Marketing Strategy and Innovation (MSc) at Maastricht University SBE
Learn to deploy real AI agents — on your own VPS, connected to Telegram, running 24/7.
Companion site to the guest lecture by Dr. Clemens Köhler (14 September 2026). This is a private educational website — not an official website of Maastricht University.
Master-level course at Maastricht University School of Business and Economics. Using Problem-Based Learning, you'll develop hands-on skills in innovation management while building your own AI agent.
Understand innovation: drivers, dimensions, and pitfalls. Learn how strategic marketing performs innovation activities and detects market trends.
Capture external ideas from customers, competitors, and policy makers. Coordinate with R&D, operations, HR, and information systems.
Apply state-of-the-art tools: Design Thinking, Growth Hacker Marketing, ideation, prototyping, and scenario development techniques.
Learn how innovations are adopted and diffused in markets. Understand how marketing influences the innovation lifecycle.
The concepts from the lecture, connected to what you build in this course.
Morgan, Whitler, Feng & Chari (2019), Figure 1: strategy formulation vs. implementation × content vs. process.
What strategy is — the decisions, frameworks, and analyses.
How strategy is made — the frontier. Agents that help make the strategy.
ChatGPT in a chat window — the coding session. You ask, it answers.
Your OpenClaw bot — running 24/7, executing tasks across channels. This is where you work now.
Your Tutorial 1 reading. Table 1 of the same paper lists the top challenges of CMOs — your bot touches most of them.
Spanjol et al. (2024), section 3.
OECD/Eurostat (2018), Oslo Manual 2018, 4th ed.: an innovation is new or significantly improved and brought into use.
Where is your bot? Where could your hackathon concept be?
Experimenting: Thomke (2020) — build a culture of experimentation. Verganti, Vendraminelli & Iansiti (2020) — your job shifts from solving the problem to designing the loop that solves it.
Adoption: Arts, Frambach & Bijmolt (2011); Joachim, Spieth & Heidenreich (2018); Aiken & Keller (2009). The hard part is never the technology — it is whether people adopt what you built.
Merriam-Webster Word of the Year 2025 — "slop": "digital content of low quality that is produced usually in quantity by means of artificial intelligence" (source).
BetterUp Labs with the Stanford Social Media Lab surveyed 1,150 full-time U.S. desk workers in September 2025: 40 % say they received workslop in the last month. Reported in Harvard Business Review, 22 September 2025 (primary study).
Your brain is not optional. Let the agent do the work. Keep the thinking.
From zero to a running AI agent in under 10 minutes. No DevOps experience required.
Two options:
Hostinger's managed OpenClaw: a Docker deployment managed through hPanel. You get the same OpenClaw, with less control and less learning.
$5.99/mo (renews $11.99/mo)
Ubuntu 24.04 LTS. Hostinger KVM 2 (2 vCPU, 8 GB RAM, $8.99/mo, renews $14.99/mo) is Hostinger's recommendation for OpenClaw. KVM 1 (4 GB RAM, $6.49/mo) is enough for a chat-only bot.
Any provider works (Hetzner, DigitalOcean, a Mac at home) — the commands are identical.
From $6.49/mo (KVM 1) · $8.99/mo (KVM 2)
Prices retrieved 2026-09-12 from hostinger.com/vps-hosting (renews at higher rates; shortest term is sensible for a course project).
SSH into your server and create a dedicated user. An agent that acts on your accounts should not run as root.
Run the official installer — it detects your OS, installs Node if needed, and installs OpenClaw:
The installer brings its own Node.js (Node 26 recommended by OpenClaw). Check:
This replaced the old setup wizard in OpenClaw 2.0. The onboarding flow walks you through:
Then verify your setup:
Details in the Telegram section below. After you approve the pairing code, your agent answers.
Your AI agent is running 24/7 on your own VPS. Open Telegram, message your bot, approve the pairing code, and start building. Welcome to Marketing Strategy and Innovation.
Once OpenClaw is installed on your VPS, connect it to Telegram in three steps. Your students get a personal AI assistant they can message directly.
Open Telegram, search for @BotFather, and send /newbot. Give it a name and username. BotFather returns your API token.
Treat the token like a password. If you paste it anywhere public, revoke it in BotFather (/revoke) and create a new one.
Back in your VPS terminal, add the Telegram channel with your bot token, then restart the gateway:
Send any message to your bot. By default OpenClaw does not answer strangers — it creates a pairing request with an 8-character code (valid 1 hour). Approve it on the server:
From now on the bot answers you. Everyone else who writes to your bot gets a code and is ignored until you approve them — that is the point.
GLM-5.2 via OpenRouter is the course default. Here's how to activate it and what to watch out for.
OpenRouter slug: z-ai/glm-5.2
OpenClaw model ID: openrouter/z-ai/glm-5.2
Context: 1,048,576 tokens
Price: from $0.48 per 1M input / $1.51 per 1M output (cheapest provider) to $1.40 / $4.40 (standard provider) — OpenRouter routes to different providers at different prices; what you pay depends on the provider that serves your request. (source, retrieved 2026-09-12)
Released: 16 June 2026 by Z.ai
Go to openrouter.ai and sign up. OpenRouter is an API aggregator — one key gives you access to hundreds of models.
Top up a small amount ($5–10 is enough for a semester of chatting; unused credits can expire after one year per OpenRouter's terms). Go to Keys → create key and set a per-key credit limit (an optional spending cap on the key). Copy your key.
Privacy: prompts and completions are not logged by default. Leave the "log prompts for a 1% discount" toggle off and the model-training toggle off.
During onboarding, select OpenRouter as your provider and paste your API key. Then set GLM-5.2:
In chat, switch models:
The -g flag switches the model for everyone; without -g it only changes your current session.
Long context is convenient but you pay for every token in the window on every message. OpenClaw compacts long sessions automatically; /new starts a fresh one.
| Route | What you need | Rough cost | When it makes sense |
|---|---|---|---|
| OpenRouter + GLM-5.2 (course default) | OpenRouter account + key with credit limit | ~$5–10/semester | Course default — cheap, 1M context, multilingual |
| Claude or ChatGPT subscription via OpenClaw subscription routes | Existing subscription (Anthropic "Claude CLI", OpenAI "Codex") | $17–20/mo (subscription) | No separate API budget; check provider's terms for agent use |
| Other open-weight models via OpenRouter | OpenRouter account | See below | Experimenting with different models |
| Local models with Ollama | Your laptop, Ollama installed | Free | Private, offline — but weaker, only small models are realistic |
Models on OpenRouter (prices per 1M input/output, retrieved 2026-09-12 from openrouter.ai/models):
| Model | Slug | Input / 1M | Output / 1M | Context |
|---|---|---|---|---|
| GLM-5.2 (course default) | z-ai/glm-5.2 | $0.60 | $2.00 | 1M |
| GLM-5.3 | z-ai/glm-5.3 | $1.40 | $4.40 | 1M |
| GLM-5.3 Flash | z-ai/glm-5.3-flash | $0.15 | $0.50 | 1M |
| DeepSeek V4 Pro | deepseek/deepseek-v4-pro-0813 | $0.58 | $1.74 | 1M |
| DeepSeek V4.1 Flash | deepseek/deepseek-v4.1-flash | $0.15 | $0.60 | 1M |
| DeepSeek V4 Flash | deepseek/deepseek-v4-flash | $0.07 | $0.13 | 1M |
| Kimi K3 | moonshotai/kimi-k3 | $2.65 | $13.28 | 1M |
| MiniMax M3 | minimax/minimax-m3 | $0.30 | $1.20 | 1M |
| Qwen3 Coder Next | qwen/qwen3-coder-next | $0.12 | $0.80 | 262K |
| GPT-6 Astra | openai/gpt-6-astra | $10.00 | $50.00 | 1M |
| GPT-5.5 | openai/gpt-5.5 | $5.00 | $30.00 | 1M |
GLM, DeepSeek, Kimi, MiniMax, and Qwen are open-weight; GPT-6 Astra and GPT-5.5 are closed-weight. OpenRouter routes to different providers at different prices; what you pay depends on the provider that serves your request.
Scores marked as vendor-reported are noted as such. For independently measured scores, the source is cited.
For this course a cheap model that you supervise beats an expensive one that you don't.
Run these commands before you connect your bot to any real account or channel.
tools.exec.mode: ask means the bot must ask before running commands on your server. The default on a plain host is full with no prompts — wrong for a machine that holds your keys.dmPolicy to open — strangers could talk to your bot without approval.openclaw update monthlyclaw to the sudo group so you can administer the box as that user. If your bot only chats and reports, remove it again with sudo deluser claw sudo — an agent that cannot escalate cannot be talked into escalating.An agent that acts on your accounts acts in your name. Review before it matters.
A live demo of Agentic AI for marketing. Describe any product — watch an AI agent autonomously research the market, define audiences, craft positioning, create multi-channel content, and critique its own work. Only your imagination is the limit.
Interactive web app — runs in your browser. No installation needed. API keys go straight from your browser to the provider you choose — never to this server.
Step-by-step instructions, discussion questions, and bonus challenges.
Not just chat. An AI agent plans and executes multiple steps autonomously — research, analysis, creation, and self-critique.
Three links. That's all it takes to go from idea to running AI agent.
Affordable VPS hosting. Ubuntu 24.04 LTS recommended for OpenClaw. From $6.49/mo.
The AI agent platform that powers your bot. Install it on your VPS, connect Telegram, and you're live.
Full documentation — installation, configuration, channels, agents, skills, and security.