Dear SaaStr: We’re thinking about deploying AI for customer support. Is it actually good enough yet, or should we wait?
Don’t wait. Do it now. Pick a leader and go.
Just don’t forget to train it. You have to train AI support with your real data, FAQs, product data, etc.
AI support isn’t perfect. It still hallucinates sometimes. Again, it has to be trained to be any good. It can’t handle every edge case. Your power users will still need to escalate to humans.
But here’s what we’ve learned deploying almost 20 AI agents across SaaStr and watching many portfolio companies do the same, as well as 1,000+ AI support deployments at SaaStr Fund porfolio company Gorgias:
AI support that’s available 24/7 and answers 60-70% of questions well is infinitely better than human support that doesn’t exist at all.
Think about it this way as a continuum … from Worst to Best:
No support at all.
Terrible. Your customers are stuck. They can’t get unstuck. They churn. You never even know why.
Support only from basic, untrained bots.
Still awful. You know the ones — they can barely parse what you’re asking, loop you through the same 3 FAQ articles, then give up. Frustration compounds. At least customers know someone theoretically works at your company. But that’s about it.
No live support — email only with 12+ hour response times.
Getting an email back tomorrow? A fundamentally suboptimal experience. Sure, you eventually get an answer. But your customer’s problem is happening right now. They’re blocked right now. They’re evaluating your competitor right now.
The data backs this up. When we looked across Gorgias’s 13,000 ecommerce customers, the average response time was an abysmal 12 hours. Even worse? 44% of those companies thought they provided excellent customer support — but only 1.3% actually did based on objective metrics.
Most of us are terrible at support. We just don’t know it.
Try your own support yourself. Calmly. And on off hours. You’ll be shocking how mediocre it is.
Well-trained AI support available 24/7/365.
Now we’re getting somewhere. Here’s what modern AI support actually delivers when properly implemented:
- Instant response. Not in 12 hours. Not in 1 hour. In seconds. At 2am on Sunday. On Christmas. Always.
- 60-70% resolution rate for well-trained systems. Is that 100%? No. But it’s 10x better than the 6% resolution rate of traditional chatbots. And it’s handling the bulk of tier-1 questions that used to clog your support queue.
- Consistent quality. AI doesn’t have bad days. It doesn’t get tired on the 47th ticket. It doesn’t ghost you because it’s overworked and looking for a new job.
- Perfect memory. It knows every conversation this customer has ever had. Every ticket they’ve opened. Every feature they use. Every pain point they’ve mentioned.
The game-changer: You can actually provide support now.
Here’s what I see happening across our portfolio:
Before AI: Startup at $500K ARR. Founders doing support themselves. Response time: 4-6 hours during working hours, next-day on nights/weekends. Coverage: Maybe 60 hours per week. Cost: Founder time + massive opportunity cost.
After AI: Same startup. AI handles 70% of inbound. Response time: Under 30 seconds, 24/7. Coverage: 168 hours per week. Cost: $500-2000/month depending on volume. Founders now only handle the 30% that need human judgment.
That’s not a marginal improvement. That’s a 10x improvement in customer experience.
Live support from outsourced, properly-trained humans.
Still great! Quality varies based on training. Probably not quite as deep as your in-house team. But getting a good answer from a human right now may be 100x better than waiting a day for email. Costs roughly $15-35/hour per agent.
Live support from your in-house, highly trained team.
The best in theory. Your team knows your product cold. They know your customers. They can make judgment calls about edge cases. They can spot patterns and feed them back to product.
But can it scale? Can you afford 24/7 coverage with humans? Can you staff Christmas Eve? Can you handle 10x growth in support volume?
For most companies under $10M ARR: No.
The hybrid model that’s winning: AI + Human escalation.
Here’s what the best companies are doing:
Tier 1 (60-70% of volume): AI handles it entirely. Password resets, basic how-to questions, billing inquiries, feature explanations, common troubleshooting.
Tier 2 (20-30% of volume): AI attempts, customer explicitly asks for human, or AI recognizes it can’t handle it and escalates immediately. Human agent picks up with full context of the AI conversation.
Tier 3 (10% of volume): Complex technical issues, feature requests requiring judgment, angry customers who need empathy, true edge cases. These go to your most experienced team members.
The result: You’ve just made your support team 3-4x more efficient while simultaneously improving response times from hours to seconds.
“But what about quality?”
The objection I hear most: “AI will give wrong answers and damage our brand.”
Fair concern. Here’s the reality:
Yes, AI makes mistakes. It hallucinates. It misunderstands context sometimes. It can’t read between the lines the way humans can.
But humans make mistakes too. Lots and lots of mistakes. Your tier-1 support person on their 47th ticket of the day? They’re making mistakes. Your outsourced team that hasn’t been properly trained on your new feature? They’re giving wrong answers.
The difference: You can measure AI’s accuracy. You can train it on every ticket. You can update its knowledge base in real-time. You can A/B test different approaches. You can set confidence thresholds and auto-escalate anything uncertain.
With humans, quality is inconsistent and hard to scale. With AI, quality is consistent and improves over time.
The math that matters:
Let’s say you’re at $2M ARR with 200 customers:
Scenario 1: No AI
- Human-only support: 2 full-time agents at $60K each = $120K/year
- Coverage: 50 hours/week (no nights, limited weekends)
- Average response time: 2-4 hours during coverage, next-day otherwise
- Tickets handled: ~4,000/month
- Cost per ticket: ~$30
Scenario 2: AI + Human
- AI platform: $24K/year
- 1 human agent for escalations: $60K/year
- Total: $84K/year
- Coverage: 168 hours/week (24/7)
- Average response time: 30 seconds for tier-1, 1 hour for escalations
- Tickets handled: ~6,000/month (can handle growth)
- Cost per ticket: ~$14
You just saved $36K while improving coverage and response times.
And that’s before you account for the value of founder time back.
“Should we wait for AI to get better?”
No.
This is like asking in 2010 whether you should wait to build a mobile app until phones get better.
AI support is already good enough to deploy. It’s improving every month. The companies deploying it now are:
- Learning what works
- Building competitive advantages in support quality
- Freeing up human agents to focus on complex, high-value interactions
- Actually collecting data to train their AI better
The companies waiting are falling behind.
How to actually do this well:
1. Start with your FAQ/knowledge base in pristine shape. AI is only as good as what you train it on. If your docs are outdated or incomplete, your AI will be too.
2. Set clear escalation rules. Define exactly when AI should hand off to humans. When a customer asks for a human. When confidence is below X%. When sentiment is negative. When it’s a billing issue over $Y.
3. Monitor religiously for the first 30 days. Review every conversation. Identify patterns in failures. Update training data. This is not “set and forget.”
4. Measure the right metrics:
- First response time (should be seconds)
- Resolution rate (% resolved by AI without escalation)
- Customer satisfaction on AI interactions
- Escalation rate and why
- False positive/negative rate
5. Keep humans in the loop. AI handles volume. Humans handle judgment, empathy, and edge cases. The best support orgs use both.
There’s No Excuse Not To Have Pretty Good 24×7 Support Now
Perfect support doesn’t exist. Even with an army of highly-trained humans, you’re going to have gaps, delays, and mistakes.
The question isn’t whether AI support is perfect. It’s whether AI support is better than what you’re doing now.
For most companies under $10M ARR: It absolutely is.
- You can’t afford 24/7 human coverage
- You can’t scale human support fast enough with growth
- You can’t consistently train humans on every product update in real-time
- You can’t afford to have founders doing tier-1 support
AI support that’s there 24/7 and handles 70% of questions well is infinitely better than human support that doesn’t exist at all.
Start with AI for tier-1. Keep humans for escalations and complex issues. Measure everything. Iterate weekly.
Your NPS will thank you. Your churn rate will thank you. And your customers who need help at 2am on Sunday will definitely thank you.
A bit more on this here.
If your competition doesn’t offer instant support,
But you do …
Watch your NPS eclipse theirs, all things being equal.
Isn’t that worth a few thousand bucks a year?
