Build a Portfolio for Career Switching

A career-change portfolio should help someone examine how you approach work relevant to the role you want.

A polished final screen or document may show presentation quality. To demonstrate judgment, include the question you addressed, the materials available, the choices you made, and the checks you performed.

You can build this evidence through authorised work, independent projects, or clearly labelled simulations. Each supports different claims. A simulation can show your approach to a task; it cannot establish that an employer adopted your work or that it produced a business result.

Begin with the hiring question

Choose a target role and identify the tasks you need to demonstrate. Use the actual role description, relevant official occupational information, and any portfolio instructions from the employer.

Translate broad requirements into observable work:

Do not build a separate project for every phrase in a vacancy. Select material that answers an important question well.

CareerOneStop recommends including work samples with descriptions of the project, your contribution, and results. It also advises keeping the collection manageable and checking its presentation and links. This is practical US job-search guidance, not evidence that a particular portfolio will secure an offer. CareerOneStop portfolio guidance.

Choose material you can honestly explain and share

For an existing project, establish what you are permitted to disclose before building the case around it. Removing names or changing numbers does not automatically make employer or client material available for reuse.

If sharing permission is unclear, choose another sample or create an original simulation without copying restricted inputs. Label a reconstruction as a reconstruction and explain what was changed.

Use precise contribution statements:

• “I wrote the analysis and prepared the recommendation; another team member collected the data.”
• “I implemented these components within a team project.”
• “This is an independent exercise using synthetic information. It was not commissioned or deployed.”

A result also needs a boundary. “The client approved the recommendation” differs from “The recommendation increased retention.” Claim only what the evidence supports.

A complete worked sample: checking event registration data

The following is a fictional teaching example. Leo, the assignment, every record, the review comments, and the conclusions are invented. It demonstrates a portfolio case structure, not a real client engagement or hiring outcome.

Leo is an administrative coordinator exploring operations analyst roles. He wants a sample that shows how he checks data and communicates uncertainty.

The brief

A fictional event organiser asks:

How many confirmed registrations are in this export, and what still needs checking before we give a planning number to the event team?

Leo defines the task narrowly. He will reconcile registrations in a single snapshot. He will not predict attendance or claim to know how many people will arrive.

The materials

He creates this synthetic CSV. The repeated names are intentional; all names are invented.

registration_id,display_name,status
R01,Ari,confirmed
R02,Sam,confirmed
R03,Jo,cancelled
R02,Sam,confirmed
R04,Priya,
R05,Alex,confirmed
R01,Ari,confirmed
R06,Alex,cancelled
R07,Mei,confirmed
R08,Noor,
R05,Alex,confirmed
R09,Luis,confirmed

He includes the CSV in the sample so that a reviewer can reproduce the checks. He also states an assumption: within this exercise, the registration ID identifies a registration, and repeated rows with the same ID and identical values are duplicate exports.

That assumption would need verification with the data owner in a real assignment. If one ID contained conflicting statuses, simply keeping the first row would not resolve the conflict.

First approach and the error it reveals

Leo initially tries removing duplicates by display name. This would combine the two records named Alex even though they have different registration IDs and statuses.

He records why the method is unsuitable: a shared name does not establish a shared registration. He changes the duplicate rule to use the stated identifier and checks whether repeated IDs contain conflicting values.

This correction belongs in the case. It shows how he identified and revised a weak decision.

Reconciliation

The sample contains twelve rows. R01, R02, and R05 each have one additional identical row, leaving nine distinct registration IDs.

The unique records reconcile as follows:

Leo treats missing status as unknown. He does not silently recode it as confirmed or cancelled.

The confirmed records are R01, R02, R05, R07, and R09. R03 and R06 are cancelled. R04 and R08 need clarification.

The decision note

His final note reads:

This synthetic snapshot contains five confirmed registrations after removing three additional identical rows. Two registrations are cancelled and two have no status. If both unknown records are later confirmed, the confirmed count would become seven. This is a range of possible registration statuses, not an attendance forecast. Confirm the two missing statuses and the meaning of the identifier before using the result operationally.

The note answers the brief and makes the remaining decision visible.

It avoids an unsupported “improvement” claim. Leo has not saved a real organisation money or improved an actual event. He has demonstrated a checking process on a deliberately small dataset.

Review and revision

In the fictional review, a reader asks, “Why is seven not your forecast?”

Leo revises the heading from “Expected attendance” to “Registration status reconciliation” and places the explanation beside the table. He also adds a note that different IDs may represent separate registrations by the same person; determining unique people would require another definition and appropriate information.

The revision improves the accuracy of the sample’s claim. It does not make the dataset more realistic than it is.

What this case demonstrates

The sample gives a reviewer evidence of problem definition, identifier checking, reconciliation, handling missing information, and explaining a conclusion.

It leaves other questions open. It does not show Leo working with large datasets, combining multiple systems, operating under workplace pressure, or obtaining stakeholder agreement. He should describe those as gaps or provide other evidence where the target role requires them.

Build the case around the evidence chain

For your own sample, copy this structure:

Give the reader a short route through the case, with supporting material available where it adds value. A concise introduction can state the task and strongest evidence; the fuller version can show decisions and checks.

The US Office of Personnel Management describes work-sample assessments as tasks that resemble job activities, evaluated under defined conditions. A self-selected portfolio is a different form of evidence. Its quality depends on what the reviewer can inspect; it does not inherit the validity of a structured hiring assessment. OPM work samples and simulations.

Review the sample before adding more projects

Ask a suitable reader to examine a specific question: “Can you follow why I excluded these records?” or “Which required capability remains unsupported?” General approval is less informative.

Use the response to decide what comes next:

Low response to applications does not establish which part of the portfolio is wrong. Role requirements, eligibility, the application, and other factors may matter. Look for specific evidence before repeatedly redesigning the presentation.

Interest questionnaires cannot fill an evidence gap. If you keep an interest report alongside your planning, the RIASEC PDF and download guidance addresses report-related questions; the portfolio itself should show relevant work.

When you have a sample, use it to refine your five-year career roadmap. If the difficulty is transferring experience into another sector, use Cross-Industry Move Strategy. You can find other resources in the career guides library.

Before sending the portfolio, check that the files open, the links work, the content is readable, and every claim matches the evidence provided.