One of the biggest misconceptions about AI is that its value lies only in external product features, chatbots, personalisation, content generation. But in reality, some of the highest-leverage uses of AI and automation happen behind the scenes.
In B2B SaaS )especially when selling into public-sector domains like the NHS) a huge portion of operational complexity happens in the onboarding phase. Every new client brings a deluge of data, most of it inconsistently formatted, unstructured, and often dumped as raw folder trees or screenshots. And if your product is replacing decades of legacy processes (as ours is), the problem compounds.
This used to be a massive manual burden. But recently, I’ve been building internal automations using n8n and lightweight AI agents, with transformative results.
The Problem: Data Onboarding Hell
Let’s take one real example:
When onboarding a new NHS Trust, our implementation team typically receives hundreds of folders, one for each staff member, containing a total of up to 68,000 files. The team’s job? Manually search each folder for a specific key document: a performance review, a scanned contract, or a roster file buried somewhere in the mess.
Even for our most experienced team members, this process could take 2 to 3 weeks of focused effort per client. That’s not just a cost problem, it’s also a morale and velocity problem. You’ve got smart people wasting their time on robotic, soul-draining tasks.
The Solution: Automation in Layers
To solve this, I built an n8n workflow that layered classic automation with agentic AI to replicate the task, only faster, cheaper, and at scale.
Here’s how it works:
Folder Traversal: Once triggered, the flow scans each folder individually, looking for files that match specific patterns based on name, type, or metadata.
File Inspection: When a match is suspected, an AI agent (often running via a hosted OpenAI node) opens the file, even if it’s an image, and extracts key content.
Sentiment/Context Matching: Using a pre-written prompt and internal knowledge base, the AI classifies whether this is the correct file or just noise.
Standardisation: For confirmed matches, the file is renamed into a consistent format and moved to a clean output folder.
Reporting: Once the scan completes, a human-readable report is generated, showing matches found, matches missed, and folders still requiring manual review.
What once took 2.5 weeks now takes a few hours, and most importantly, the output is structured, traceable, and easily reviewed.
What This Actually Changed
This wasn’t just a time saver. It was a mindset shift.
The implementation team regained focus. Instead of drowning in folders, they could focus on client comms, risk flags, and onboarding quality.
The business saved thousands per implementation especially important at scale.
Confidence in delivery increased because the output was consistent, and we could audit every decision the automation made.
How Far Can You Push It?
Once you’ve built the flow, you can take it further. For example:
Attach a webhook to expose the automation through a lightweight Lovable front-end, letting non-technical team members use it safely.
Chain it into more complex flows, e.g., once all files are matched, trigger a prefilled Airtable record or upload to a client workspace.
Use embedded memory or vector stores to let the AI “learn” your match patterns over time.
Limitations and Gotchas
This isn’t magic. It takes work.
n8n has a learning curve. You need some comfort with conditional logic, scripting, and flow design to build more advanced use cases.
Not all workflows are suited for n8n alone. Sometimes you need to plug in other tools (e.g., OCR, regex filters, hosted AI functions).
AI sometimes gets it wrong. That’s why the final output still includes a manual review list.
But even with these trade-offs, the value has been massive. This isn’t about replacing jobs, it’s about removing the parts of the job nobody should be doing anymore.
Final Takeaway
When people talk about AI in product design, they usually mean something flashy - chatbots, voice assistants, personalisation engines.
But here’s the truth: some of the most powerful uses of AI are invisible. They’re buried in your internal tools, sitting quietly between your folders, removing the friction that used to steal your team’s energy.
If you’re serious about scaling your product org, not just the product itself, this is where you start.