Every few months a new vendor emails the firms I used to build for, offering an AI chatbot that will supposedly turn their website into a lead machine. The technology has improved; the pitch has not changed; and for most New Zealand accounting and advisory practices, the answer is still no. Here is what a chatbot actually does on a financial services site, the specific risks it carries in this sector, and the narrow cases where it is worth a second look.
The pitch is good. A chatbot on your website, the vendor explains, means you never miss an enquiry again. It works around the clock, greets every visitor, answers their questions instantly, qualifies the serious ones, and books them in - all while you are asleep or in a meeting. For a firm that worries about leads slipping away after hours, it sounds like cheap insurance.
I will state my position plainly, because that is what this site is for: for most New Zealand accounting and advisory firms, a website chatbot solves a problem they do not have, and introduces several they did not previously have. There are exceptions, and I will get to them. But the default answer for a small professional-services practice is no, and it is worth understanding why before a vendor talks you into a monthly subscription for one.
Start with traffic. A three-partner accounting firm or a solo financial adviser does not run a high-traffic website. Most days it sees a modest trickle of visitors, and a large share of those already know the firm - they were referred, they are checking you out before they call, or they are an existing client hunting for your phone number. This is not an e-commerce site fielding hundreds of cold shoppers an hour.
Drop a chatbot into that and what you get is a widget that mostly sits idle, then occasionally intercepts someone who would have rung you anyway. Worse, it invites questions it cannot answer well. Can you handle a trust return? Should I restructure my mortgage? A person knows how to field those. A bot either guesses or stalls, and either way the visitor leaves less impressed than when they arrived. The realistic traffic behind these sites does not support the premise that you are quietly bleeding after-hours leads a bot would have caught.
Here is the problem unique to this sector. A visitor types a question that drifts toward financial, tax, or mortgage advice - they nearly always do - and an AI chatbot, built above all to be helpful, answers it. Now you have a tool on your own website improvising guidance in a regulated domain, under your firm's name.
For a financial advice provider, that is a direct compliance hazard: the line between general information and regulated financial advice is exactly the line an eager language model is worst at respecting. Even outside the advice regime, an answer that turns out to be wrong is a misleading-conduct risk and, more immediately, a credibility problem. Professional firms sell judgement and accuracy. A confidently incorrect bot undercuts the very thing a prospective client arrived to assess. The regulator does not care that the chatbot said it - the firm owns what its website tells people.
The second sector-specific issue is data. Everything typed into a chat box is personal information, and on a financial firm's site it is often sensitive - income, debts, a business situation, sometimes a name attached to all of it. Most chatbot products are offshore software-as-a-service, which means those transcripts leave the country, sit on servers you do not control, and in a growing number of cases are used to train the vendor's model.
The Privacy Act 2020 has plenty to say about that - collection, storage, and especially disclosure to an overseas provider. The practical point is simpler than the law: before any chat tool goes on your site, you should be able to say where the data lives, who can read it, how long it is kept, and whether it trains an AI. In my experience firms install the widget first and ask those questions never.
None of this means hiding from your visitors. The opposite. For a relationship-driven, high-trust decision like choosing an accountant or adviser, the goal is to make reaching a human as easy as possible - not to deflect people into a chat queue.
That means the obvious things, done well: a phone number visible on every page, a contact form that actually works rather than quietly failing, and, increasingly, a booking link that lets someone put a first conversation in the diary without the email back-and-forth. People evaluating a firm to hand their finances to do not want to negotiate with a widget. They want to know a real person is on the other end, and exactly how to reach them.
Look at what a chatbot would actually be asked on a firm's site and a pattern emerges: fees, services, location, how to get started, what to bring to a first meeting. These are not dynamic questions that need a conversation. They are the same handful of things, every time.
Which means the better tool is content. A services page that explains plainly what you do and who you do it for, and a genuine FAQ that answers the predictable questions once, serves every visitor instantly - with no privacy exposure and no risk of an improvised wrong answer. It has a side benefit, too: that material is exactly the sort of specific, useful content that helps you show up in search. The work people imagine a chatbot doing is, almost always, work that is better done as a well-written page that sits there earning its keep.
To be fair to the technology, there are situations where chat earns its place. A larger brokerage or a multi-office firm with genuine website volume may field enough repetitive enquiries that triage is worth automating. A tightly-scoped assistant that does one job - book a call, route a message to the right office, take a callback request - can be useful precisely because it is not pretending to advise.
The distinction is scope. A bot that does one defined task, never improvises, and hands off to a person at the first sign of a real question is a tool. A bot turned loose to engage visitors and answer whatever they ask is a liability wearing a friendly avatar. If you do add one, the safe version is narrow, scripted, hosted somewhere you have actually checked, and never more than a click from a human.
So before you sign up, put the pitch through a few plain questions. Do we actually have the problem this solves - can we point to measured traffic and genuinely missed enquiries, not a vague fear of them? Where will the chat data live, who can see it, and is it used to train a model? Can the bot be scoped so it never strays into advice? Who maintains and monitors it once the novelty wears off? And over a year, what does it cost against what it realistically returns?
A good vendor can answer those without flinching. If the answers are vague, that is your answer. For most New Zealand financial firms the honest conclusion is the one I started with: the money and attention a chatbot would absorb are better spent making it easy to reach a human and answering the obvious questions in plain sight. The technology is not the problem. The mismatch between what it does and what these firms actually need is.
The web industry is very good at selling financial firms tools built for businesses that look nothing like them. A chatbot is one of the clearer examples: impressive in a demo, mismatched in practice. Spend the subscription on making your site easier to contact and clearer to read, and you will have solved the real problem the bot was only pretending to address.
