A founder running a professional services consultancy asked me recently whether to invest in an AI chatbot or bring in a part-time virtual assistant. Her enquiries were piling up out of hours, clients were waiting longer than they should, and she was already stretched across delivery and business development. It’s a question worth thinking through carefully, because the cost ranges look very different depending on which direction you go, and the risk of choosing the wrong one runs in both directions.
What’s the choice you’re actually facing?
The question is about matching the tool to what you’re actually trying to handle. AI virtual assistants sit in a price range from roughly £15 to £400 a month depending on features and volume. Human virtual assistants typically cost £800 to £1,600 a month for twenty hours a week. Which one fits depends on your enquiry volume, task complexity, and how clearly your processes are documented.
For many owner-managed businesses, a hybrid works better than either option alone. AI handles volume and repetition while a human VA or existing staff covers exceptions, sensitive conversations, and high-value sales contacts. Understanding both options lets you design that split sensibly rather than committing to one by default.
The cost per AI-handled conversation sits at roughly £0.40, compared with £5 to £10 for a human agent managing a routine query, according to industry benchmarks. At those numbers, the crossover point is lower than many owners expect.
When an AI virtual assistant makes sense
AI virtual assistants earn their keep in high-volume, repetitive work where the answer to a question is predictable. Entry-level tools at £15 to £60 a month handle FAQ deflection, lead capture and out-of-hours responses for owner-managed businesses with fewer than 200 conversations per month. Mid-range options at £60 to £200 a month add CRM integration, multi-channel coverage, and basic analytics once volumes climb higher.
The case is clearest when three things line up: you have a steady flow of similar enquiries, those enquiries follow patterns you can write down, and your business can absorb a small configuration project upfront. Verizon’s 2024 research into small business operations found that owners lose an average of 21.8 hours a week to repetitive administrative tasks. An AI assistant won’t recover all of that, but for the predictable fraction, it handles the volume at a fraction of the cost.
Entry-level tools like Tidio start at around £25 a month and suit businesses handling basic website enquiries or social media messages. HubSpot’s Starter chatbot builder is bundled at roughly £45 a month and connects to its CRM for lead capture and routing. For professional services firms with a steady inbound flow, mid-range platforms like Intercom Fin Lite (from around £60 a month for 500 conversations) offer escalation logic and analytics that a basic widget won’t provide. No-code deployments at this level typically take two to fourteen days to go live, with return on investment achievable within three to six months when the use case is well scoped.
When a human VA is still the better option
A human VA holds the edge wherever tasks require judgement, sensitivity, or handling something genuinely unexpected. Complex client complaints, bespoke proposals, difficult billing conversations, and high-trust client interactions need discretion that no chatbot replicates reliably. If your firm receives only a handful of enquiries per week, or each one is genuinely different from the last, an AI assistant will frustrate more customers than it deflects.
The calculus shifts when your enquiry mix is highly variable, your service offering changes frequently, or you lack the internal capacity to configure and maintain an AI system. Keeping an AI assistant accurate requires someone to update its knowledge base whenever your pricing, services or policies change. If that update process won’t happen consistently, the assistant will start delivering stale or incorrect answers, creating complaints and, in some circumstances, liability.
Human VAs adapt on the fly. Skilled VAs globally cost around £6 to £50 per hour depending on location and experience, with a UK-based part-time arrangement at twenty hours a week running around £1,600 per month. For firms receiving fewer than ten enquiries a week, each involving a degree of judgement, that cost is often well justified. A human VA is also significantly easier to manage if you’re not comfortable with technical configuration.
What it costs to get the call wrong
Getting the decision wrong cuts in both directions. Over-buying means paying £300 a month for a platform when your volume is fewer than fifty monthly enquiries, producing negative ROI. Under-buying means losing leads and insight when a basic chatbot handles hundreds of contacts poorly. A poorly configured AI assistant also introduces compliance costs that owner-managed businesses in regulated sectors must factor in from the outset.
For any business processing customer data through an AI tool, UK GDPR obligations apply. The ICO is explicit that controllers must conduct Data Protection Impact Assessments where AI processing is likely to create high risk for individuals, and its guidance on AI covers transparency, data minimisation and security obligations for chatbot and automated-response deployments. Mapping data flows from your chatbot into your CRM, updating your privacy notice, and setting appropriate retention periods takes time, and sometimes requires professional input.
The National Cyber Security Centre’s guidance on AI system development notes that integrations secured poorly at the API level can expose customer data or be exploited through prompt injection attacks. A misconfigured chatbot with broad CRM access, reached from a public web page, creates incident response and reputational costs that can easily exceed any subscription saving.
Firms regulated by the Financial Conduct Authority face an additional layer. The FCA’s Consumer Duty requires that all communications to retail customers, whether produced by a person or generated by a machine, are fair, clear and not misleading. An AI assistant that implies personalised advice or gives overconfident answers to product questions creates suitability and mis-selling risk. The ICO’s enforcement database documents cases where organisations faced significant consequences for security lapses that AI misconfigurations could equally produce.
What to ask before you commit
Before signing up, push your vendor on four areas: pricing model, data protection, integration depth, and your ability to supervise and correct the system. A well-scoped AI assistant returns its cost within three to six months when aligned to high-volume, predictable work. Getting specific answers before you sign up separates vendors who have genuinely implemented this at your scale from those selling a feature they haven’t configured in practice.
Ask specifically how the pricing model scales when your enquiry volume doubles. Find out what is included in setup and who does the integration work, because Vonage’s own guidance notes that defining conversation flows, training the assistant on your data, and iterative testing all take meaningful time that vendors sometimes treat as your problem, not theirs. Ask where customer data is stored and in which jurisdiction, what retention periods apply, and whether the vendor can provide a data processing agreement and support you through a DPIA if the ICO requires one. Confirm that you can set explicit guardrails on what topics the assistant will and won’t respond to, and that your staff can review transcripts and step in when the assistant is wrong.
For firms in regulated sectors, ask whether the vendor has experience working with FCA-regulated businesses, and how they adapt when regulatory expectations shift. One point many vendors gloss over is who is responsible for keeping the knowledge base current. That responsibility sits with you. If your pricing, policies or services change and the assistant is not updated, the consequences fall on your business.
Define the metrics you will track before you go live. Deflection rate, first-contact resolution, and lead-to-conversion are the figures that tell you whether the tool is earning its subscription or sitting idle.
If you want to work through which option fits your situation and what a sensible first configuration looks like, book a conversation.
