What You Will Be Able To Do

• Explain AI, ML and LLMs in plain words
• Name dependable task families by department
• Apply four questions to any vendor claim
• Describe model limits such as hallucination
• Score readiness on five dimensions

Why Every Board Is Asking About AI

• Since 2022, chat tools put language models in every employee's hands
• Tasks like summarizing a contract now take about a minute
• Interest is high, but results depend on how work is redesigned
• This course treats AI as a business capability

Three Families Of AI

• Machine learning: learns patterns from past data
• Generative AI: creates text, code, images, audio
• Vision and speech: turn scans and calls into data
• Real solutions combine families with ordinary software

What Works Today, By Department

• Customer service: draft replies, summarize tickets
• Finance: extract invoice fields, flag odd transactions
• HR: draft ads, summarize survey comments
• Legal and sales: first-pass review, call summaries

Operations And Software

• Operations: demand forecasts, maintenance alerts
• Operations: summaries of incident reports
• Software: code completion and test writing
• Software: explaining old code to new team members

How A Language Model Behaves

• It predicts likely next words from patterns in training text
• It sounds confident whether or not it is right
• It does not know your company unless you supply facts
• Answers can differ between runs

Where AI Struggles

• Hallucination: invented facts in a confident tone
• No built-in knowledge of your prices or policies
• Weak at exact arithmetic without extra tools
• Cannot always explain why it answered

Choose Tasks By Cost Of Error

• Cheap, visible errors: good fit, such as email drafts
• Costly or hidden errors: add checks or use another tool
• Always name the person who is responsible
• Start with human in the loop

Error cheap and easy to spot  -> start here
Error costly or hard to spot  -> strong checks first
Error harms safety or law     -> keep a person deciding

Hype Versus Value: Four Questions

• Which task exactly?
• Which people will use it?
• Which measure will change?
• Compared with what baseline?

Evidence From Customer Support

• AI assistant for support agents, studied by Brynjolfsson and colleagues
• About 14 percent more issues resolved per hour on average
• About 34 percent for the least experienced agents
• Gains differ by skill and role

Two Audiences, Two Jobs

• Leaders: pick few use cases, fund data and training, own risk
• Employees: learn tools, report what works, protect company data
• Best early ideas often come from people doing the work
• Staff already use tools before policy exists

Five Readiness Dimensions

• Strategy: agreed places where AI should help
• Data: right people can find and use clean data
• People: skills and time to learn
• Technology: secure access to models and data
• Governance: clear rules and accountable owners

Scoring And What To Do With It

• Score 1 (not started) to 4 (working well)
• Write one line of evidence per score
• Find the lowest score and plan two actions
• Repeat in week 12 to see movement

Recap

• AI is a business capability built on three families of tools
• Value is measured against a baseline in a named task
• Models are fluent but can be wrong, so keep people in the loop
• Readiness has five dimensions, and the weakest sets the pace

Next Week

• Finding use cases
• Map one process step by step
• Score ideas on value and feasibility
• Separate quick wins from bigger bets

Questions And Exercise

• Exercise: score readiness and test one vendor claim
• Capture one baseline from your own work
• Compare scores with a colleague from another department
• Questions from the room