Crafting Excellence in Software
Let’s build something extraordinary together.
Rely on Lasting Dynamics for unparalleled software quality.
Mohamed Boukrim
Nov 11, 2026 • 10 min read

A man whose grandmother had just died asked Air Canada's website chatbot about bereavement fares. The bot told him he could book now and claim the reduced fare afterward, within 90 days of the ticket being issued, while the airline's own bereavement page said such fares could not be claimed retroactively. When the refund was refused and the case reached the British Columbia Civil Resolution Tribunal, Air Canada argued that the chatbot was a separate legal entity responsible for its own actions. In February 2024 the tribunal called that “a remarkable submission” and ordered the airline to pay C$650.88 in damages, plus interest and fees.
The software was not really the problem, because nobody had decided which questions the bot could answer alone, which sources it had to answer from and when it should hand the customer to a person. An AI chatbot for business pays off when someone writes it a job description, the way you would for a new hire, and the companies that skip that step end up with a bot their customers try to get past.
This article is for the people who decide whether to buy one. It covers what a good chatbot does, the five questions that make up its job description, what the invoice really measures, what the law now requires and when it is worth having a chatbot built rather than bought.
A chatbot earns its place with three things a team of people struggles to offer: it answers at any hour, in the customer's language, and gives the same answer the hundredth time as the first. Every routine question it closes is one your staff no longer have to answer, which leaves them the conversations where a person changes the outcome, such as the complaint, the exception or the customer about to leave.
European companies are moving, and according to Eurostat, 20.0% of EU enterprises with ten or more employees used AI in 2025, up from 13.5% a year earlier. What holds the others back is more telling. Among firms that had considered AI but did not adopt it, only 20.68% said it was not useful for their business, while 70.89% cited a lack of relevant expertise (Eurostat, Statistics Explained), which makes know-how, not value, the real obstacle.
So treat your AI chatbot for business as a hire, since a good new employee arrives with a defined role, a manual to answer from and a colleague to call when a question is above their pay grade, and a chatbot needs the same three things. The five questions that follow make up its job description, covering whether it will answer or act, which conversations it takes first, how you pay for it, how it tells customers it is a bot and whether you buy it or have it built.
Start with what the bot is allowed to do, because that shapes every other choice. A chatbot answers: it tells the customer the opening hours, the status of an order, the returns window or which plan fits their team. An AI agent acts: it issues the refund, moves the booking or changes the delivery address inside your own systems.
The difference sets the price of an error, since a wrong answer leaves a misinformed customer, which a follow-up message can fix, whereas a wrong action leaves a transaction to reverse. Many products sold as chatbots today can do both, so where the line sits is a business decision, not a setting to leave at its default. Most companies start with answering and add actions one at a time, once the answers have proven reliable.
The best first jobs come up often, carry little risk and already have a written answer somewhere you trust, whether a FAQ, a price list, a policy page or the order system. Delivery times, opening hours, password resets and warranty terms all qualify. A bot answering from those sources is fast and consistent, and you can check its work against the document, which is where an AI chatbot for customer service pays for itself first.
The conversations to keep with people at the start are the ones where the honest answer is “it depends”: exceptions, complaints, refunds outside policy and anything emotional. That is exactly where the Air Canada bot went wrong, and it was not alone.
Let’s build something extraordinary together.
Rely on Lasting Dynamics for unparalleled software quality.
In April 2025, Cursor's support bot told users they could no longer use the product on more than one device, a policy that did not exist, and the AI Incident Database records that the answer reportedly led to subscription cancellations. Neither case is an argument against chatbots, but both teach the same lesson. In each case the bot answered a question it had no written source for, on a topic that needed a person, and a business answers for its bot as it does for its staff.
The fix therefore belongs to the business, which has to give the bot a source for every answer, permission to say “I don't know, let me get a colleague” and a visible way to reach a person. That handoff is part of the product, not an admission of defeat. Our guide to chatbot best practices covers how to design those flows, and our piece on AI customer service looks at how the whole support team changes once AI takes over the routine work.

One test helps before you delegate anything: would you let an employee in their second week answer this alone, without checking with anyone? If yes, it is a good job for the bot; if not, it is not ready yet.
Help-desk platforms such as Intercom and Zendesk now base chatbot pricing on results: you pay when the bot resolves a conversation, on top of a monthly fee per seat for your human staff. That favors the buyer, because the vendor earns only when the bot does its job, as long as you know how the vendor defines a result.

Intercom's AI agent, Fin, costs $0.99 per “outcome,” charged once per conversation, on top of seats at $29, $85 or $132 per month depending on the plan, with a discount on the cheapest plan for new customers. Intercom publishes the definition on Fin's pricing page, where a service outcome counts when no further help is requested after Fin's last answer or when Fin completes a procedure you have set up to end in a handoff to a person. A conversation simply passed to your team without an outcome is not charged.
Zendesk also charges per automated resolution and, since May 2026, has an AI model review each conversation once it ends, so that only the ones it verifies as resolved count against your allowance. Both definitions are reasonable, but “no further help requested” covers the customer who found the answer and the customer who gave up, and the invoice cannot tell them apart. You can, by tracking how many customers come back with the same question within a few days. If that number stays flat while resolutions rise, the outcomes are real.
To estimate the cost, skip the price ranges you will find online and do the arithmetic with your own numbers: conversations per month, times the share the bot actually handles, times the price per outcome, plus seats, plus the hours someone spends keeping the answers current. As an illustration, 2,000 conversations a month with 40% handled by the bot is 800 outcomes, or $792 a month at $0.99, before seats.
Upkeep is the cost people forget and the lever that matters most. A bot is only as current as its sources, so every price change, new product or updated policy has to reach them, or it will answer confidently from last quarter's rules, which is why that content needs a named owner from day one.
We read these prices on the vendors' own pages on 10 November 2026. They change often, so check them before you sign.
From idea to launch, we craft scalable software tailored to your business needs.
Partner with us to accelerate your growth.
In the European Union, a chatbot has to say so. Article 50(1) of the AI Act (Regulation (EU) 2024/1689) has applied since 2 August 2026, and it requires that people be told they are interacting with an AI system, clearly and from the first interaction, unless that is obvious to a reasonably well-informed person. The European Commission's FAQ on Article 50 says the “obvious” exception should be read restrictively, and the only grace period in the article concerns the marking of AI-generated content, not this duty.
The obligation is easy to meet, because it formally sits with the provider, the company that builds and sells the system, so a reputable off-the-shelf tool should already support it. What you control is the configuration, and that is where it can slip. A human first name, a photo of a staff member or an opening line like “Hi, I'm Sam from the team” can undo it. An honest name and a first message that says “I'm the company's AI assistant” do the job. If you build your own chatbot, you are the provider, and the disclosure is yours to design.
Breaching the transparency obligations can cost up to €15 million or 3% of worldwide annual turnover, with proportionality for small and medium-sized companies. The same idea is spreading beyond Europe: in October 2026, New York City proposed rules that would require businesses to disclose automated customer service and give customers an easy route to a human agent. None of this is legal advice: our piece on AI governance covers the decisions a company has to make about its own use of AI, including what the AI Act, as amended in 2026, asks of it, and it is worth asking your vendor, in writing, how disclosure is handled.
If the first job of your AI chatbot for business is to answer questions from your own documents, an off-the-shelf tool is usually the fastest route to value. The vendor brings the expertise and designs the disclosure, which covers two of the barriers Eurostat found most often among European companies that held back, namely lack of expertise (70.89%) and unclear legal consequences (52.52%). The third, data protection (48.83%), is where the choice gets more interesting.
The table below is not a ranking of the best AI chatbot for business but a map of what each kind of tool is for.
| Option | Good for | How you pay | Who designs the AI disclosure | When it stops being enough |
|---|---|---|---|---|
| AI add-on to your help desk (e.g., Intercom Fin, Zendesk AI agents) | Teams already on that help desk; answering from help-center content | Per resolution, plus seats | The vendor; you configure it | When the bot must act in systems the help desk can't reach |
| Website chat tool (e.g., Tidio, ChatBot.com) | Pre-sales questions and simple FAQs on smaller sites | Monthly subscription; check how conversations are counted | The vendor; you configure it | When answers need live data from your own systems |
| General AI assistants (e.g., ChatGPT, Claude) | Your staff: drafting, research, summaries | Per user, per month | Not designed as a customer-facing chatbot | They were never meant to face customers |
| Custom AI chatbot | Acting inside your systems; regulated data; high volume | Build cost, plus running and maintenance | You, as the provider | It grows with the job when it is built to |
A custom AI chatbot becomes the better investment when the bot has to act inside your own software, for example checking live stock, changing a booking or opening a claim, or when the data it handles, such as health, financial or insurance records, has to stay in your own infrastructure. It also pays off once volume grows to the point where monthly outcomes times the price per outcome exceed what it costs to own and maintain a bot of your own.
For many companies, that is the second step rather than the first, because the off-the-shelf bot proves which conversations are worth automating, and a custom one, built through AI software development, takes on the ones that need real access to your systems. That step goes faster when a first bot has already shown what your customers actually ask.
Five answers give you a plan, and one month gives you the evidence. Pick a single conversation type from your list of good first jobs, one rather than three. For the first two weeks, let the bot draft and have a person approve every answer before it goes out, since many help-desk AI tools offer a draft or copilot mode for exactly this. In weeks three and four, let it answer alone and have someone review a sample every day.
Over the month, keep track of the conversations the bot handled and those it handed to a person, the customers who came back with the same issue, the complaints that mention the bot, and the invoice set against the staff hours saved. If the invoice is lower than the cost of the hours saved and repeat contacts and complaints did not rise, give the bot its second job. If not, fix the sources and run another month; either way you will have spent weeks, not a year, finding out.
We design and build high-quality digital products that stand out.
Reliability, performance, and innovation at every step.
The businesses that get real value from an AI chatbot for business did not buy the cleverest one. They gave it a narrow job, good sources and an easy way to say “let me get a colleague,” and once it had earned it, they gave it the next job.
Does your chatbot need to do things inside your own systems, not just answer questions? Tell us what it has to do.
This article was written by Mohamed Boukrim at Lasting Dynamics, an EU-based custom software development company that also builds AI software. It is written for the people who decide whether to put an AI chatbot in front of their customers, and the prices and rules it cites are dated and linked so you can check them before you sign.
It is software that answers customers' or staff's questions in plain language on your website, app or messaging channels, using your own documents as its source. When it can also do things in your systems, such as issuing a refund, it is usually called an AI agent.
Help-desk tools such as Intercom and Zendesk charge for each conversation the bot resolves, plus a monthly fee per seat, and Intercom's Fin costs $0.99 per outcome (read on 10 November 2026). For your own estimate, multiply monthly conversations by the share the bot handles and the price per resolution, then add seats and upkeep.
A chatbot answers questions, while an AI agent takes actions inside your systems. A wrong answer can be corrected with a message, but a wrong action has to be reversed.
In the EU, yes. Since 2 August 2026, Article 50(1) of the AI Act requires people to be told they are dealing with an AI system unless it is obvious. The duty sits with the provider, but a human name and a staff photo can undermine it, so configure your bot honestly.
Buy first if the bot's job is to answer questions from your own documents. Build when it has to act inside your own systems, keep regulated data in your own infrastructure or run at a volume where per-resolution fees exceed the cost of owning it.
Transform bold ideas into powerful applications.
Let’s create software that makes an impact together.
Mohamed Boukrim
I am a Software Engineer and Backend Developer with a relentless focus on software quality and robust architecture. I don't believe in shortcuts; outstanding results are the direct product of daily commitment and hard work. Alongside my team, I leverage Agile methodologies and continuous process evaluation to push boundaries and optimize performance. I approach every challenge with a highly competitive mindset: I work tirelessly to reach the top, and once there, I keep pushing to maintain the lead.