AI for Customer Success Webinar February 19th

This session is about using AI to boost your customer success, meaning the perceived value your clients get from working with you. It is built for service providers, agencies, and consultants who want their clients to feel the work being done: onboarding documents, automated reports, dashboards, and a full client portal, most of it costing you almost nothing to produce. We start with quick wins anyone can use today, then go hands-on with a live build in Claude Code.

What customer success actually is (2:03)

Unlike the previous webinars on sales and marketing, customer success is not a business vertical. It is everything surrounding your company that lets customers see perceived value in your services (7:03). When a customer buys a service, you can simply deliver it and maybe they will be happy. But if you package that service with assets, onboarding, and documentation, the customer sees a huge increase in perceived value, even though your workload barely grows, because AI now automates most of it.

The key components (8:30): structure the onboarding journey, with a page or PDF stating the scope, deliverable dates, and a checklist for the first call. Track your milestones, so instead of clients asking "any updates?" you communicate progress constantly. Keep regular check-ins and business reviews, like a monthly report of everything you did. Customer success wraps around sales, marketing, and customer support all at once: after a sales call, the client gets a clean proposal exactly as requested; in support, you go from reactive to proactive by giving people everything in advance.

For tools (4:02), any of ChatGPT, Perplexity, Gemini, or Claude works, but I recommend Gemini or Claude. I ran the demos on the Claude desktop app. Fun fact: if your business pays for the Google Standard plan, Gemini is already included, so paying separately for ChatGPT is just losing twenty dollars a month. For the advanced parts we used Nano Banana Pro (which was down that day), Antigravity, and Claude Code; Cursor Pro works too.

Quick win 1: AI onboarding documents (10:40)

We played a persona game for the whole session: a recruitment agency that just signed a tech client, with a package of nine hires in eight weeks. The first prompt, which I shared during the session, gives Claude the business context (who we are, what we sold, the client details) and asks for an onboarding playbook as an HTML page: welcome and engagement overview, working processes, kickoff agenda, communication cadence, a "what we need from you" checklist, and intake questions (13:32).

That checklist matters more than it looks. If you work in a business where you need customer credentials or information, asking three separate times because you forgot during onboarding does not look professional. A solid onboarding process fixes that permanently, and the generated page is reusable for every new customer.

Two upgrades. First, connect your LLM to Google Drive (13:47): if your pricing or packages live in a document, Claude or Gemini can reference them so every detail is correct. Second, output formats: you can convert the page to PDF, but my advice is to stick with HTML, because PDFs cut off at awkward moments. If you run your business in Notion, connect Claude through Connectors and it creates the onboarding page natively, checklists and key contacts included, no Markdown pasting needed. And if you use Stripe or another payment service, you can wire it up so a new payment automatically triggers the onboarding generation.

Quick win 2: automated weekly reporting (18:06)

Reporting is how clients know you are actually working, especially for volume work. I created a fake CSV of one week of candidate sourcing for our pretend client and shared it on the session's Notion vault so everyone could test with the same data. The prompt takes that raw data and generates a full weekly report (23:10): week one of eight, executive summary, KPIs, how many people were hired, how many are at offer stage, how many were sourced, the breakdown by role, and sourcing channel effectiveness.

Because reports can now be generated instantly, you can communicate more effectively, with more KPIs, and more often: instead of monthly, go fortnightly or even weekly. It costs you nothing to produce and the perceived value is huge. On consistency (25:06): iterate until you find a report you like, then keep it as a template. From then on you hand over the new raw data plus the template and ask it to keep everything the same. A report that looks different every week does not look professional either. And on accuracy: the AI does not freehand the numbers. It generates code, code always gives the same output, and that code fills the template with the correct data.

Quick wins 3 and 4: voice reports and interactive dashboards (26:21)

Quick win three is mostly for fun: drop the report into NotebookLM and generate a one-minute or three-minute voice report of everything done for the customer. An attendee suggested posting it to Slack weekly, which is a genuinely good use case I had not considered.

Quick win four is the interactive dashboard (27:13). Instead of a static report where the client wonders about one precise number, give them a dashboard where they filter and split the data themselves. It is still a plain HTML file, so it shares just as easily, and you can deliver the static report plus the dashboard together.

The client portal (30:30)

Everything above converges into one thing: the client portal, a single page (Notion or web) where your customer always finds everything about the collaboration: weekly reports, the onboarding playbook, the contract, emails, call summaries. A lot of professionalism gets lost in scattered communication, and a portal fixes it.

The demo portal (33:00) showed the full recruitment pipeline: people sourced this week, hires completed, days remaining in the collaboration, channel activity per day, top candidates, plus a live activity feed, so when your team moves a project from one stage to another, the client sees a live notification. It also holds the weekly reports and the playbook. Two honest caveats: this level needs deployment (Vercel or similar) with passwords and basic security per customer, and if you handle sensitive data you should get technical help. For the simple version, Notion is fine. As a bonus (38:21), you can hand raw data to Nano Banana Pro and get an infographic-style report; it will not be consistent between runs, but it is another format your clients might love. It was down during the session, so the fallback demo used the non-Pro model, which looked noticeably worse.

Live build: client portal with Claude Code (40:01)

For the live build I opened Claude Code with a prompt that Claude Code itself had generated: prerequisites, scaffold the project, build the dashboard UI, step by step with checkpoints so we could watch the portal evolve. The prompt was too big for the chat, so I shared it on the Notion vault. Within minutes we had a live server and a working dashboard, then added the AI chatbot, then ran a design pass: improve the front-end with the design skill, add a loading animation to the assistant, and fix the formatting so it stops printing raw asterisks. The result looked visibly more polished on reload.

The chatbot deserves a highlight (36:18): it answers questions about everything inside the portal and can cite its sources, so your customer can always verify the information. Back in late 2024 everyone called this RAG and it was complex to set up; now it takes minutes. Honestly, I do not even know the internals, I just asked Claude Code to build it. The full portal took around thirty minutes.

Along the way I gave a Claude Code crash course. On permissions (44:11): you have accept, plan mode, review (it asks before every step, great while learning), and bypass for people like me who are not scared of death. Three ways to use it (47:50): inside the Claude desktop app's code window, the friendliest option if terminals intimidate you; as an IDE extension; or in the terminal, which is what I use, purely a matter of preference. The golden rule: always have Claude Code working on something, then go grab a coffee. I also mentioned WisprFlow, the voice tool that let me go from three Claude Code instances to six at once.

Client portals in Notion (54:30)

For non-technical folks, the quick win version: plain Claude (not Claude Code), connected to Notion through Connectors, prompted to create a customer portal for the client with everything a good portal needs, and told to ask its questions first so it knows precisely what you want. It builds the pages natively in your Notion workspace.

Q&A highlights (1:01:42)

How do I add login authentication to a portal? Use Supabase: it handles authentication very well, and even no-coders manage it.

Can Claude Code connect to Zapier, n8n, or Make? Yes, but I find it a dumb use case: instead of having Claude Code build workflows, have it build the automation as code and host it directly.

Are custom-coded automations less maintainable than n8n workflows? They are just as maintainable. It is just code.

Is Antigravity as good as Claude Code for this? For workflow building, yes, it is on par. I use Claude Code because I have the budget and love the tool; if you are on a budget or a student, use Antigravity, no shame in that.

Opus or Sonnet in Claude Code? I always use Opus on the hundred dollar plan and only hit limits during very intense work.

Where do clients run this? They do not need Claude Code at all, only the deployed script, and Modal is about five dollars a month. Publishing the dashboard itself goes through Vercel.

Workflows versus scripts (1:02:49)

A big Q&A theme deserved its own diagram: should automations be n8n or Make workflows, or code? A no-code workflow is just a huge visual script; the nodes hide code underneath. LLMs are far better at writing raw code than at wiring nodes together, so asking an LLM to build an n8n workflow uses it below its potential. Instead, have it write the script and put it on the web with a platform like Railway or Modal. Modal is what I use: think of it as n8n but for a Python script, and it is nearly free, around five dollars a month. Your client does not need Claude Code, they only need the deployed script. Workflows are not dead: they are still the right choice when a team needs to see and edit the automation visually. Choose per customer and per use case.

Quick win 5: email thread analysis (1:08:30)

Last quick win: take your customer email threads, especially the tense ones, drop them into Claude, and ask it to analyze and propose improvements. On my synthetic angry-customer thread it flagged things like no visible follow-up, no specificity on timelines, and missed upsell signals, then produced actionable items. Connect Claude or Claude Code to your mailbox and it can go through the real threads itself. That is customer success too: your complaints archive is a service-improvement roadmap.

I also demoed my job monitoring dashboard (1:00:03), built over a weekend with pure Claude Code and Modal, no n8n: it tracks every job post mentioning AI tools like ChatGPT, Zapier, Claude, Copilot, n8n, or Make, with live notifications, company size, and a qualification for each, powered by a job post aggregator we covered in a dedicated earlier webinar.

That was the session: onboarding playbooks, automated reporting, voice reports, dashboards, a live-built client portal with a chatbot, and email thread analysis. If you want the next one live, or the full back catalog, head to the webinars page.