What's the difference between an AI agent and a chatbot?

A chatbot responds to questions — you ask it something, it answers. An AI agent takes actions — you give it a goal, it plans steps, uses tools, and executes tasks autonomously. A chatbot tells you your account balance. An AI agent processes your refund, updates the CRM, sends a confirmation email, and logs the interaction. Chatbots are reactive. Agents are proactive.

How much does an AI agent cost to build?

A basic AI agent (single task, 2-3 tool integrations) costs £5,000-£10,000. A multi-step agent handling complex workflows across multiple systems costs £10,000-£20,000. Running costs are £200-500/month for hosting and API fees. Chatbots are cheaper: £2,000-£8,000 to build, £100-300/month to run. The cost difference reflects the complexity — agents do more, so they cost more.

Are AI agents safe to use in business?

With proper guardrails, yes. AI agents need human oversight for consequential actions (financial transactions, account changes, customer communications). Build in approval workflows for high-stakes decisions, audit logging for everything the agent does, rollback capability for mistakes, and clear boundaries on what the agent can and can't do autonomously. Under GDPR Article 22 and the EU AI Act, automated decisions affecting individuals require human oversight.

When should I use a chatbot instead of an AI agent?

Use a chatbot when: you mainly need to answer questions (FAQ, product info, support queries), the interaction is conversational and doesn't require actions in other systems, and you want lower cost and complexity. Use an AI agent when: you need the system to take actions (process orders, update records, trigger workflows), the task involves multiple steps across different tools, and you want to automate entire processes, not just conversations.

Do AI agents need GDPR compliance?

Yes, and more than chatbots do. Agents access multiple systems, process data across tools, and take autonomous actions — which means more personal data processing, more data flows to document, and more potential for automated decisions affecting individuals. Every system the agent connects to needs a Data Processing Agreement. The agent's actions need audit logging. And if it makes decisions about people, GDPR Article 22 and EU AI Act human oversight requirements apply.

AI Agents vs Chatbots: What's the Difference and Which Does Your Business Need?

Every vendor in 2026 is calling their product an "AI agent." Microsoft Copilot is an agent. Google's Gemini is an agent. Salesforce has Agentforce. Anthropic has tool-using Claude. If you believe the marketing, agents are everywhere and you need one yesterday.

But here's the thing. Most businesses don't actually need an AI agent. A well-built chatbot would handle their problem for half the price.

And some businesses absolutely need an agent — but they're wasting money on a chatbot that can't do what they need.

This article helps you figure out which camp you're in.

The Core Difference (in Plain English)

A chatbot answers questions. You ask it something, it responds. It's a conversation.

An AI agent does things. You give it a goal, it figures out the steps, uses tools, takes actions across your systems, and reports back when it's done.

That's the whole difference. Everything else — cost, complexity, compliance burden — flows from it.

How This Plays Out in Practice

Chatbot scenario: A customer visits your website at 11pm and types "Do you offer next-day delivery?" The chatbot checks your FAQ database and responds: "Yes, next-day delivery is available for orders placed before 2pm. The cost is £5.99."

Done. Question asked, question answered.

Agent scenario: A customer emails saying they received the wrong item. The agent reads the email, identifies the order number, checks the order management system to confirm what was shipped vs. what was ordered, initiates a return label, creates a replacement order with the correct item, updates the customer's record in the CRM, sends the customer a confirmation email with the return label and replacement tracking number, and flags the warehouse pick error for the ops manager.

Seven steps across four systems. No human involved. The customer gets a resolution in under two minutes instead of waiting for someone to read their email in the morning.

That's the difference between answering questions and doing work.

Chatbot vs Agent: The Full Comparison

	Chatbot	AI Agent
What it does	Answers questions, provides information	Takes actions, executes multi-step tasks
How it works	Question → answer	Goal → plan → tool use → execution → verification
Data access	Reads from 1-2 sources (FAQ, knowledge base)	Reads from and writes to multiple systems
Autonomy	None — waits for user input	High — works independently within boundaries
Typical integrations	Website widget, messaging apps	CRM, email, databases, accounting, calendars, APIs
Build cost	£2,000-£8,000	£5,000-£20,000
Monthly running cost	£100-300	£200-700
Setup time	2-4 weeks	4-10 weeks
Complexity	Low-medium	Medium-high
Compliance burden	Lower (less data processing)	Higher (more systems, more data flows, more autonomous decisions)
Best for	Customer support, FAQ, information delivery	Workflow automation, operations, multi-step processes

When a Chatbot Is Enough

Don't overthink this. A chatbot is the right choice if the problem you're solving is primarily conversational.

Customer Support and FAQ

Your support team answers the same 40 questions every day. Delivery times, pricing, return policies, account setup. A chatbot trained on your knowledge base handles 60-70% of these without any human involvement. The rest get escalated.

Cost to build: £3,000-£5,000 for a well-trained chatbot with your company's knowledge base, personality, and escalation rules.

ROI timeline: 1-3 months. If your team currently spends 15 hours per week on repetitive support queries, a chatbot pays for itself fast.

Product Recommendations

Customer describes what they're looking for. The chatbot asks clarifying questions, then suggests relevant products from your catalogue. Good for e-commerce, especially if you have a large product range.

Internal Knowledge Base

Your team has questions about company policies, procedures, benefits, IT setup. Instead of everyone pinging HR or searching through a 200-page handbook, they ask the chatbot. It finds the answer in seconds.

Lead Capture

Visitor lands on your website. The chatbot engages, asks qualifying questions, captures their details, and passes the lead to your sales team. It's not doing anything complex — just having a structured conversation and collecting information.

The pattern: In all these cases, the interaction is someone asking and the system answering. There's no need to take actions in other systems. No multi-step workflows. No autonomous decision-making.

If that describes your use case, build a chatbot. It's cheaper, faster to deploy, and simpler to maintain. We wrote a full breakdown of what chatbots cost and how to build them right.

When You Need an Agent

You need an AI agent when the work involves doing, not just talking.

Order Processing and Fulfilment

The agent monitors incoming orders, verifies payment, checks stock levels, triggers fulfilment, generates shipping labels, sends customer confirmations, and handles exceptions (out of stock, address issues, fraud flags). Each order touches 3-5 systems. An agent handles the entire flow.

Multi-Step Customer Service

Not just "answer the question" but "fix the problem." Process refunds. Cancel and rebook orders. Update account details across systems. Verify identity, check eligibility, apply the change, confirm with the customer. This is agent territory.

Financial Operations

Invoice processing: receive invoice, extract details, match to purchase order, flag discrepancies, route for approval, update accounting system. Expense management: receive receipts, categorise spend, check policy compliance, route approvals, sync to accounting. Each involves reading documents, making decisions, and updating records.

Employee Onboarding

New hire accepted the offer. The agent creates their accounts, assigns training modules, schedules orientation meetings, sends welcome materials, provisions equipment requests, adds them to the right Slack channels, and follows up on incomplete steps. Twenty tasks across eight systems, executed consistently every time.

Data Consolidation and Reporting

The agent pulls data from your CRM, accounting platform, and project management tool. It compiles a weekly report, identifies trends (revenue up 12%, project completion rate dropping, three clients haven't been contacted in 30+ days), and sends it to the relevant people with action items highlighted.

The pattern: Multiple systems. Multiple steps. Decisions based on data. Actions that change things. If the task requires moving between tools and making judgement calls, that's an agent.

For a deeper look at what agents can do and how to deploy them, see our guide on AI agents for business.

The Cost Difference (and Why)

Chatbot Costs

Component	Cost
Build (knowledge base, training, UI, deployment)	£2,000-£8,000
Monthly running (hosting + AI API)	£100-300
Annual total (first year)	£3,200-£11,600

Agent Costs

Component	Cost
Build (integrations, logic, testing, compliance)	£5,000-£20,000
Monthly running (hosting + AI API + monitoring)	£200-700
Annual total (first year)	£7,400-£28,400

Why Agents Cost More

It's not because the AI is fancier. The language model powering both might be identical. The cost difference comes from three things:

1. Integrations. A chatbot connects to one data source — your FAQ or knowledge base. An agent connects to 3, 5, maybe 8 different systems. Each integration needs building, testing, error handling, and maintaining. When Salesforce updates their API, your agent integration needs updating too.

2. Testing. A chatbot's failure mode is giving a wrong answer. Annoying but fixable. An agent's failure mode is taking a wrong action — sending the wrong refund, creating a duplicate order, emailing the wrong customer. You need extensive testing for every workflow path, every edge case, every possible failure.

3. Compliance. More systems means more data flows. More data flows means more to document. Agents that make decisions about people trigger GDPR Article 22 requirements and, from August 2026, the EU AI Act's transparency and oversight rules. This isn't optional work — it's legally required.

For a complete breakdown of automation costs and ROI, our article on AI workflow automation costs covers the numbers in detail.

The Compliance Difference

This is where a lot of businesses get caught out. They assume compliance is the same for both. It isn't.

Chatbot Compliance (Simpler)

Privacy policy that covers AI-assisted support
Data Processing Agreement with your AI provider (OpenAI, Anthropic, etc.)
EU AI Act transparency: tell users they're talking to AI, not a human
Data retention policy for conversation logs
DPIA if processing personal data at scale

Manageable. A few documents and the right technical setup.

Agent Compliance (Bigger Surface Area)

Everything above, plus:

DPAs with every system the agent connects to. CRM, email platform, accounting tool, order management — each one needs a Data Processing Agreement if personal data flows through it.
Audit logging for every action. What did the agent do? When? What data did it access? What decision did it make? You need a complete trail. Not just for regulators — for when something goes wrong and you need to understand what happened.
Human oversight mechanisms. If the agent makes decisions that affect people (refund approvals, account changes, service level determinations), GDPR Article 22 gives individuals the right to request human review. You need a process for that.
Automated decision-making documentation. The EU AI Act requires transparency about how automated systems make decisions. You need to document the logic, not just the outcome.
Data flow mapping. Data moves between systems. You need to know what goes where, what's processed at each step, and what's stored. More connections means a more complex map.
Rollback capability. When an agent takes a wrong action, you need the ability to undo it. That means building reversibility into every workflow.

This isn't meant to scare you off agents. It's meant to make sure you build them properly from day one. Retrofitting compliance after deployment costs 3-5x more than building it in from the start.

Decision Framework: 4 Questions to Ask

If you're unsure which you need, work through these:

1. Does the task require actions in other systems?

If your use case is purely conversational — answering questions, providing information, capturing form data — a chatbot is sufficient. If it involves updating records, triggering workflows, or making changes across platforms, you need an agent.

2. How many systems are involved?

One or two (website + knowledge base)? Chatbot. Three or more (CRM + email + accounting + order system)? Agent. The number of systems is the clearest indicator of complexity.

3. Does it need to make decisions?

If the system just retrieves and presents information, chatbot. If it needs to evaluate conditions, choose between options, and act differently based on context (approve/reject, escalate/handle, flag/ignore), that's agent territory.

4. What's the cost of getting it wrong?

If a chatbot gives a wrong answer, a customer might be briefly annoyed. If an agent takes a wrong action — processes an incorrect refund, sends confidential data to the wrong person, double-charges an account — the consequences are real and potentially expensive. Higher stakes means you need the extra testing and compliance work that comes with a proper agent build.

Quick scoring: If you answered "no, 1-2, no, low" — build a chatbot. If you answered "yes, 3+, yes, high" — build an agent. Mixed answers? Start with a chatbot for the conversational piece and add agent capabilities for specific workflows later.

The Hybrid Approach (What Most Businesses Actually Need)

Here's what we see most often: businesses don't need a pure chatbot or a pure agent. They need a chatbot that handles the front-end conversation, connected to agent-like workflows for specific tasks.

Example: A property management company.

Chatbot layer: Tenant asks about rent due dates, maintenance request procedures, lease terms. The chatbot handles these conversations from a knowledge base. Cost: £3,000.
Agent layer: Tenant reports a leak. The agent creates a maintenance ticket, notifies the assigned contractor, schedules the repair, sends the tenant a confirmation with the time window, and follows up 48 hours later to confirm the repair was completed. Cost: £5,000-£8,000 for this workflow.

Total: £8,000-£11,000. The chatbot handles 70% of interactions. The agent handles the 30% that require action. You get cost efficiency where volume is high and automation where it matters most.

The Bottom Line

The question isn't "should I get an AI agent or a chatbot?" It's "what does the work actually require?"

If the work is answering questions, a chatbot is cheaper, simpler, and faster to deploy. Don't overcomplicate it.

If the work involves taking actions across multiple systems, making decisions, and executing multi-step processes, you need an agent. Don't underbuild it.

Either way, build compliance in from the start. The regulatory environment in 2026 — GDPR enforcement, EU AI Act obligations from August — doesn't leave room for "we'll sort that out later."

Need help figuring out which approach fits your business? Talk to us. We'll assess your use case, recommend the right solution, and build it with compliance baked in — not bolted on. You can see our full range of AI development and compliance services.

Related reading: