that need account data.
What does a week-by-week Fin rollout look like?
| Week | Step | Work | Done when |
|---|---|---|---|
| 0 | Analyze | Pull a month of conversations, group them by topic, record the baseline (volume, first response time, CSAT), pick 3 to 5 launch topics, write the success scorecard | The scorecard names the topics, target metrics and the go/no-go owner |
| 1 | Train: Content | Fix or write the articles for the launch topics, add internal articles and snippets, set content targeting by audience (it relies on clean attributes), let new sources finish indexing | Every launch topic has one current, non-contradictory source |
| 2 | Train: Guidance | Tone and answer length, clarifying questions, escalation rules, handoff Workflows and office hours | Each Guidance rule does one job; escalation routes to a named team |
| 3 | Train: Procedures | One or two Procedures for topics that need account data, data connectors set live, write limits agreed | Each Procedure has a written success criterion |
| 4 | Test | Batch tests per audience and language, Simulations for every Procedure, Previews as real users, the four test cases | Sign-off from the support lead against the scorecard |
| 5 | Deploy | Fin live to an internal audience, then one channel and one segment; usage limits set | A day of internal conversations reviewed with no blocking misses |
| 6 to 7 | Deploy and hypercare | Widen the audience, add email as its own deploy, read Fin conversations daily, fix within 24 hours | Two weeks at or near target on the scorecard |
| 8 | Analyze | Exit review, hand over to the weekly cadence, choose the next topics | Owner signs off; the monthly review is on the calendar |
Small teams compress this. A startup on Essential with good docs and no Procedures can do weeks 0 to 2 in a week and launch in the second; Essential has no Workflows builder (Intercom pricing), so it deploys with Simple deploy. Enterprises stretch it: several brands, regions and Procedures per team is the 90-day shape Deployment Services describes.
What you need to provide on any route: Intercom admin access, a month of conversations, your refund, credit and account policies in writing, API documentation and credentials for each Procedure, two or three agents as testers, and one named go/no-go owner.
One timing trap. Fin's pricing page gives a 14-day trial with unlimited outcomes. If Fin isn't on your workspace yet, a trial started on day one ends in week 2. Do the week 0 audit and the week 1 content work first, then start the trial so it covers testing and the first live week.
Week 0: audit past conversations and content gaps
Pick launch topics from your own volume, not a vendor's list. With the Pro add-on (from $100 a month on fin.ai/pricing), Topics Explorer groups conversations for you. Without it, export a month of Intercom conversations and tag a sample of 200 by hand. For each topic, note three things: how many conversations a month, whether the answer is the same for everyone or depends on the customer's account, and whether a current article exists. Topics that are high-volume, same-for-everyone and already documented go first. Topics that need Stripe or CRM data go into week 3. Topics where the right answer depends on a human judgment (security reports, contract disputes) stay with people, and you write the escalation rule for them in week 2.
Weeks 1 to 3: content, Guidance and Procedures
Content is most of the work. When Fin gives a wrong answer, the cause is usually two articles that disagree or one that's out of date; our guide to Fin not answering from your content covers the sync and indexing problems. Intercom's content gap recommendations flag missing, unclear, duplicated and contradictory content, but they need the Pro add-on and they work from live conversations, so they help from week 5 onward.
Guidance comes next. Keep each rule to one purpose; in an Intercom community thread on Fin repeating itself, the accepted answer names conflicting or overlapping Guidance rules as the most common cause. Escalation rules decide when Fin hands over, and a Workflow decides who gets the conversation, so build both together (Fin handoff setup).
Procedures are for multi-step jobs with branching, code checks and data connectors to systems like Stripe or Shopify, not simple FAQs. Budget engineering time: someone has to expose your API and agree which writes Fin may make (a refunds example).
Week 4: test with Fin's three tools and four cases
Intercom's testing comparison (updated July 31, 2026) describes three tools:
| Tool | What it does | Limits | Use it for |
|---|---|---|---|
| Previews | You chat with Fin yourself, "Testing as" a real user, lead or audience | No usage limit; conversations show in your inbox but stay out of reporting | Spot checks while building |
| Batch tests | Fin answers a list of questions; you rate each Good, Acceptable or Poor | Up to 50 questions per test group; no charge | Coverage across topics, audiences, brands and languages |
| Simulations | An AI plays the customer through a Procedure and judges pass or fail against your criteria | 250 to 12,500 runs a month, depending on workspace volume | Every Procedure, before launch and after each change |
Two details trip teams up. Previews include draft Guidance, but "a data connector must be set live before it can be triggered in preview. Draft connectors won't fire." And Intercom says "running Simulations before every Procedure launch is strongly recommended", so treat a Procedure without a passing Simulation as unfinished.
Then write test conversations in the four shapes Faye uses: a happy path that resolves end to end, an escalation that should reach a person with context collected, an edge case (a lead asking about an invoice, a customer on a legacy plan, a refund on a partly shipped order) and a tone case where the facts are right and the policy has to hold. Take them from real conversations in your week 0 sample. Our post on testing agents on real data before launch explains why replays beat invented questions.
Weeks 5 to 7: staged deployment and hypercare
Intercom's chat deployment article tells you to "set Fin live over chat to a small test or internal audience (e.g., yourself and teammates)" first. After a day of internal use, widen by audience and channel, one change at a time so you can tell which change moved the numbers:
- Internal team only, Messenger.
- Logged-in users on one plan or region, Messenger.
- All users, then leads and visitors.
- Email, as its own deployment; forwarding alone doesn't put Fin on email.
- WhatsApp, social and SMS last.
Set usage alerts and a hard limit before step 2 (Fin usage alerts and limits). Hypercare is a partner's word, not Intercom's; Faye defines it as "close monitoring of real conversations in the weeks following launch". In practice: someone reads every conversation Fin touched on the new slice each day, sorts misses (wrong content, wrong Guidance, missed handover, Procedure failure), and fixes the source within 24 hours. Agree the rollback rule before you widen: if a slice's CSAT or reopens fall below the scorecard floor for two days, narrow Fin's audience in Deploy back to the previous step and fix before trying again. Rollouts that skip this are the ones that get switched off a month later.
Week 8 onward: analysis that keeps going
After the exit review, the cadence is weekly: read the performance dashboard, open a random sample of conversations Fin marked resolved, and work the Recommendations list if you have Pro. Add new topics monthly, each with its own cycle of content, test and staged deploy.
Who owns each task: a RACI for the Fin rollout
R is responsible (does the work), A is accountable (signs off), C is consulted, I is informed. Fin Deployment Services or a partner takes some R cells if you hire one; the A cells stay with you either way.
| Task | Support lead | Intercom admin | Content owner | Engineer | Agents (testers) | Partner or Deployment Services |
|---|---|---|---|---|---|---|
| Success scorecard and launch topics | A | C | C | I | C | R if hired |
| Content fixes for launch topics | A | I | R | I | C | C |
| Guidance and escalation rules | A | R | C | I | C | R if hired |
| Handoff Workflows, assignment, office hours | C | A/R | I | I | I | C |
| Procedures and data connectors | A | R | C | R | I | R if hired |
| Write and refund limits | A | C | I | C | I | C |
| Batch tests, Simulations, four test cases | A | R | C | C | R | R if hired |
| Go-live decision per audience | A/R | C | I | I | I | C |
| Usage limits and budget | A | R | I | I | I | I |
| Hypercare daily review | A | R | R | C | C | R while engaged |
| Weekly analysis after week 8 | A | R | R | I | I | Scale engagement or retainer |
What Fin may change in a customer's account, and the decision to let it talk to a new audience, stay with your support lead even when a partner does the rest.
The Fin launch checklist
Copy this into your project doc and tick it in week 4:
- Success scorecard written: launch topics, target involvement and resolution, a satisfaction floor (CX Score needs Pro; otherwise your CSAT survey), go/no-go owner
- Baseline recorded from week 0 (volume, first response, CSAT per topic)
- One current article per launch topic; duplicates and contradictions removed
- Content targeting set for each audience, brand and language
- Guidance rules each do one job; tone and answer length set
- Escalation rules cover security, legal, cancellations and angry customers
- Handoff Workflows route to named teams; office hours checked
- Old Workflows on new conversations and "customer sends any message" reviewed so they don't pull Fin back in
- Every Procedure published, data connectors set live, write limits written down
- Every Procedure passes its Simulations
- Batch tests run per audience and language, Poor answers fixed
- Four test cases run in Preview as real users
- Usage alerts and a hard limit set
- Internal audience live for at least a day
- Hypercare reviewer named for weeks 5 to 7, with a 24-hour fix rule
- Weekly review on the calendar after week 8
Which KPIs should you track, and what counts as good?
Intercom defines Fin's core metrics in its performance report article (updated June 24, 2026):
| Metric | Intercom's definition | What to watch |
|---|---|---|
| Involvement rate | Conversations where Fin participates and attempts an answer, out of all new conversations | Rises as you widen the audience; flat means Fin is being bypassed |
| Resolution rate | Conversations Fin resolved, out of those it was involved in | Split confirmed from assumed resolutions; assumed ones hide silent wrong answers |
| Automation rate | Involvement rate × resolution rate | The share of total volume Fin closes; the number finance cares about |
| CX Score (Pro add-on, per fin.ai/pricing) | Share of positive ratings (4 or 5 out of 5) on conversations Fin handled alone | Compare with your human CSAT on the same topics |
| Escalations and Procedure handoffs | Outcome types in the same report | Escalations by reason point at the Guidance or content to fix |
Add one metric Intercom's dashboard won't give you: reopens and repeat contacts on the same topic within a week of a Fin resolution.
A warning on comparisons. Intercom changed the metric definitions in mid-2026: conversations where Fin was "constrained" (a workflow or escalation rule fired first, a usage limit hit, a teammate stepped in) no longer count as involved. Involvement went down and resolution went up, with billing unchanged. Existing workspaces moved over July 1 to 8, 2026, so a resolution rate from before July isn't comparable with one after.
Published reference points, all vendor or partner claims rather than independent benchmarks:
- Fin Deployment Services cites a 72% average resolution rate across its deployments.
- Salesforce cited a 76% average Fin resolution rate when it completed the acquisition on September 10, 2026 (sourced in our Fin explainer).
- Faye's partner guide (published May 21, 2026, updated September 22, 2026) calls 60% to 70% involvement and 40% to 50% resolution a healthy deployment, with the best above 70% resolution and a few above 90%. Its rollout post reports one client moving from 36% to 62% of conversations handled by Fin over three months.
Use your own week 0 baseline as the target, not these. The gap between 40% and 72% is mostly your content and your topic mix, and nobody can tell you where you'll land before reading your conversations.
What does Fin Deployment Services cover, and what stays with you?
Deployment Services sells two engagements. Launch, for new customers, lists a success scorecard, use case prioritization, knowledge optimization, Guidance and task configuration, channel implementation, pre-launch testing and post-launch optimization. Scale, for existing customers, adds performance benchmarking, workspace optimization, a task expansion plan, workflow expansion and a team AI readiness review. Named roles include an AI Deployment Consultant, an AI Agent Engineer and a CX Architect. The price is "scoped to your goals" and not published.
| Area | Deployment Services (Launch) | Stays with you |
|---|---|---|
| Choosing topics and targets | Success scorecard, use case prioritization | Approving them, and the baseline data |
| Content | Knowledge optimization | Policy facts, and keeping articles current after every release |
| Guidance and tasks | Configuration | Refund, credit and write limits |
| Channels | Channel implementation | Workflow and routing decisions for your teams |
| Testing | Pre-launch testing | Sign-off, and testers who know your customers |
| After launch | Post-launch optimization (length not published) | Everything once the engagement ends, unless you buy Scale |
| Your own APIs | Not stated on the page; ask | Access, credentials and the engineer who owns the endpoint |
Fin's own page also says "You can set Fin up yourself, and many teams do." Deployment Services is worth its price when Procedures touch billing or account systems, or when several teams, brands or regions are involved. Fin earns per outcome, so its own team's incentive and yours line up on resolution count; where they can part is on assumed resolutions, which bill even when the customer simply went quiet. Read those yourself.
What does the plan cost?
Here's the arithmetic for a B2B SaaS team with 2,000 new conversations a month on 5 Advanced seats, after the rollout. Assumptions: 65% involvement and 45% resolution (inside Faye's healthy range), 60 Procedure handoffs a month, and the Pro add-on for Recommendations.
| Line | Arithmetic | Monthly |
|---|---|---|
| Intercom seats, Advanced, billed annually | 5 × $85 | $425 |
| Fin outcomes | 2,000 × 65% = 1,300 involved; 1,300 × 45% = 585 resolutions; 585 + 60 handoffs = 645 × $0.99 | $638.55 |
| Pro add-on | From $100 | $100 |
| Software total | $1,163.55 |
Automation rate is 65% × 45% = 29.25% of volume. Fin bills Procedure handoffs at $0.99 too, which surprises teams who budget only for resolutions; our Fin pricing breakdown models other volumes.
People time is the bigger cost, and nobody's price list shows it. Our estimate for the 8-week plan, by role:
| Role | Hours over weeks 0 to 8 |
|---|---|
| Support lead | 27 (3 a week) |
| Intercom admin | 40 |
| Content owner | 50 |
| Engineer (Procedures, data connectors) | 24 |
| Agents as testers | 16 |
| Total | 157 |
At an assumed $50 an hour fully loaded, that's $7,850 to roll out. After week 8, the weekly review and content upkeep run about 5 hours a week, roughly 22 hours or $1,083 a month at the same rate. A partner swaps some of those hours for a fee: Cloudstride publishes EUR 4,000 to 7,000 for a 4-to-8-week project that includes Fin, and the content owner and support lead hours stay with you either way.
So the first year looks like one-off rollout time plus about $2,247 a month in software and upkeep at this volume. The outcome line grows as Fin gets better, which is the trade per-outcome pricing makes.
When should you hire a partner instead, or not use this plan?
Hire Deployment Services or a partner when you're migrating to Intercom at the same time, when you run several brands or workspaces, or when nobody internal can give the content owner's 50 hours. Our partner ranking covers who does it and what they charge. Don't use this plan as written if your help center is close to empty; write the core articles first, because every week after week 1 depends on them.
Fin rollout plans by situation
| You are | Plan | Why |
|---|---|---|
| A startup on Essential, under 500 conversations, good docs | Weeks 0 to 2 in one week, Simple deploy, skip Procedures | Seats from $29 and a small outcome bill; no Workflows on Essential; Topics, Recommendations and CX Score need Pro |
| A B2B SaaS team, 2,000 to 5,000 conversations, one or two billing topics | The full 8 weeks | Procedures and data connectors need their own week and real testing |
| An ecommerce brand on Intercom with Shopify | The full plan, with order lookups as the first Procedure | Order status and returns are the volume; they need account data |
| An enterprise with several brands or regions | Deployment Services or a partner, about 90 days | Routing, change management and per-brand content are the work |
| A team already on Fin with a flat resolution rate | Restart at week 0 on the topics with the most escalations | The fix is usually content and Guidance, not a new tool |
| A team with no one who can own weekly analysis | Budget a retainer, or look at a vendor that runs it for you | Fin's results fall back without the weekly review |
Running a different agent inside Intercom
The last row is where Macha fits, and it's ours. Macha runs its own AI agent inside your existing Intercom instead of configuring Fin, so it's a Fin competitor, not a Fin service (Macha vs Fin compares the two). The Macha team does the plan above for you: it analyzes your past Intercom conversations, fills the knowledge gaps for the first topic, builds the agent and runs it in safe mode, with drafts as Intercom internal notes, until your team approves them. Live in safe mode within 7 days, or your first month is free. Your part is about 15 minutes to connect Intercom and the tools your answers come from, then about 5 minutes a day reviewing drafts.
It bills about $0.40 per ticket for every conversation the agent works, resolved or not, which includes the done-for-you setup and a dedicated success manager who handles new topics, product changes and new tools. At the 1,300 conversations a month in our example, that fits the $599/month for 1,500 tickets plan, and Intercom seats still apply. Macha suits teams on Intercom that want the rollout and the weekly review done for them. It's the wrong fit if you want Fin Voice, Fin's own reporting, or to edit every rule yourself. The trial includes the same setup: $50 of free usage (about 125 tickets), no credit card, no time limit. See Macha on Intercom, or ask for the free support audit to see which topics an agent could take first.
Common questions
How long does a Fin implementation take?
For a mid-size team with one or two topics that need account data, about 8 weeks from audit to the end of hypercare. Fin Deployment Services promises a pilot in weeks and an enterprise rollout in under 90 days; a small team with good docs can launch in about two weeks. For how Fin's timeline compares with other vendors and agencies, see how long AI customer support takes to implement.
What are the steps to implement Fin?
Intercom's four: Train (content, Guidance, Procedures), Test (Previews, batch tests, Simulations), Deploy (channels and audiences) and Analyze (performance, topics, recommendations). Run Analyze first on past conversations to pick launch topics, then again every week after launch.
What test cases should you run before launching Fin?
Run batch tests of up to 50 questions per group across your audiences and languages, a Simulation for every Procedure, and four hand-written cases from real conversations: a happy path, an escalation, an edge case and a tone case. Test in Preview as a real user, because audience targeting changes which content Fin can use.
Can you roll Fin out to a small group first?
Yes, and Intercom recommends it: its chat deployment article says to set Fin live to a small test or internal audience, such as yourself and your teammates, before everyone. After that, widen by segment and channel one step at a time, with email as its own deployment.
What is a good Fin resolution rate?
There's no independent benchmark. Fin Deployment Services cites a 72% average, Salesforce cited 76%, and Faye, an Intercom partner, called 40% to 50% resolution with 60% to 70% involvement healthy in its guide updated September 22, 2026. Measure against your own baseline, and note that Intercom changed how resolution rate is calculated in July 2026.
Why is my Fin resolution rate low?
Usually because of content: missing, outdated or contradictory articles on the topics with the most escalations. Next come overlapping Guidance rules, Procedures that never trigger because an article answers first, and audiences that leave Fin out of whole segments. Start with escalations by reason and the Recommendations list.
How we researched this
On October 14, 2026 we checked Fin's Deployment Services page, Fin's overview, Intercom's articles on testing (Simulations, batch tests and Previews), Previews, Procedures, content gap recommendations, chat and email deployment, the Fin performance report and the 2026 metric change, plus fin.ai/pricing and intercom.com/pricing. Partner claims come from Faye's rollout post (published October 9, 2026) and partner guide, and Cloudstride's tiers from our partner ranking, checked the same day. The week-by-week plan, the RACI and the hour estimates are our synthesis from those sources, not a schedule Intercom publishes. The cost example uses assumed rates (65% involvement, 45% resolution, 60 handoffs, $50 an hour), and the arithmetic is shown in each table. We didn't run an Intercom workspace or hire a partner for this page.
