What does an AI chatbot for customer service cost?
An AI chatbot for customer service costs SEK 5,000–20,000 per month as an off-the-shelf platform service, or SEK 400,000–1.5 million to build custom. The choice comes down to case volume, integration needs and the demand for control. Always factor the cost of a bad bot, more escalations and damaged trust, into the calculation.
For an AI chatbot there are two buying paths with completely different economics: subscribe to an off-the-shelf platform, or build your own solution. Both can be right – but they solve different problems, and choosing a path on the wrong criterion means you either pay too much or get a bot that can’t do the job.
The two buying paths in kronor
| Buying path | Typical cost | Fits when |
|---|---|---|
| Off-the-shelf platform | SEK 5,000–20,000/month | General questions, standard flows, a fast start without integrations |
| Custom-built solution | SEK 400,000–1,500,000 + operations | The bot must reach your systems, handle customer data and follow your rules |
The platform is cheap to start but limited to what the platform can do: answer from your content and follow predefined flows. The custom-built bot can look up the customer’s actual order, change bookings and follow your tone and your business rules – but it requires an investment and an operating budget.
Breakeven: do the math on case volume
The honest way to compare the paths is per case. A worked example:
Suppose your customer service handles 4,000 cases a month and that an average case costs SEK 60 in staff time to handle manually. If a platform bot at SEK 12,000 a month resolves 30 percent of the cases, it saves 1,200 cases × SEK 60 = SEK 72,000 per month. An easy deal.
Now suppose instead that a custom-built bot at SEK 900,000 resolves 55 percent thanks to integrations that let it actually do things – then you save SEK 132,000 a month, minus say SEK 15,000 in operations. The extra saving over the platform is SEK 45,000 per month, and the extra investment is earned back in under two years.
Flip the numbers for a smaller company with 400 cases a month: the platform’s saving becomes SEK 7,200 per month – barely above the platform cost – and the custom-built bot never adds up. Volume decides. Run your own numbers before you request any quote.
The cost of a bad bot belongs in the calculation
The decision basis isn’t complete without the downside. A bot that answers wrong, misunderstands or traps users in loops with no route to a human costs in three ways:
- More escalations, not fewer. The customer who has already explained their case to an uncomprehending bot starts over with an agent – now irritated. The handling time per escalated case goes up.
- Brand damage. The chatbot is often the customer’s first encounter with your service. A bad bot signals that you’re deprioritizing the customer, and screenshots of absurd bot answers spread readily.
- Incorrect statements. A bot that promises the wrong price or the wrong terms creates cases that are more expensive than the ones it was meant to solve.
This is the strongest argument against choosing the cheapest possible solution for complex needs: the difference between a bot that resolves 30 and 55 percent shows up in the calculation, but the difference between a bot that helps and one that irritates shows up in your customer base.
Don’t forget the operating items in both tracks
Whichever path you take, there are ongoing costs beyond the license or the build. For the platform: any volume surcharges and the internal work of keeping the knowledge content current – a bot that answers based on last year’s terms does more harm than good. For the custom-built one: model calls that scale with conversation volume, hosting, monitoring and maintenance, often SEK 10,000–30,000 a month for a mid-sized solution. The content work is easy to underestimate in both tracks: someone on your side has to own the bot’s knowledge, review the logs and feed in what’s missing.
How to choose your path
Start with your case data: what questions come in, in what volumes, and how many require lookups in your systems? If most are general questions, a platform is often enough – start there and learn from the logs. If the most common cases require customer data, the platform is a dead end, and then it’s better to go straight for a custom solution, ideally via a contained pilot. At Weapp we’ve built customer-facing digital services like Sejfa and are happy to help with the breakeven math for your case flow – get in touch.
Frequently asked questions
When is an off-the-shelf chatbot platform enough?
When the questions are general, the volume is moderate and the bot doesn't need to do things in your systems. A platform that answers from your FAQ and website goes a long way for common questions about opening hours, deliveries and terms. If the bot needs to look up a customer's case or change a booking, integrations are required, and then the math changes.
What is included in the price of a custom-built chatbot?
Conversation design, a connection to your knowledge base, integrations with case and business systems, permissions, escalation flows to human support, testing and operations. The integrations and testing are usually the largest items; the language model itself is a smaller part of the cost.
How do you measure whether a chatbot is good?
Track the resolution rate without human involvement, the share of correct answers on a test set, customer satisfaction in rated conversations, and the share of escalations that happen smoothly. Measure before and after launch; a bot that only moves contacts from email to frustrated chats hasn't solved anything.
Can the chatbot give wrong answers about our terms?
Yes, any language-model-based bot can answer incorrectly. The risk is managed by having the bot retrieve answers from your documentation instead of its general memory, with clear limits on what it may answer and with regular review of the logs. For binding statements, the bot should refer to a human.
How long does it take to launch an AI chatbot?
A platform-based bot can be up and running in 2–6 weeks including content work. A custom-built solution with integrations usually takes 3–6 months. The content, making sure your knowledge base is correct and current, is often more time-critical than the technology.