SaaS: What’s the role of Humans in Customer Service in 2025

Rodrigo Cardenete
Rodrigo Cardenete
Founder at BUNCH
Last Update:
October 8, 2024

Sam Altman, founder of OpenAI, made it clear back in 2022 when they launched ChatGPT: the first industry to get disrupted by AI models would be customer service, with up to 70% of tasks eventually automated.

SaaS can benefit from that. AI will have a profound impact on service quality if done well, leading to the magic word: retention.

Economy of scale and cost optimization are pushing companies to replace human agents with AI. It’s a bit unsettling, but also inevitable. At BUNCH we have been helping SaaS companies with Customer Support Outsourcing Services since 2017 and we know the value of great CSX.

But what’s the real cost? Can AI agents fully replace the premium experience of human touch?

In this article, we will explore:

a) What AI can and can’t do in customer service for SaaS

b) The role of humans

c) What to do if you run a SaaS and want to get the most out of AI + humans

A Word of Caution

AI bots are a huge help. They excel at understanding your knowledge base, examples, and user intent. They’re perfect for that first contact with your users—whether they’re paying or not. If done right, bots can bring down your CSX costs. And, of course, they’re available 24/7.

But we’ve also seen the downsides that can damage the customer experience if not managed properly. Disruption hasn’t come as fast as we thought. It’s happening slowly, but surely. Here’s why the transition is taking longer than expected:

  1. Limited trust and understanding of the technology
  2. SaaS products often are not well documented (at least not in a way AI can use)
  3. AI hallucinations are too frequent if not mitigated by inference
  4. Customer service isn’t always a priority—integrating it with existing processes takes time, especially with other things on the roadmap (if you’re in SaaS, you get this)

In fact, handing over all customer service to AI with today’s tech is causing a drop in customer satisfaction in 90% of cases.

So, how do you use it?

a) What Are AI and Humans Doing Today for SaaS Customer Service

AI is here to stay, and most companies can benefit from AI models trained on their guidelines, capable of replying 24/7 to customers. Sounds great for everyone.

But is AI enough? No, it’s not. AI agents hallucinate, make mistakes, and can even create legal liabilities.

We’ve noticed a shift in the types of inquiries that can be handled by AI:

  • What AI does well now: Features and Benefits Walkthroughs, How To’s, Basic Troubleshooting, Subscription plans, Compliance Queries, Onboarding Help
  • What humans handle now but AI will: Billing Enquiries, Impersonating User Journey, Data Migration, Database Queries, Product Demos, User Permissions
  • What only humans will do: Technical Issues (L3), Legal Discussions, Edge Cases Support, Contract Negotiations (Enterprise), Custom Features (Enterprise)

In essence, AI agents today excel at managing information that’s repetitive or standardized, based on rules pulled from your knowledge base, where low emotional engagement is required (they can’t go beyond the first few interactions). You don’t want AI in situations that demand judgment or empathy.


b) The Role of Humans: QA, Escalations, Sensitive Cases, and Personalization

Will humans lose their jobs to AI? That’s the question we keep getting from journalists and bloggers.

Bad news: there’s no simple answer.

From what we’ve observed, yes, some people will lose entry-level positions. But the role of humans in customer service will actually increase. As AI takes over most L1 and trivial cases, the average CSR will become more skilled, handling more knowledge-based work. Unfortunately, this isn’t great for entry-level reps (and this applies to most industries), but we believe it’s a net positive for the entire industry, especially SaaS.

The role of humans in SaaS customer experience:

  1. QA: Humans still need to audit a large portion of AI outputs (that’s industry consensus), especially in upper support levels.
  2. Escalations: Unfortunately, most LLM models tend to act overconfident and often can’t assess more complex problems beyond their initial detection.
  3. Sensitive Cases: You don’t want AI handling issues involving privacy, legal liabilities, confidentiality, or trade secrets. Those require human care (and probably shouldn’t be outsourced either).
  4. Personalization: Some customers need a customized experience. While AI agents can adapt to a user’s context, nothing beats a human in maintaining relationships—probably the one thing we’re truly unique at.

c) I Run a SaaS. What Should I Do?

There are simple approaches that will allow you to get best of both worlds:

1. Document (more than ever) your SaaS for Users and for Bots

Knowledge bases for SaaS are nothing new. If you’re in SaaS, you know how important documentation is for both your users and your team.

But there’s something new to consider: bots are about to become the primary readers of all your documentation. And they understand natural language.

Your AI bots will use this documentation to support users (and much more if you use AI agents to automate internal processes—but that’s a whole other topic for another day).

It doesn’t stop there. Your users and prospective clients will use their own bots to understand your product. People are using search tools like Perplexity (Google with AI answers and many more) to explore your SaaS features before they even land on your webpage. And those bots will learn from your knowledge base.

Write your documentation for both users and bots

Quick example: Some SaaS companies tend to hide known bugs or platform limitations, making them accessible only to internal reps. Documenting bugs is often seen as exposing flaws.

But this will backfire. You’ll end up escalating those cases to human agents, which is exactly what you want to avoid. Make that info available to bots from the start to reduce your reliance on human CSRs.

2. Try Different AIs

Your current ticketing system probably has or will soon have AI integrated (likely just a wrapper).

That doesn’t mean it’s the best fit for you, and it doesn’t mean you should just settle for their chatbot.

You may need an external AI agent or even build your own. Test what’s best for your SaaS, not just what’s easier within your current tech stack.

3. Use Human Agents in the Loop 24/7

It sounds obvious, but don’t leave an AI model alone with your customers. AI agents are great for L1 inquiries, but you need to be ready to escalate to humans when certain triggers are fired. AI will handle the bulk of the volume, but humans are necessary to manage edge cases.

Set up a team of 24/7 agents working alongside your model. You’ll get the best of both worlds.

By the way, human 24/7 teams sound expensive to build, but we solved this challenge long ago by creating Shared 24/7 Customer Support SaaS Teams. Basically, you get a team trained on your product that covers 24/7 support, but you only pay for the time they’re actively used, optimizing for idle time. This is the perfect use case for combinining humans with AI.

4. Use Human Inputs to Fine-Tune Your Models Progressively

This is a pro-tip: Use human interactions with real customers (QA audits, escalations, and human-handled tickets) to fine-tune your model and make it smarter.

Training datasets are the foundation of machine learning. Typically, they’re massive collections of data points collected from various sources, often unrelated to your product. The model understands your documentation (during inference or retrieval), but it’s not really trained on your specific data.

Every interaction with a real user is a chance to change that.

Store every single interaction with your clients and use it to fine-tune your model

Over time, you’ll have a model that not only understands your documentation but has also been trained on actual data—real tickets and chat sessions—getting smarter with every interaction.

About the Author

Rodrigo Cardenete
Rodrigo Cardenete
Rodrigo is co-founder of BUNCH. With background in design, operations and development, he has taken different roles as COO and CMO

Stay in the Loop!

Subscribe to our newsletter and get the latest updates, exclusive content, and insights on Data Ops, Machine Learning, and emerging tech startups.

Related Content

A Brutal Disruption in Image Annotation Services in 2025

AI data labeling just got disrupted. Generalist models are out, and expert-driven, specialized data is in. Here’s how the landscape is evolving faster than anyone expected.

How AI Spawned "Second-Gen Bias" in Content Moderation Services

AI created a whole new layer of bias nobody saw coming: second-gen bias. Learn how this hidden flaw impacts content moderation and why humans are key to keeping AI in check

How BUNCH Became a 24/7 Operations Powerhouse

Our 24/7 outsourcing services ensure seamless, efficient operations for businesses worldwide. From shift scheduling to cultural sensitivity, we guarantee continuous support in all time zones.

How We Are Obsessed About Data Quality and Why

We understand the importance of reliable data quality for training datasets and precision in moderating user-generated content. Learn how we apply rigorous QA in all our processes.

Our Managed Services Model

Learn how our managed services are designed to shoulder all operational responsibilities, offering clients streamlined, process-based operations under a flat monthly fee, allowing them to focus on growth.