AI Foundations
Module 06
Foundation
Pitching ShopMate Internally: Maya needs to convince her business partner to invest two weeks of developer time in ShopMate. She scores each feature by time saved vs effort to build. The top three -- product descriptions, customer email replies, and review summaries -- take 5 days to build and save 15 hours per week. The business case is straightforward.

Enterprise AI Strategy

Deploying AI at scale requires more than selecting a model. It requires executive alignment, governance, change management, and a phased rollout strategy. These fundamentals apply regardless of which AI platform or tools your organisation deploys.

Build vs Buy: The First Strategic Decision

Before writing a single line of code, every organisation faces this question: should we build our own AI capabilities, buy an off-the-shelf solution, or combine the two?

Build (API + Custom Development)

What: Use model provider APIs (Claude, GPT, etc.) to build custom applications tailored to your workflows.

Pros: Full control over the user experience, deep integration with internal systems, competitive differentiation, ownership of IP.

Cons: Requires engineering talent, longer time to deploy, ongoing maintenance burden, responsible for safety and reliability.

Best for: Core business processes where AI is a differentiator; organisations with engineering capacity; use cases requiring deep system integration.

Buy (SaaS AI Products)

What: Adopt pre-built AI products (e.g., AI customer service platforms, AI writing tools, AI code assistants).

Pros: Fast deployment, vendor handles maintenance and safety, built-in best practices, lower initial cost.

Cons: Limited customisation, vendor lock-in, data leaves your control, generic rather than tailored to your workflows.

Best for: Non-differentiating use cases (e.g., meeting transcription, general writing assistance); organisations without engineering capacity; rapid pilots.

i
The Hybrid Approach

Most organisations end up with a hybrid: buy commodity AI capabilities (meeting transcription, general writing assistance, code completion) and build custom solutions for core business processes where AI creates competitive advantage. ThreadCo uses off-the-shelf code completion tools for development but builds a custom ShopMate agent because customer service is their differentiator.

Executive Alignment and Sponsorship

Executive Sponsor

Every successful enterprise AI programme has a named executive sponsor with the authority to allocate budget, resolve cross-functional blockers, and communicate the AI vision to the organisation. Without this, programmes stall at pilot stage. The sponsor does not need to be a technologist -- they need to be someone who can make decisions, remove obstacles, and keep the programme visible at the leadership level.

The AI Vision

Before selecting tools or building prototypes, articulate a clear AI vision: What business outcomes will AI drive? Which strategic priorities does it support? What does success look like in 6 months, 12 months, and 3 years? A vision without metrics is a wish. Define measurable objectives tied to business KPIs, not vanity metrics like "number of prompts sent."

Cross-Functional Alignment

AI adoption touches every function: engineering builds it, legal reviews it, security secures it, HR manages the workforce impact, finance funds it, and business units use it. Establish a cross-functional steering committee that meets regularly (bi-weekly or monthly) to align priorities, resolve conflicts, and share learnings across the organisation.

Communication Strategy

How you communicate about AI internally determines adoption. Be transparent about what AI can and cannot do. Share concrete examples and metrics. Address fears directly -- job displacement anxiety is real and ignoring it breeds resistance. Frame AI as a tool that makes existing roles more effective, not a replacement. Celebrate early wins publicly to build momentum.

AI Governance

Governance is the framework of policies, processes, and accountability structures that ensure AI is deployed safely and effectively. Establish it before broad deployment -- retrofitting governance after incidents is far more expensive and disruptive.

AI Centre of Excellence

A dedicated team (even if small: 2-3 people to start) that owns AI governance, maintains best practices, evaluates new tools, and supports business units in their AI initiatives. The CoE is not a gatekeeper -- it is an enabler. It provides templates, training, review processes, and approved tools that make it easy for teams to adopt AI safely. Without a CoE, every team reinvents the wheel and makes avoidable mistakes.

Acceptable Use Policy

A clear, written policy that defines: which AI tools are approved, what data can be sent to AI models, prohibited use cases (e.g., automated hiring decisions without human review), requirements for human oversight, and consequences for policy violations. Keep it concise -- a 50-page policy that nobody reads is worse than a 2-page policy that everyone follows. Update it quarterly as the technology evolves.

Data Classification

Not all data can be processed by AI equally. Establish clear tiers: Public (can be sent to any AI tool), Internal (can be sent to approved enterprise AI tools with DPAs), Confidential (can only be processed by self-hosted models or with explicit approval), Restricted (never sent to AI -- PII, health records, classified information). Map every use case to a data tier before deployment.

Monitoring and Compliance

Governance without monitoring is policy without enforcement. Implement: usage tracking (who is using which tools, how much), cost monitoring (AI spend by team, use case, and model), quality monitoring (output accuracy, user satisfaction), compliance monitoring (data policy adherence, approval workflow completion). Dashboard these metrics and review them monthly with the steering committee.

Measuring ROI

AI investments must demonstrate measurable business value. Executives will not continue funding programmes that cannot show returns. Here is how to build a credible ROI case.

ROI CategoryMetricsHow to MeasureThreadCo Example
Time savingsHours saved per week/monthTime-motion study: measure task duration before and after AIEmail replies: 2 min manual vs 15 sec with ShopMate = 15 hrs/week saved
Cost reductionLabour cost saved minus AI cost(Hours saved x hourly rate) - AI platform costs(15 hrs x $25/hr) - $50/month = $1,575/month net savings
Quality improvementError rate, consistency, customer satisfactionBefore/after comparison of quality metricsProduct description consistency: 60% -> 95% brand-voice compliance
Revenue impactNew revenue enabled or conversion rate improvementA/B testing AI-generated vs manual contentAI product descriptions: 12% higher conversion rate in A/B test
Scale enablementTasks that were impossible before AICount tasks that could not be done at all without AIPersonalised email campaigns for 2,000 products -- previously infeasible
!
ROI Pitfall: Measuring the Wrong Things

Common mistake: measuring adoption (number of users, prompts sent) instead of outcomes (time saved, revenue impact, quality improvement). High adoption of a tool that does not improve outcomes is waste, not success. Always tie your metrics to business outcomes that executives care about.

Change Management

Technology adoption fails when it ignores the human element. AI adoption is especially sensitive because it triggers legitimate concerns about job security, skill relevance, and professional identity.

Address Fear Directly

Do not pretend that AI will not change roles -- it will. Be honest about how roles will evolve. Frame the change as augmentation: "AI handles the repetitive parts so you can focus on the work that requires your judgment, creativity, and relationships." Provide concrete examples of how each role changes. People fear the unknown far more than they fear known changes.

Training and Upskilling

Invest in training before deploying AI tools. This programme you are completing right now is an example. People adopt tools they understand and resist tools they do not. Training should cover: what the tool can do, what it cannot do, how to use it effectively, when not to use it, and how to verify its outputs. Hands-on practice is essential -- lectures alone do not build competence.

Champions Network

Identify 5-10% of your workforce as "AI champions" -- enthusiastic early adopters who can support their peers. These are not IT staff; they are business users who understand their team's workflows and can translate AI capabilities into practical benefits. Invest extra training time in your champions. They will do more for adoption than any top-down mandate.

Feedback Loops

Create structured channels for users to report problems, request features, and share successes. Weekly "AI office hours" where users can bring real problems. A shared Slack/Teams channel for tips and questions. Monthly "show and tell" sessions where teams demonstrate creative AI uses. These loops accelerate learning and surface issues before they become problems.

Running Effective Pilot Programs

Pilots are how you prove value, identify problems, and build confidence before scaling. A well-designed pilot produces evidence that drives the scale decision.

Pilot ElementDescriptionExample
ScopeOne specific use case, one team, defined time periodShopMate email replies for the customer service team, 6 weeks
Success criteriaMeasurable outcomes defined before the pilot startsReduce average reply time from 12 min to 3 min; maintain 90% CSAT
Baseline measurementMeasure current performance before the pilotCurrent: 12 min/reply, 87% CSAT, 40 hrs/week on email
Participants20-50 users, mix of enthusiasts and scepticsFull customer service team (3 people) plus 2 from sales
Support structureTraining, documentation, escalation path, weekly check-ins2-hour training session, Slack channel, weekly 30-min review meeting
Exit criteriaClear decision framework: scale, iterate, or stopScale if: reply time < 5 min AND CSAT > 85%. Iterate if one met. Stop if neither.
i
Pilot Anti-Patterns

Too many use cases: Trying to prove everything at once proves nothing. Pick one. No baseline: You cannot show improvement if you did not measure the starting point. Wrong participants: A pilot with only enthusiasts will overstate benefits; only sceptics will understate them. No exit criteria: Without predefined success criteria, the pilot becomes an endless experiment that never leads to a decision.

Phased Rollout Strategy

Phase 1: Foundation (Weeks 1-6)

Activities: Executive alignment, governance framework, acceptable use policy, vendor selection, security review, training programme launch. Deliverables: Approved tool list, data classification policy, training materials, pilot plan. Exit criteria: Policy approved by legal and security, training content ready, pilot team identified.

Phase 2: Controlled Pilot (Weeks 7-14)

Activities: Deploy AI tools to 20-50 users for specific use cases, intensive support and monitoring, weekly feedback sessions, measure against baseline. Deliverables: Pilot results report, refined use cases, updated training based on real-world learnings. Exit criteria: Success criteria met for at least one use case, no critical safety or compliance issues.

Phase 3: Departmental Expansion (Months 4-6)

Activities: Expand to 2-3 departments, add 2-3 new use cases based on pilot learnings, establish champions network, refine governance processes. Deliverables: Department-level adoption metrics, ROI evidence, updated best practices. Exit criteria: Demonstrated ROI in multiple departments, governance processes operating smoothly, champions network active.

Phase 4: Enterprise Scale (Month 7+)

Activities: Organisation-wide rollout, full training programme, AI Centre of Excellence fully operational, advanced use cases (agents, custom apps), continuous improvement cycle. Deliverables: Enterprise adoption dashboard, quarterly business reviews, ongoing training curriculum. Exit criteria: This phase is ongoing -- focus shifts to optimisation, new use cases, and keeping pace with technology evolution.

Building a Use-Case Portfolio

Prioritise use cases on two axes: business impact and implementation feasibility. Start with high-impact, high-feasibility quick wins to build organisational confidence.

High FeasibilityLow Feasibility
High ImpactDO FIRST: Quick wins that prove value fast. E.g., AI-assisted customer email replies, automated report generation, code review assistance.PLAN FOR: Strategic investments that need infrastructure or data work. E.g., custom AI agents, predictive analytics, personalisation engines.
Low ImpactNICE TO HAVE: Low-effort experiments that build skills. E.g., meeting transcription, internal FAQ bot, document summarisation.AVOID: High effort, low return. E.g., fully autonomous systems for niche tasks, AI projects without clear business sponsors.

Scaling AI Adoption

Moving from pilot to enterprise-wide adoption requires addressing challenges that do not exist at small scale.

Infrastructure

At scale, you need: centralised API key management, cost allocation by team/project, rate limiting and usage quotas, single sign-on integration, audit logging, prompt template management. Most organisations underestimate the infrastructure required to run AI at scale. Build it incrementally as you grow, but plan the architecture upfront.

Cost Management

AI costs can grow rapidly with adoption. Implement: model tiering (use the cheapest model that meets quality requirements for each task), prompt optimisation (shorter prompts = lower cost), caching (identical requests should not call the API twice), batching (process bulk work during off-peak hours for lower rates), and budget alerts per team and project.

Quality Assurance

At scale, you cannot manually review every AI output. Build automated quality systems: output validation against expected formats, regression testing for prompt templates, A/B testing for prompt improvements, user feedback collection, periodic human audits of random samples. Treat AI quality like software quality -- it requires continuous testing and monitoring.

Knowledge Sharing

The biggest scaling challenge is not technology -- it is knowledge. Create: a central prompt library (best prompts and templates), a use-case gallery (successful deployments), an internal blog or newsletter, regular training sessions, and a community of practice. Organisations that share AI knowledge effectively adopt AI 3-5x faster than those that leave each team to figure it out alone.

Governance at Scale

Governance ElementPilot StageEnterprise Scale
PolicyInformal guidelinesFormal acceptable use policy, reviewed quarterly
ApprovalAd hoc review by pilot leadStructured approval workflow with SLAs
MonitoringManual spot checksAutomated dashboards with real-time alerts
TrainingHands-on workshop for pilot groupMandatory onboarding module + role-specific tracks
Incident response"Ping the pilot lead"Defined escalation path, SLAs, post-incident review process
Cost controlSingle budget lineChargebacks by department, per-team quotas, automated alerts
i
Next Step: Choose Your Tool Track

You have completed the AI Foundations track. Now choose a tool track based on your role and the tools your organisation has adopted. AI assistants are ideal for knowledge workers, analysts, and enterprise deployments. Windsurf is purpose-built for software engineering and AI-native coding workflows.

Hands-On Exercises

1
Exercise 1: Build a Business Case

Choose one AI use case for your team. Build a one-page business case including: (a) The problem it solves (with current cost in hours/dollars). (b) The AI solution. (c) Build vs buy decision with justification. (d) Estimated cost (AI platform + development time). (e) Projected ROI over 6 months. (f) Key risks and mitigations. Present this to a colleague and refine based on their questions.

2
Exercise 2: Draft an Acceptable Use Policy

Write a 1-2 page AI acceptable use policy for your organisation. Include: approved tools, data classification rules (what can be sent to AI), prohibited use cases, human oversight requirements, and incident reporting procedures. Compare your draft with a colleague's -- what did they include that you missed?

3
Exercise 3: Use-Case Prioritisation

Brainstorm 10 potential AI use cases for your organisation. Score each on two dimensions: business impact (1-5) and implementation feasibility (1-5). Plot them on the 2x2 matrix from this module. Which three would you tackle first? Write a one-paragraph justification for each. Share with your manager for feedback.

4
Exercise 4: Design a Pilot

For your top-priority use case from Exercise 3, design a complete pilot plan. Include: scope (one use case, one team), duration (4-8 weeks), success criteria (with specific numbers), baseline measurement plan, participant selection (how many, from which teams), support structure (training, check-ins, escalation), and exit criteria (scale, iterate, or stop). Use the pilot framework from this module as a template.

5
Exercise 5: Change Management Plan

Write a change management plan for introducing AI tools to your team. Address: (a) What concerns will team members have? (List at least five.) (b) How will you address each concern? (c) Who will be your AI champions? (d) What training will you provide and when? (e) How will you collect and act on feedback? (f) How will you measure adoption success beyond just usage numbers? This exercise builds the skills for leading AI adoption in your organisation.

Start Claude Track --> Start Windsurf Track -->
<-- Safety & Ethics AI Foundations Complete