Your experience is a head start

What you will be able to do

See your years in another field as a strength in AI, not as lost time. You learn what domain knowledge is, why AI projects fail without it, and three ways to use it: contribute to AI work, lead it, or build it.

Before you startWhat to know and to have readyAbout 35 minutes

What you need to know

• The five areas and the three levels, from Skills inventory: Five areas, not one and Skills inventory: Three levels.
• This lesson is for everyone. If you are a fresh graduate, your field of study, a part-time job or a serious hobby is a domain too.

What you need

• Your first inventory, under ## Skills inventory in notes.md. You wrote it in Skills inventory: Guided practice: your first inventory. If it is missing, do that guided practice again.
• This lesson adds the heading ## My head start.

You do not start from zero

Many people who want to move into AI say the same thing: “I have no AI background, so I start from zero.” If you have worked in another field, this sentence is wrong.

AI is almost always used in a field: in a bank, a hospital, a law firm, a shop or a support team. The technical team can learn how to build the AI. It is much harder for them to learn how that field really works. People who already know the field are rare, and AI teams need them in every phase of a project.

Remember Grace, the support lead at Larkfield in One project, many roles. She did not write one line of code. But without her, the team would have used old documents and tested with wrong answers. It would also have missed the five refund errors.

What domain knowledge is

Domain knowledge is the knowledge of one field of work that you get by doing the work: its words, its rules, its people and its risks. A domain is that field, for example accounting, nursing, law, retail or customer support.

Domain knowledge is more than facts that you can read in a manual. It has these parts:

Part	What it means	Example at Larkfield
The real meaning of words	Words in the field have exact meanings that outsiders mix up.	“Damaged” means broken in delivery, which is a delivery claim. “Faulty” means it stopped working, which is a warranty claim. The two follow different rules.
Rules and their exceptions	The written rule, and every case where it does not apply.	Returns are accepted for 30 days, but for 60 days for orders made in December.
How the work is really done	The real steps, which are often different from the manual.	Agents check the customer’s order history before they read the policy.
What the data really means	How values were recorded, by whom, and why.	“Closed” on a ticket does not always mean “solved”. Some agents close tickets to meet a time target.
Which mistakes are expensive	The errors that cost the most money, trust or safety.	A wrong promise of a refund is serious. A wrong delivery time is a small problem.
What people trust	What users need to see before they act on an answer.	Agents trust an answer when they can see the source and the date.
The laws of the field	Regulations, professional standards and duties.	Consumer law gives customers some rights that no shop policy can remove.

None of this is in the AI model. Most of it is not in the data either. It is in the heads of the people who do the work.

Why AI needs it: the data does not explain itself

AI learns patterns from data. But data records what happened, not why it happened. Without somebody who knows the why, a model can learn a pattern that is true in the data and dangerous in real life.

A well-known example comes from health research in the 1990s. Researchers trained models to predict the risk of death for patients with pneumonia, a lung infection. The aim was to decide which patients could be safely treated at home.

One model learned a rule: patients with asthma have a lower risk of death. Doctors saw at once that this was wrong. Asthma makes pneumonia more dangerous, not less.

The data was correct. Hospitals knew that asthma patients were at high risk, so they sent them directly to intensive care. The extra care worked, and fewer of them died. The model learned the effect of the treatment, not the risk of the disease. If the model had been used, it could have sent home the patients who needed the most care. Rich Caruana and colleagues described this case in a research paper in 2015.

The engineers could not see the problem in the numbers. A doctor saw it in seconds. That is the value of domain knowledge.

Where domain knowledge enters a project

Each phase of an AI project has a question that only a domain expert can answer well.

Text description of the diagram

• Frame: which problem costs the most today?
• Data: what does each value really mean?
• Build: what are the rules and the exceptions?
• Evaluate: what is a correct answer, and which mistakes are serious?
• Operate: will people trust it, and what changed in the field?

Look at Evaluate. To measure an AI system, somebody must decide what a correct answer is. For a legal question, a medical summary or a tax rule, only a domain expert can decide that. This is why many AI teams hire domain experts to write test cases and judge the outputs.

A head start, not the whole path

Domain knowledge gives you a head start. It does not replace the other skills. There are two opposite mistakes:

• “I start from zero.” You hide your experience and compete with new graduates on programming alone. You lose your strongest advantage.
• “My experience is enough.” You do not learn how AI works. Then you cannot tell the team what is possible, you cannot judge their results, and they cannot use your knowledge well.

Everybody who works with AI needs AI literacy: how models learn from data, how they fail, what they cost, and how teams measure them. This course and the courses after it give you that. How much programming you add depends on the way that you choose.

Three ways to use your experience

Three ways to use domain experience in AI. You can start at any of them, and move between them.

Text description of the diagram

• Contribute: subject-matter expert on an AI team. Coding level 0.
• Lead: AI product, evaluation, adoption or governance in your field. Coding level 0 to 2.
• Build: analyst, data scientist or engineer in your field. Coding level 2 to 3.

Contribute plus AI literacy can lead to Lead. Lead plus programming can lead to Build.

Contribute

You work with an AI team as the subject-matter expert. You define correct answers, write test cases, review outputs and explain the field to the team. This can be part of your current job. Many companies look for experienced people inside the business to help their AI projects.

Fastest start. You need AI literacy, and you need to explain your knowledge clearly.

Lead

You lead AI work in your field. Here are some examples:

• AI product manager for products in your field, for example AI tools for accountants.
• Evaluation lead, who decides how the team measures quality, and which mistakes are not acceptable.
• AI adoption lead, who changes how your department works with AI, trains colleagues and measures the results.
• AI governance and risk in your field, for example AI rules in a bank or a hospital.
• Founder of a company that solves a problem that you know from the inside.

Domain experts are well placed to lead AI products. The hardest questions in an AI product are often domain questions: Which problem matters? What is a correct answer? Which mistakes are not acceptable? Will people trust it? A leader who can answer these questions saves the team months. Engineers know how to build. A leader must also know what to build and why.

Needs: AI literacy, product and project skills, and often some data skills (level 1–2).

Build

You learn the technical skills and become a builder who also knows the field. Examples are a data scientist in health care, an ML engineer in a bank, or a “legal engineer” who builds tools for lawyers. Some of these hybrid roles have their own names, such as clinical informatics in health and quantitative analyst in finance.

Takes longest, because you add programming (level 2–3). But the result is rare: few people can both build the system and know when its answers are wrong.

You can start at any of the three, and move between them. Many people start by contributing, inside their current job, and then move to leading or building.

Worked example: Daniela translates her experience

Daniela, the accountant from Skills inventory, writes what her experience gives an AI team. She writes each item as “I know …, so I can …”:

I know …	… so I can …
the 20 ways that supplier invoices go wrong: duplicates, wrong tax, wrong supplier number	write test cases for an AI that reads invoices, including the difficult ones.
which accounting errors auditors care about, and which they do not	tell the team which AI mistakes are serious.
that month-end is the busiest week, and that nobody tries new tools then	plan a pilot that people will actually use.
the tax rules for our three countries	check whether the AI’s tax answers are correct.
how to explain numbers to the finance director	present the AI’s results honestly to the people who decide.

With this table, Daniela sees two good first routes: contribute to an invoice AI project in her company now, and work toward leading AI products for finance teams. She also writes what she must learn: AI literacy now, and basic Python and SQL later, so she can check data herself.

Guided practice: translate your experience

1. In notes.md, add a heading: ## My head start.
2. Write the name of your domain. If you are a fresh graduate, use your field of study, a job that you had, or a field that you know well.
3. Write three lines as “I know …, so I can …”, like Daniela’s.
4. Under the table, write which of the three ways (contribute, lead or build) interests you most, and one reason.

Check: each “I know” line must be specific to your field. “I know how to work hard” is true for every field, so it does not count. “I know that customers often say ‘broken’ when they mean ‘it arrived broken’” is specific.

Independent variation: choose a person from a different field than yours, such as a teacher, a nurse or a shop manager. Write two “I know …, so I can …” lines for them. This shows you how much domain knowledge other people have, and it prepares you for the next lesson.

Common mistake

Hiding your past career

Symptom: your CV and profile start with “Aspiring AI engineer” and list only online courses. Ten years as a pharmacist are one line at the bottom. You get no interviews, or only interviews for junior jobs where you compete with new graduates.

Cause: you believe that your past career is not relevant to AI. So you hide your rarest skill, and show only the skills where you are a beginner.

Fix:

1. Put your domain first: “Pharmacist with 10 years of experience, now building skills in AI evaluation.”
2. Add your “I know …, so I can …” lines as evidence.
3. Look for AI work in your field first: health AI companies, or AI projects in pharmacies and hospitals. There, your experience is worth the most.

Check your understanding

In the pneumonia example, the data was correct and the model learned the pattern correctly. So what went wrong?

Show the answer

The pattern was real, but its cause was not the one that the model assumed. Asthma patients died less often because they got intensive care, not because asthma is safe. Data shows what happened, not why. Only someone who knew how hospitals treat asthma patients could see that the rule was dangerous.

Check your understanding

A retail store manager says: 'I know the shop floor, so I can lead an AI product for shops without learning anything about AI.' What is missing?

Show the answer

AI literacy. To lead an AI product, they must know how models learn and fail, what they cost, and how quality is measured. Without that, they cannot judge what the team proposes, or explain the risks to the people who decide. Their domain knowledge is the head start. AI literacy is what makes it usable in an AI team.

Recap

• If you have worked in another field, you do not start from zero. AI teams need people who know the field.
• Domain knowledge includes the real meaning of words, rules and exceptions, how the work is done, what the data means, which mistakes are expensive, what people trust, and the laws of the field.
• Data shows what happened, not why. Domain experts see dangerous patterns that the numbers hide.
• Domain knowledge is a head start, not the whole path. Everybody needs AI literacy.
• Three ways to use it: contribute, lead or build.

Next, you see what this looks like in finance, health, legal, retail and other fields.

Next: Domain knowledge in practice