Every few months a business owner asks me whether a chatbot will let them cut their support team. The pitch they heard was simple: replace salaries with software, save money forever. The reality of chatbots vs live agents cost is more interesting, and more honest, than that pitch.
The right answer is almost never "all bots" or "all humans." It is a specific ratio for your specific volume and question mix. To find it, you have to actually run the numbers instead of trusting a vendor's slide. So let us run them.
I will use plausible figures for an Indonesian SME. Adjust to your own reality, but the shape of the math holds.
The Cost of a Live Agent, Per Conversation
Start with what you already pay. A competent support agent in the Jakarta area might cost around Rp 5,000,000 per month all-in once you count salary, BPJS, and overhead. Say that agent works about 22 days a month, roughly 8 hours a day.
A realistic agent handles maybe 6 to 8 conversations per hour when the questions are mixed. Take 7 as an average:
- 7 conversations per hour times 8 hours times 22 days is about 1,232 conversations per month.
- Rp 5,000,000 divided by 1,232 is about Rp 4,060 per conversation.
So a human conversation costs you roughly Rp 4,000 each. That number is your benchmark. Everything a bot proposes should be compared against it.
The Cost of a Chatbot, Honestly Counted
Vendors quote you the running cost and stay quiet about the rest. A fair chatbot cost has three parts: build, maintenance, and platform.
Build. A genuinely useful bot, one that answers your real questions rather than a generic FAQ, costs something to design and train. Call it Rp 15,000,000 to Rp 40,000,000 for a proper build tied to your catalog and policies. Amortize it. Over 24 months, a Rp 30,000,000 build is Rp 1,250,000 per month.
Maintenance. This is the cost everyone forgets. Products change, prices change, promos come and go. A bot that is not updated becomes a liar, and a lying bot costs you customers. Budget realistically for ongoing updates, say Rp 2,000,000 per month of someone's time.
Platform. Messaging API fees and hosting. For a mid-volume SME, perhaps Rp 1,500,000 per month.
So the bot costs roughly Rp 4,750,000 per month before it answers a single question. The magic is what happens as volume rises. Whether the bot handles 2,000 conversations or 20,000, that monthly cost barely moves.
| Monthly volume | Bot cost per conversation | Human cost per conversation |
|---|---|---|
| 2,000 | ~Rp 2,375 | ~Rp 4,000 |
| 5,000 | ~Rp 950 | ~Rp 4,000 |
| 20,000 | ~Rp 238 | ~Rp 4,000 |
The bot wins decisively on volume. But cost per conversation is only half the story.
Where Each One Actually Earns Its Keep
Cost is not value. A cheap answer that loses a sale is expensive. Here is the split I have seen hold up across retail, services, and F&B:
Bots win on repetitive, high-volume questions. Store hours, order status, shipping fees, return policy, "is this in stock." These are the questions that eat your agents' day and require zero judgment. A bot answers them instantly, at 2 AM, without getting tired or rude.
Humans win on conversion and complaints. When a customer is deciding whether to buy a Rp 2,000,000 item, a human who reads hesitation and offers reassurance closes deals a bot cannot. And when a customer is angry, a bot saying "I understand your frustration" often makes it worse. A real person who takes ownership saves the relationship.
The mistake is forcing one tool to do both jobs. Bots are terrible at empathy. Humans are wasteful on "what time do you open."
The Hybrid Model That Usually Wins
The design that actually saves money looks like this:
- The bot takes every incoming conversation first.
- It resolves the repetitive questions instantly, which is often 60 to 70 percent of volume.
- The moment it detects a purchase intent, a complaint, or its own uncertainty, it hands off to a human with the full chat history attached.
- Your agents now spend their time only on conversations where a human changes the outcome.
The result is fewer agents handling higher-value work, and a bot absorbing the drudgery. You are not replacing your team. You are aiming them.
This handoff logic is not something you should fully automate and walk away from. Knowing exactly where a machine should stop and a person should take over is its own discipline, and I wrote about it in Human in the Loop: Where AI Still Needs a Supervisor. If your support volume is still small, a bot may be premature entirely, which ties back to MVP Scoping: What to Cut First and What to Never Cut.
The Practical Takeaway
Do not ask "bot or human." Ask three questions:
- What is my true human cost per conversation? Calculate it before believing any vendor.
- What fraction of my conversations are repetitive and judgment-free? That fraction is your bot's job.
- Where does a human actually change the outcome, in sales and in complaints? Protect those conversations for your team.
On the real chatbots vs live agents cost comparison, the winner is almost always a hybrid tuned to your volume. Bots for scale, humans for the moments that matter. Build the handoff well and you spend less while your customers feel more looked after, not less.