You've read the articles. You've attended the webinars. You know AI is important. But you're still running your business the same way you were running it a year ago. This is the practical, week-by-week playbook we use to transform a traditional SME into an AI-first operation in 90 days. No theory. Just steps.
Before You Start: The Prerequisites
This playbook assumes:
- You run an SME with 2-50 employees
- You have at least one person who's comfortable with technology (doesn't need to be a developer)
- You're willing to invest 2-4 hours per week over 90 days
- You have a budget of £500-2,000 for the transformation (tools, services, infrastructure)
If you don't meet these prerequisites, get there first. This playbook moves fast and assumes basic readiness.
Weeks 1-2: Audit and Map
Week 1: Process Documentation
Spend this week documenting every recurring process in your business. Not the big strategic stuff — the daily, weekly, and monthly operational tasks that consume time without creating proportional value:
- Email management and responses
- Invoice processing and payment tracking
- Report generation
- Data entry between systems
- Meeting scheduling and follow-up
- Customer enquiry handling
- Social media posting
- Inventory/stock management
For each process, document: who does it, how long it takes, how often it happens, and what tools are involved. A shared spreadsheet is fine. Don't overthink the format.
Week 2: Prioritisation
Score each process on two dimensions:
- Time consumption: How many hours per week does this take? (1=low, 5=high)
- AI suitability: How repetitive, rule-based, and data-driven is it? (1=creative/judgmental, 5=highly repetitive)
Multiply the scores. The highest-scoring processes are your automation targets. Pick the top 3. These become your first agent projects.
Weeks 3-4: Foundation
Week 3: Infrastructure Setup
Set up your AI infrastructure:
- Google Cloud account: Create a project. Set up billing with a budget alert at £100/month.
- OpenClaw deployment: Deploy the OpenClaw gateway on Cloud Run. Follow the quickstart guide — it takes about 2 hours including channel configuration.
- Communication channels: Connect 1-2 channels — we recommend starting with Telegram (fastest setup) and email. WhatsApp and Slack can be added later.
Week 4: First Agent
Deploy your first agent — targeting the highest-scoring process from Week 2. For most businesses, this will be one of:
- Email triage agent: Classifies incoming emails, drafts responses for review, archives noise
- Morning briefing agent: Compiles a daily operational summary from your key systems
- FAQ/enquiry agent: Answers common customer questions via your website or messaging channel
Start with strict guardrails. The agent should flag, categorise, and draft — but not send, execute, or decide without human approval. You'll relax these guardrails over time as you build trust.
Weeks 5-8: Expand and Optimise
Weeks 5-6: Second Agent
Your first agent should be running reliably by now. If it's not, spend these weeks debugging and refining before adding more. If it is: deploy your second automation target.
Key lesson from our experience: the second agent is always easier than the first. You've already solved the infrastructure problems. You understand how OpenClaw's configuration works. The time-to-deploy drops from days to hours.
Weeks 7-8: Integration and Refinement
Connect your agents to each other and to your business systems:
- Link the email triage agent to the briefing agent — triaged email summaries feed into the morning report
- Connect the enquiry agent to your CRM — new leads are automatically logged
- Build shared memory — agents access the same Firestore database for consistent context
Also: review the first month's performance data. How many hours has the agent actually saved? Where does it struggle? What tasks is it handling that you didn't originally plan for? Adjust and optimise.
Weeks 9-12: Scale and Systematise
Weeks 9-10: Third Agent + Proactive Automation
Deploy your third automation target. But more importantly: shift from reactive agents (responding to inputs) to proactive agents (initiating actions on schedules):
- Weekly reporting to clients or stakeholders
- Daily inventory/stock checks
- Automated follow-ups on overdue invoices or unanswered proposals
- Content scheduling and distribution
Proactive agents are where the real leverage appears. Instead of replacing human effort on existing tasks, they create new capabilities that didn't exist before — things that would have been nice to do but "we never had time."
Weeks 11-12: Documentation and Measurement
Document everything:
- Agent configurations: What each agent does, its guardrails, its tools, its schedule
- Performance metrics: Hours saved per week, tasks handled per day, error rates, escalation rates
- Cost analysis: Infrastructure costs vs. time savings vs. revenue impact
- Lessons learned: What worked, what didn't, what surprised you
This documentation serves three purposes: onboarding new team members, planning future automations, and measuring ROI for stakeholders.
Day 90: Where You Should Be
By the end of 90 days, a well-executed transformation typically looks like:
| Metric | Before | After 90 Days |
|---|---|---|
| Agent deployments | 0 | 3-5 |
| Hours saved per week | 0 | 10-20 |
| Monthly infrastructure cost | £0 | £25-75 |
| Processes documented | 0 (or informal) | All recurring ops |
| Team AI comfort level | "I've used ChatGPT" | "I manage AI agents daily" |
| Data organisation | Scattered across tools | Centralised in Firestore |
Common Mistakes to Avoid
- Trying to automate everything at once. Start with one agent. Get it working. Then expand. The businesses that try to deploy five agents in week one end up with five broken agents in week three.
- Removing human oversight too early. Keep a human in the loop for the first 30 days of any agent deployment. Review the agent's outputs regularly. Build trust through verification, not faith.
- Ignoring the data foundation. Agents are only as good as the data they access. If your business data is scattered across personal inboxes, random spreadsheets, and sticky notes — fix that first.
- Expecting perfection. Agents make mistakes. Humans make mistakes. The question isn't "does it work perfectly?" — it's "does it work better than the alternative?" An agent that correctly handles 90% of email triage saves more time than a human who correctly handles 98% but takes 10x longer.
- Treating this as an IT project. This is a business transformation. The technology is the easy part. The hard part is changing how your team thinks about work — from "I do this task" to "I oversee the agent that does this task."
Ninety days. That's all it takes to go from "we should probably look into AI" to "our AI agents handle 10-20 hours of work per week autonomously." The playbook works. We've used it ourselves. We've guided clients through it. The only variable is whether you start.
