How Fullscript Built 1,500 n8n Workflows in One

Quick disclosure before we get into it: this story comes from a conversation published on n8n's own YouTube channel. That's not a disqualifier — the tactical detail here is genuinely useful — but you should know the halo-glow is baked in. I'll flag where the numbers need more scrutiny.

Okay. So here's what actually caught my attention about this one.

Every few weeks I see another "we deployed AI company-wide" announcement and I click through expecting concrete details and get vibes. Templates! Buy-in! Culture! And then nothing. No texture. No honest accounting of what broke. I almost skipped this one. What made me stop was a single detail buried in the conversation: credential setup. Sahar Rahmani, Fullscript's Director of Internal AI, identified API credential onboarding — not the AI concepts, not the workflow logic — as the specific wall that stopped non-technical people cold. That's so specific and so true that I had to keep reading.

Fullscript is a health tech company with over a thousand employees handling PHI and PII data. They can't just plug every shiny new AI tool into their stack. Rahmani says they built over 1,500 n8n workflows in under a year, with more than 100 making it to production. Those are Fullscript's own figures — unverified externally — but the structural reasoning behind how they got there is worth unpacking regardless of the exact count.

The credential problem nobody talks about

Here's the thing about enterprise AI rollouts: everyone fixates on "will people actually use it" and skips the more boring question of "can people actually get started." Rahmani figured out that the real friction wasn't learning n8n. It was the moment a non-technical person hit "add credential" and suddenly needed to know what OAuth2 meant.

Her fix was pre-configured global credentials — click this, you have Gemini access; click that, you're connected to Slack — so the first workflow a new user builds doesn't immediately turn into a 45-minute hunt through API dashboards. She paired this with templates built around the most common actual requests she was seeing: "how do I pull data from somewhere and post it to Slack?" Turns out that covers like 60% of what most teams want automation to do.

The templates had sticky notes with step-by-step instructions. Ctrl+A level instructions. I know some engineers reading this are cringing, but that's the point — the goal wasn't to build power users overnight, it was to get people to their first successful run so they'd come back.

The hackathon was a feedback engine, not just a launch event

The company-wide AI hackathon sounds like standard enterprise rollout theater, but the way Rahmani describes it, it was actually functioning as demand sensing. They set up a dedicated room with a "Get Stuff Done" team whose only job was to help people build. And because everyone was in the same room at the same time, patterns surfaced fast. Rahmani noticed that a huge percentage of the solutions people wanted to build ended with "...and then send a Slack message." That insight directly shaped which templates she built next.

She also maintains a dedicated help channel where AI questions come in constantly. She reads it regularly. When the same question surfaces repeatedly, she turns it into a template. That's not a complicated system — it's just sustained attention, which most rollouts don't have because most rollouts don't have a dedicated person.

The workflows that actually moved the needle

Sales: The team was manually cross-referencing conference attendee lists against their customer base — existing customers, past contacts who never converted, and brand-new names. Per Rahmani, that process took close to a week per list. The n8n workflow does it in seconds. That's a genuine time-to-lead improvement, and in sales, speed of response has real compounding effects on conversion.

Finance: Invoice reconciliation. The kind of work that everyone pushes to tomorrow because it's tedious but consequential. Rahmani says this one now lives with the finance team directly — they own it, they run it, the AI team just provides support. That handoff model is interesting to me: build it, train the business owner, become support. It's how this kind of thing actually scales.

Customer support: After each call, agents were responsible for summarizing, tagging, and logging the interaction. Automating that post-call work doesn't replace the agent — it just means the agent can actually be present during the call instead of mentally drafting their summary while still talking to a customer. Rahmani described the feedback from the support team: "it help us to be more focused on the call rather than being worried to take a note." That's the version of "AI frees up humans" that I actually believe, because it's specific.

Legal: This is the one I found most technically interesting. Contract review is sensitive enough that Fullscript uses a local language model — running on-prem, not sending contracts to an external API. Rahmani also mentioned using n8n's evaluation node to validate outputs before anything reaches the legal team. Running local inference for compliance-sensitive workflows and adding an evaluation step is genuinely thoughtful architecture, not just "we connected GPT-4 to everything and hoped."

The compliance workflow — an automated process that strips PHI from recorded calls — is what Rahmani says (per Fullscript's own estimate) saved around 3,000 hours in a single year. I can't verify that independently, and you should read it as a company-supplied figure. But the mechanism is credible: if your team is manually reviewing calls for protected health information, automating that scan removes a real bottleneck.

The "hype alive" problem is the actual hard part

Here's what I think is the most underrated thing Rahmani said in this whole conversation:

"It's part of my job to make sure the hype is alive — like everybody still knows this is here, people using it."

Most discussions of enterprise AI adoption treat launch as the finish line. Build the templates, run the hackathon, done. Rahmani is describing something different: her job is perpetual re-enrollment. New templates every week or two. Town hall demos, specifically featuring non-technical employees who built things, because watching the security team showcase a workflow lands differently than watching the AI team do it. Dedicated workshops from n8n 101 through more advanced sessions, all recorded so late-joiners can catch up.

This is the part that doesn't show up in the playbook PDFs. It's ongoing editorial work — keep the signal coming, keep surfacing wins, keep lowering the activation energy for the next person to try something. It's the same logic as a newsletter: if you go quiet for six weeks, people forget you exist.

Which brings me to the number that should give everyone pause: Rahmani mentioned 77 active users per month. At a company with over 1,000 employees, that's under 8%. Now, she noted that it's not always the same 77 people — there's rotation — and the 1,500 workflows weren't all built by active monthly users simultaneously. But still. If your benchmark for successful company-wide adoption is "most of the company is regularly building," 77/month is a much more honest picture of where even a well-run rollout actually lands. That's not a failure; it might be exactly right for what automation work looks like in practice. But it's worth sitting with rather than skimming past.

The role itself is the strategy

Rahmani's title — Director of Internal AI — sounds like a lot of companies are creating this position. I'm skeptical that's true at scale, and this conversation doesn't give us a date for when she took the role, so I can't tell you whether she was ahead of the curve or riding it. What I can say is that the Fullscript model only works because someone owns it full-time. Not as a side project. Not as one PM's 20% time. Someone whose entire job is figuring out what's blocking adoption, building the next template, going to the finance team's offsite to show them what's possible, and answering Slack questions at 4pm.

The technology is almost secondary. n8n is a capable platform, but the reason Fullscript is at 1,500 workflows and most companies are at 10 isn't the tool. It's that they made AI adoption someone's actual job and gave that person real authority to move across departments.

The question every company watching this conversation should ask isn't "should we use n8n." It's: "who's the Sahar Rahmani for us, and do we actually have budget for them?"

— Yuki Okonkwo, AI & Machine Learning Correspondent