Every agency has been asked about chatbots. Clients hear the buzzword, see them on competitor websites, and assume they need one. The question is not whether you can build one. It is whether you should.
The honest answer: chatbots are brilliant in some contexts and a waste of money in others. Knowing the difference is what separates agencies that build profitable chatbot services from ones that deliver expensive disappointments.
Lead capture chatbots: usually worth it
This is the strongest use case and the one most agencies should start with. A lead capture chatbot sits on a client’s website and qualifies visitors by asking a short sequence of questions. Name, email, what they need, budget range, timeline.
Why it works: It engages visitors who would not fill in a static form. The conversational format feels lower commitment, so more people start the process. Completion rates on well-designed lead capture bots run 15-25% higher than traditional forms.
Implementation time: 2-4 days for a straightforward build. This includes designing the conversation flow, connecting it to the CRM, and setting up notifications.
Ongoing maintenance: Minimal. Maybe 1-2 hours per month reviewing conversation logs, tweaking questions, and updating routing rules.
Tools that work well: Tidio, Drift, and Intercom all handle lead capture effectively. For agencies that want more control, building on Voiceflow or Botpress gives you a white-label option you can resell across clients. This is a natural candidate for a productised service.
Typical pricing: Agencies charge £1,500-4,000 for setup and £200-500/month for management. Margins are strong because the maintenance is light.
Customer service chatbots: often oversold
This is where agencies get into trouble. A client wants to automate their customer service with a chatbot, and the agency says yes without fully understanding the complexity.
Customer service chatbots need to handle dozens (sometimes hundreds) of different query types, access order data, process returns, check stock levels, and escalate gracefully when they cannot help. That is not a two-day build.
The problem: Clients expect the chatbot to handle 80% of enquiries. In practice, poorly implemented customer service bots handle 30-40% and frustrate customers on the rest. The result: unhappy customers, a frustrated client, and an agency dealing with scope creep.
When it works: Clients with a clearly defined and limited set of common queries (under 20 distinct query types), an existing knowledge base, and realistic expectations. E-commerce brands with standard questions about shipping, returns, and sizing are a good fit.
When to skip it: Clients with complex, nuanced customer interactions. Professional services firms, B2B companies with long sales cycles, or any business where the customer service interaction is itself a selling opportunity. In these cases, a chatbot gets in the way.
Implementation time: 3-6 weeks for a proper build. Longer if integration with backend systems is required.
Ongoing maintenance: 4-8 hours per month minimum. Conversation logs need regular review, the knowledge base needs updating, and edge cases need handling. This is where agencies underestimate the work and margins erode.
Internal knowledge bots: underrated
This is the use case most agencies overlook, and it is arguably the best one.
An internal knowledge bot sits inside a company’s Slack, Teams, or intranet and answers employee questions by searching internal documentation. New starters asking about processes, developers checking API documentation, sales teams looking up pricing structures.
Why it is underrated: The ROI is immediate and measurable. If 50 employees each save 15 minutes a day by asking a bot instead of searching through Google Drive or bothering a colleague, that is over 3,000 hours saved per year. At even a modest average hourly cost of £25, that is £75,000 in recovered productivity.
Implementation: These bots work by indexing a company’s internal documents (Google Drive, Notion, Confluence, SharePoint) and using retrieval-augmented generation (RAG) to answer questions grounded in that content. Be sure to review data privacy considerations before connecting client document stores.
Tools: Stack AI, CustomGPT, and Glean all offer no-code or low-code options. For agencies with developers, building on LangChain or LlamaIndex gives more flexibility.
Implementation time: 1-2 weeks for a basic deployment. 3-4 weeks if the client’s documentation needs organising first (it usually does).
Ongoing maintenance: 2-3 hours per month. Mostly re-indexing when documentation changes and reviewing question logs to identify gaps.
Website chatbots vs messaging platform bots
Agencies often default to website chatbots, but the platform matters.
Website chatbots
Best for lead capture and basic customer service. The visitor is already on the site, exploring. A well-timed chatbot that offers help after 30 seconds on a pricing page can convert browsers into leads.
Limitation: The conversation dies when they leave the page. There is no persistent relationship unless you capture their contact details early in the flow.
WhatsApp and messaging platform bots
Better for ongoing customer relationships. The conversation persists. The customer can come back days later and pick up where they left off. Notifications work naturally.
For agencies serving retail, hospitality, and service businesses, WhatsApp bots are often more effective than website chatbots. Open rates on WhatsApp messages sit at 90%+ compared to 20-30% for email.
Consideration: WhatsApp Business API has costs attached (per-conversation pricing) and requires approval. Factor this into the client proposal.
Slack and Teams bots
The right platform for internal knowledge bots. Employees already live in these tools, so adoption is natural. No need to train people to go to a new interface.
When a contact form beats a chatbot
Sometimes the simple answer is the right one. A contact form beats a chatbot when:
- The client’s website gets fewer than 500 visits per month. There is not enough traffic to justify the investment. A clear, well-designed form with a fast response process will outperform a chatbot at this scale.
- The buying decision is complex. If the visitor needs a 30-minute consultation to understand the product, a chatbot cannot bridge that gap. Better to route them to a booking link.
- The client cannot maintain it. A chatbot with outdated information is worse than no chatbot at all. If the client will not commit to maintenance (or pay you to do it), skip it.
Building a chatbot service line
If you are adding chatbots to your agency’s offering, structure it properly. This falls under the broader question of building an AI services division inside your agency.
Packages that work:
- Lead capture bot: £2,000-3,500 setup, £300/month management. High margin, low maintenance. Your bread and butter.
- Knowledge bot: £4,000-8,000 setup, £500/month management. Sell to mid-size companies with 50+ employees. Strong ROI story.
- Customer service bot: £8,000-15,000 setup, £800-1,500/month management. Only take this on with clear scope boundaries and realistic client expectations.
The rule: Start with lead capture. Prove the value. Expand from there. Do not lead with customer service bots unless you have the technical depth to deliver them well.
This is part of Tool Drop, a series reviewing AI tools and approaches through an agency lens. Subscribe to the newsletter to get new articles weekly.
