Designing AI chatbots that feel like a strategist, not a chatbot

Most AI tools still feel like spreadsheets with a chat box bolted on. We wanted Katelyn to feel like a strategist — someone who knows your data, understands your goals, and always guides you to the next best move.

After 18 months of development, countless iterations, and some humbling failures, we've learnt what it takes to create conversational AI that B2B leaders actually trust. This isn't about building another chatbot. It's about engineering a digital strategist that transforms how marketing teams work.

The problem with most chatbot UI designs

Let's be honest: most chatbots are terrible. They're reactive, disconnected, and often patronising. They dump data without context, ask questions they should already know the answers to, and feel more like obstacles than assistants.

The uncanny valley of AI interactions is real. When a chatbot tries too hard to be human, it fails. When it's too robotic, it's useless. Most chatbot user interface designs get stuck in this middle ground — neither genuinely helpful nor authentically conversational.

What B2B leaders actually need isn't another dashboard disguised as chat. They need strategic thinking, contextual awareness, and actionable guidance. They need what we call a digital consigliere, not a digital assistant.

From chatbot to strategist: The mindset shift

When we started building Katelyn, we didn't ask "How do we build the best AI chatbot?" We asked "What would the world's best marketing strategist do?"

This fundamental shift changed everything about our conversational design approach:

From tool to teammate: Traditional chatbots are tools you use. Strategic AI becomes a teammate you work with. The difference shows in every interaction — from how it remembers context to how it suggests next steps.

From reactive to proactive: Most chatbots for marketing wait for questions. A strategist anticipates needs. Katelyn doesn't just answer "What's my ROI?" She notices when ROI is dropping and alerts you with solutions.

From data dumper to insight curator: Any chatbot can pull numbers from a database. A strategist knows which numbers matter right now and why. It's the difference between "Here are 50 metrics" and "These 3 pages could add £43,000/mo - a 12% revenue increase."

From chatbot to strategist: This isn't about adding personality to a bot. It's about embedding strategic thinking into every response, every suggestion, every interaction.

Emotional design: What users should (and shouldn't) feel

Great conversational UX isn't just about functionality — it's about emotion. We mapped out exactly how users should feel when working with strategic AI:

What users should feel

Relief: "Finally, someone's keeping track of this chaos for me." We achieve this through comprehensive monitoring and intelligent summarisation. Instead of forcing users to check multiple dashboards, Katelyn proactively surfaces what matters.

Confidence: "I trust these recommendations — they're grounded in ROI, not guesswork." Every suggestion includes clear reasoning and expected outcomes. We show our working, cite specific data, and always frame recommendations in business impact.

Momentum: "I know the next best move, and I can act now." Strategic AI should create forward motion, not analysis paralysis. Each conversation ends with a clear next step, sized appropriately to the user's capacity.

Partnership: "Katelyn's got my back — she anticipates what I need." Through consistent interaction patterns and learned preferences, users feel supported rather than supervised.

What users should NOT feel

Overwhelmed: No walls of text, no data dumps, no 47-point action plans. Information is revealed progressively, always respecting cognitive load.

Patronised: No "Hi there! I'm your friendly AI assistant!" nonsense. Professional tone, peer-level interaction, respect for expertise.

Disconnected: No jarring jumps between chat and external dashboards. Everything flows within the conversation, maintaining context and focus.

Distracted: No unnecessary animations, notifications, or "helpful" interruptions. The interface respects attention as a precious resource.

The chat + cards model: Why pure chat fails

After testing dozens of chatbot UI approaches, we discovered pure chat has fundamental limitations:

• Information density problems: Complex B2B data doesn't fit in chat bubbles
• No visual hierarchy: Everything looks equally important
• Lost context: Critical information scrolls away
• No actionable elements: Can't click, expand, or interact with pure text

Our solution: the chat + cards model. We first pioneered this conversational design pattern working with Microsoft back in 2019, when we built one of the first AI-powered content intelligence systems for a global agency. The system helped 5,000 employees find content across 1.9 petabytes of data - all through natural language in Microsoft Teams.

"Best use of Microsoft Teams I've seen," was the feedback - and this was before ChatGPT existed. We used Adaptive Cards and message extensions to create what's now become our foundation for all strategic AI interfaces. The pattern was so effective that Microsoft featured it as an exemplar of conversational AI design.

This conversational design pattern combines the best of both worlds:

Progressive disclosure in action

Chat maintains the conversational flow while cards provide evidence and actions. Here's how it works:

Level 1 - Summary: Start lightweight with key metrics

Katelyn: "I've found £43,000/mo uplift across 3 pages - a 12% revenue increase."
[Collapsed ROI card: "3 pages, 6h effort, 162x ROI"]

Level 2 - Details: User expands for more context

[Expanded card shows:
- Individual page breakdowns
- Lens scores for each page  
- Specific improvement recommendations]

Level 3 - Deep dive: Full modal for implementation

[Modal includes:
- Step-by-step implementation guide
- Technical requirements
- Expected timeline
- Success metrics]

This progressive approach respects user time while providing depth when needed. It's the difference between overwhelming and empowering.

Conversational flow design: The psychology matters

Designing great chatbot conversations isn't just about the words — it's about timing, context, and flow. Here's what we've learnt about conversational UX psychology:

Response timing creates trust

Instant responses feel inhuman. Delays build anticipation. We've calibrated response times based on query complexity:

• Simple queries: 1-2 seconds (feels responsive)
• Analysis requests: 3-5 seconds with "thinking" indicator
• Complex operations: Progressive updates ("Analysing page 1 of 50...")

Context maintenance across turns

Nothing breaks trust faster than a chatbot that forgets what you just discussed. Our architecture maintains conversation state, user preferences, and historical context. When you return a week later, Katelyn remembers where you left off.

A complete conversation flow

Let's look at great chatbot examples in action. Here's how to design chatbot conversation that builds trust and drives action:

User: "How's my site doing?"

Katelyn first checks data freshness: "I have analysis from 4 days ago showing strong performance. Would you like me to refresh with the latest data first?"

User: "Use what you have for now."

Katelyn provides layered insights: "Your site's performing well overall — we've identified £43,000/mo in potential uplift - a 12% revenue increase. The biggest opportunity is on three key pages that are underperforming for your ICP."

[Shows collapsed card with summary metrics]

"Want to see the specific pages and quick wins?"

User: "Yes, show me."

Katelyn expands the details: [Card expands showing page-by-page breakdown]

"Your About page has the highest potential — £14k/month with about 2 hours of work. Shall I create a sprint brief for these improvements?"

User: "Create the brief."

Katelyn confirms before executing: "I'll create a brief covering all three pages with:

• Specific copy improvements
• Technical requirements
• Implementation timeline
• Success metrics

This will be saved to your sprint planning folder. Proceed?"

Each step builds trust through transparency, maintains user control, and manages cognitive load effectively.

Guardrails and limitations: The boundaries that build trust

The best AI chatbot knows its limitations. Strategic AI that tries to do everything inevitably does nothing well. Here's how we've built intelligent boundaries into our conversational design:

What strategic AI can do

• Access and analyse data: Pull from multiple sources, identify patterns, surface insights
• Identify opportunities: Spot ROI potential, funnel leaks, competitive gaps
• Suggest strategic moves: Recommend prioritised actions based on impact and effort
• Execute with permission: Create briefs, update plans, trigger workflows — always with confirmation

What it cannot and should not do

• Make decisions without consent: Every action requires explicit approval
• Access unauthorised data: Clear boundaries on data access, always transparent
• Promise unrealistic outcomes: Honest about limitations and uncertainties
• Replace human judgment: Augments decision-making, never replaces it

Intelligent escalation patterns

Knowing when to involve humans is crucial for chatbots for marketing. Here's how we handle escalation:

Technical boundaries:

Katelyn: "This integration requires API access I don't have. 
Would you like me to create a technical brief for your development team?"

Strategic boundaries:

Katelyn: "This strategy change could impact your sales process significantly. 
I recommend reviewing with your sales leader first. 
Shall I draft a brief outlining the implications?"

Confidence boundaries:

Katelyn: "I'm seeing unusual patterns in this data that I'm not confident interpreting. 
Would you like me to flag this for expert review?"

These boundaries don't diminish the AI's value — they enhance trust by showing self-awareness and respecting expertise.

Technical sophistication without the jargon

Building conversational AI that feels natural requires serious technical architecture. Here's how we've solved key challenges while keeping the user experience simple:

Separating questions from actions

We use what's called CQRS (Command Query Responsibility Segregation) — but users just experience it as "answers are instant, actions are thoughtful."

When you ask "What are my top ROI pages?", that's a query. The answer comes from pre-calculated data, delivered in milliseconds. When you say "Run a full audit", that's a command. It triggers deeper analysis, with progress updates along the way.

This separation means the chatbot UI always feels responsive, even when handling complex operations.

Async operations that feel human

Heavy analysis doesn't block conversation. While Katelyn analyses your 50-page site, you can continue asking questions, refining parameters, or working on other tasks.

We humanise the waiting with natural progress updates:

• "Starting analysis..."
• "Found some interesting patterns..."
• "Nearly done — calculating projections..."

State management across sessions

Every conversation builds on the last. Our architecture maintains:

• User preferences and context
• Historical decisions and outcomes
• Pinned insights and active projects
• Team-wide knowledge sharing

This creates continuity that makes the AI feel less like a tool and more like a team member who's always up to speed.

Lessons from 18 months of building strategic AI

Building the best AI chatbot isn't a destination — it's an evolution. Here's what we got wrong initially and how we fixed it:

Over-automation killed trust

Initial approach: Katelyn would automatically implement high-confidence improvements. User reaction: "Wait, what did you just change?" Fix: Always preview, always confirm, always provide undo options.

Under-explaining created confusion

Initial approach: Brief, efficient responses to save time. User reaction: "How did you calculate that?" Fix: Progressive disclosure — summary first, then reasoning available on demand.

Perfect grammar felt robotic

Initial approach: Flawless, formal communication. User reaction: "This doesn't feel natural." Fix: Conversational tone with minor variations, matching user formality level.

Instant everything reduced value perception

Initial approach: All operations completed instantly. User reaction: "That can't be thorough analysis." Fix: Appropriate processing times that signal thoughtful work.

These lessons shaped our core insight: great conversational design isn't about perfection — it's about feeling authentically helpful.

Implementing these principles in your own AI

Whether you're building chatbots for marketing teams or customer service, understanding how to design chatbot conversation effectively requires both strategic thinking and technical sophistication.

The interface layer: Where personalisation happens

The secret to conversational AI that feels truly strategic lies in what we call the interface layer. This is where LLMs personalise every response based on:

• Conversation history: Every previous interaction informs the current response
• Implicit preferences: Learning from user choices without explicit configuration
• Context awareness: Understanding not just what users ask, but why they're asking
• Goal anticipation: Focusing on what the user is trying to achieve, not just answering questions

For example, when a CMO consistently asks about ROI first, the interface layer learns to lead with financial impact. When they repeatedly dive into technical details, it adjusts to provide more depth upfront.

This isn't about simple personalisation — it's about building a mental model of each user's priorities, communication style, and decision-making patterns.

Start with emotional goals, not features

Before writing a single line of code, map out how users should feel. What emotions indicate success? What feelings signal failure? Design every interaction to move users toward positive emotional states.

Design conversations before code

Write out full conversation flows. Test them with real humans reading scripts. Identify confusion points, refine language, and validate the flow before any development begins.

Test with real users early

Your team understands your chatbot too well. Real users will break it in ways you never imagined. Test early, test often, and test with people who match your actual audience.

Iterate based on confusion points

Track where conversations fail. When users abandon sessions, ask why. When they repeat questions, understand what's unclear. Each confusion point is an opportunity to improve.

Quick wins for better chatbot UX

• Add "working on it" states with contextual messages
• Show confidence levels for recommendations
• Always explain reasoning (available on demand)
• Offer alternatives when confidence is low
• Build in graceful failure paths

The future of strategic AI in B2B

Our journey with conversational AI design began back in 2019 with Microsoft, where we built one of the first AI-powered content intelligence systems. Since then, we've spent years perfecting these conversational design principles, culminating in Katelyn — our most sophisticated implementation yet.

What we've learnt applies whether you're building AI for your team or your customers: respect user control, build trust through transparency, and always focus on outcomes over outputs.

The best AI chatbot isn't the one with the most features or the cleverest responses. It's the one that makes users feel supported, confident, and capable of achieving more than they could alone.

Strategic AI isn't about replacing human intelligence — it's about augmenting it in ways that feel natural, helpful, and trustworthy. And as we continue to push the boundaries of what's possible with AI-powered solutions and strategic advisory, we're seeing that the future belongs to those who can blend technical sophistication with human understanding.

Ready to explore what strategic AI could do for your business?

Whether you need AI that transforms your marketing team's capabilities or want to provide intelligent experiences for your own customers, the principles remain the same. But execution makes all the difference.

Talk to our experts about building AI that thinks like a strategist →

---

About the author: Our team has been designing conversational AI since 2019, starting with Microsoft's AI content intelligence platform. This article shares real lessons from developing multiple AI systems, including Katelyn — our strategic AI advisor that helps marketing teams operate like departments 10x their size.