The AI Landscape For Business
You learn what AI can reliably do inside a company today, how to tell real value from hype, and how ready your own organization is. You finish with a scored readiness self-check you can share with colleagues.
You Will Be Able To
- Explain AI, machine learning and large language models in plain words to a colleague.
- List the task families where AI delivers dependable value in finance, HR, customer service, operations, legal, sales and software.
- Separate a vendor claim from measured value using four simple questions.
- Describe the limits of language models, including hallucination and missing company knowledge.
- Score your organization on five readiness dimensions and name the weakest one.
Why Companies Are Talking About AI Now
Artificial intelligence (AI) is software that performs tasks that normally need human judgment, such as reading text, spotting patterns or writing a first draft. For many years it lived inside specialist teams. Since 2022, tools built on large language models (LLMs), which are programs trained on huge amounts of text to predict and produce language, have reached ordinary employees through a simple chat window. A manager can now ask a tool to summarize a forty page contract in a minute. That shift is why AI appears on almost every board agenda.
This course treats AI as a business capability, not a technology project. Over twelve weeks you move from understanding the landscape to a written roadmap for your own organization. Every week speaks to two groups: leaders, who decide where money and attention go, and employees, who use the tools every day. Both groups need the same shared vocabulary, so we build it first.
Three Kinds Of AI You Will Meet
Machine learning (ML) is the older and broader idea: models that learn patterns from past data instead of following hand-written rules. A bank uses ML to score the chance that a payment is fraud. A factory uses it to forecast demand. These systems usually take structured data, meaning rows and columns, and return a number or a label.
Generative AI creates new content such as text, images, code or audio. LLMs are the generative systems most companies meet first. Computer vision and speech recognition are a third family: they read images, scanned documents and phone calls and turn them into data that other systems can use. Most real business solutions combine two or three of these families with ordinary software.
| Family | Typical input | Typical output | Business example |
|---|---|---|---|
| Machine learning | Rows and columns | Number or label | Fraud score, demand forecast |
| Generative AI (LLMs) | Text, images, code | New text, code or media | Ticket summary, contract draft |
| Vision and speech | Scans, photos, calls | Text and fields | Invoice reading, call transcript |
Most real solutions combine two of these families with ordinary software.
What Works Today
Dependable value sits in a handful of task families. Language models are strong at drafting, summarizing, translating, classifying and extracting information from text. Classic ML is strong at forecasting, scoring and detecting unusual events. Vision tools read invoices, forms and photos. Speech tools turn calls into searchable text.
The examples below are common in mid-size and large organizations:
- Customer service: draft replies, summarize long tickets, suggest answers to agents during a call.
- Finance: extract fields from invoices, flag unusual transactions, draft variance commentary.
- HR: draft job descriptions, summarize survey comments, answer routine policy questions.
- Legal: first-pass contract review, clause comparison, summaries of regulations.
- Sales: call summaries, proposal drafts, research on an account before a meeting.
- Operations: demand forecasts, maintenance alerts, summaries of incident reports.
- Software: code completion, test writing, explanations of old code.
Where AI Struggles
A language model produces fluent text, but fluent is not the same as correct. A model can state a wrong fact in a confident tone. People call this a hallucination. A model also knows nothing about your company, your prices or last week's policy change unless you give it that information. Answers can vary from one run to the next, and the model cannot always explain why it answered as it did.
These limits shape what to automate. Tasks where an error is cheap and easy to spot suit AI well, for example a draft email that a person reads before sending. Tasks where one error causes legal, financial or safety harm need strong checks or a different tool. Arithmetic is a simple case: a spreadsheet beats a language model at calculation, so good solutions pair the model with proper tools. In every example above, a person stays responsible for the result. This pattern is called human in the loop, and it is the safest way to start.
Hype Versus Value
Hype describes what a technology might do someday. Value is a measured improvement in cost, speed, quality or risk inside your own processes. A useful habit is to ask four questions of any claim: which task, which people, which measure, compared with what. A statement such as 'our assistant boosts productivity' answers none of them.
Research gives a realistic picture. In a widely cited study of customer support agents, Brynjolfsson, Li and Raymond found that an AI assistant raised the number of issues resolved per hour by about 14 percent on average, and by about 34 percent for the least experienced workers. These are meaningful gains. They also show that benefits differ by role and skill level. Expect real but modest gains in specific tasks, measured against a baseline, rather than overnight transformation of the whole company.
- Hype signals: no named task, no baseline, demos instead of data, savings stated as 'up to'.
- Value signals: one task, a named group of users, a before and after number, an owner.
Gains were largest for the newest workers, which shows benefits differ by skill level. Source: Brynjolfsson, Li and Raymond, Generative AI at Work, NBER Working Paper 31161
- Pick one task
- Measure baseline
- AI drafts
- Person checks
- Compare results
Every credible AI claim can be traced through these five steps.
Two Audiences, Two Jobs
Leaders set direction. They choose a small number of use cases, fund the data and training work behind them, and own the risk. A leader who asks for 'an AI strategy' without naming a business problem usually gets a pile of disconnected experiments.
Employees decide whether any tool becomes real. They learn what the tool does well, tell the organization what works and what does not, and follow the rules on company data. Staff often adopt AI tools before the company has a policy. That is a risk we cover in week 5, but it is also a signal: people want help, and the best early ideas often come from the people doing the work.
A Readiness Self-Check
Before choosing projects, take stock. Score your organization from 1 (not started) to 4 (working well) on five dimensions. Strategy: do leaders agree on where AI should help? Data: can the right people find and use clean data? People: do staff have the skills and the time to learn? Technology: do systems allow secure access to models and data? Governance: do clear rules exist on who may use what, and who is accountable?
The point is not a perfect score. The point is to find the weakest dimension, because it limits everything else. A company with excellent tools and no rules on data creates risk. A company with clear rules and messy data cannot deliver results. You will return to this score in week 12 when you write your roadmap, and it will have moved.
How To Use This Course
Weeks 1 to 6 build the foundations: finding use cases, preparing data, choosing between build, buy and partner, handling security and law, and designing a pilot. Weeks 7 to 12 turn these into practice: workflows by department, change management, costs and return, the operating model, scaling, and a final roadmap.
Each week has a study document, a slide deck and one exercise. Read the document first, use the slides to discuss it with colleagues, and do the exercise with real examples from your own work. Keep one shared notes file for the whole course. By week 12 it becomes the raw material for your roadmap.
A practical tip for the weeks ahead: share what you learn. Run a short monthly session where people show one task they improved with AI, how they checked the result, and what went wrong. These sessions spread good habits quickly and surface risks early, at almost no cost.
Key Terms
- Artificial intelligence (AI)
- Software that performs tasks that normally need human judgment, such as reading, predicting or drafting.
- Machine learning (ML)
- Models that learn patterns from past data instead of following hand-written rules.
- Large language model (LLM)
- A program trained on large amounts of text that predicts and produces language.
- Generative AI
- AI that creates new content such as text, images, code or audio.
- Hallucination
- A confident but wrong or invented statement produced by a language model.
- Human in the loop
- A design where a person reviews or approves the AI output before it is used.
- Baseline
- The measured result of a process before any AI is added, used for comparison.
- Use case
- A specific task in a specific team where AI could help, described with a measurable goal.
- Readiness
- How prepared an organization is across strategy, data, people, technology and governance.
Exercise: Score Your Readiness And Test One Claim
- Copy the five readiness dimensions into a spreadsheet and score each from 1 to 4 for your organization or team.
- Write one sentence of evidence next to each score, such as 'no data owner named for customer records'.
- Circle the lowest score and write two actions that would raise it by one point.
- Find one vendor or news claim about AI productivity. Apply the four questions: which task, which people, which measure, compared with what.
- Pick one task from your own week that you think AI could draft. Write down how long it takes now. This is your first baseline.
- Share the sheet with one colleague from a different department and compare scores. Note where you disagree.
Further Reading
- Prediction Machines: The Simple Economics of Artificial Intelligence, Ajay Agrawal, Joshua Gans and Avi Goldfarb, Harvard Business Review Press
- Generative AI at Work, Erik Brynjolfsson, Danielle Li and Lindsey Raymond, NBER Working Paper 31161
- Co-Intelligence: Living and Working with AI, Ethan Mollick, Portfolio
- The AI Index Report, Stanford Institute for Human-Centered AI, annual report
- The State of AI, McKinsey and Company, annual global survey