Turn a CSV/CSA Maturity Profile into an Owned Improvement Plan
A CSV/CSA maturity profile tells you where to look. An improvement plan says who will do what, what will show it was done and who will check. This guide walks from one to the other in nine steps, with an action register you can copy.
Written by CSV/CSA Maturity Editorial TeamPublished 10 min read
On this page6 sections
A maturity profile becomes an improvement plan when a small number of its prompts are turned into actions that each have an owner, a stated piece of evidence that will show completion, a known dependency and a date the owner set. Start by confirming the scope. Then separate what the team does not know from what it knows to be weak, because the first calls for verification and the second for a change in practice.
This guide is for team leaders who hold a complete or partial self-reported profile from the CSV/CSA Maturity Assessment, which covers computer system validation (CSV) and computer software assurance (CSA) practices. CSV here does not mean comma-separated values. The output is a planning document that lives in your own tools and follows your own quality system. It is not a compliance plan, and nothing in it predicts a future score.
Nine steps from profile to plan
1. Check the scope
A profile describes one team, one site or one system portfolio. Confirm that every respondent answered for the same one. If two teams with different practices were averaged into a single set of answers, the actions that fall out will fit neither. Narrow the scope and repeat the affected answers before planning. Our how-to on running a scoped, evidence-led self-assessment covers this preparation.
2. Separate unknowns from low practice descriptions
Two kinds of answer drive a plan, and they need different responses:
- A low practice description (0 or 1) was selected by someone who knows the practice. The response is a change in how the work is done.
- "I do not know yet" means nobody could support a choice. The response is to find the record and then choose.
Unknown is not zero. The model never scores an unknown answer as 0. It leaves it out of numeric coverage, and a profile with any unknown answer is partial: it shows a possible range and no overall score or band. The assessment page explains how unknown answers are handled in the scoring.
3. Review the priority review prompts
Three of the 18 questions carry a priority review prompt: how relevance to regulated work is decided (Q02), how assurance effort is connected to function-level risk (Q05) and how assurance records are protected (Q15). An answer of 0 on any of them asks you to review that practice independently of the overall score. An answer of "I do not know yet" asks you to verify the evidence. Neither changes the arithmetic.
The prompts exist because an average can hide a weak practice. In one of the model's published examples every answer is 3 except Q15, which is 0. The profile still reads 92.9, in the band the rubric calls Measured, with the Q15 prompt beside it. Look at these prompts before you look at the lowest area score.
4. Choose a small number of actions
The profile proposes at most three primary actions. Its ordering is fixed by rule: priority questions answered 0 come first, then priority questions answered unknown, then other unknown answers, then answers of 0 or 1 with the lowest first. A later action in a practice area that already has a primary action is merged into it, so the three fall in different practice areas where possible.
That order is a starting sequence, not a risk ranking. Someone who knows the systems may change it. Keep the first plan close to that size all the same. Every action in the register needs an owner's time and a second person to verify it.
5. Assign an owner role, then a person
The model suggests an owner role for each practice area. Use the role to find the person, and put the person's name in your own register.
| Practice area | Suggested owner role | Completion signal |
|---|---|---|
| System inventory | System owner and quality lead | A sampled new or changed system is reflected in the inventory and its relevance decision. |
| Intended use | Process owner with users and quality | Another reviewer can explain the boundary and expected behavior without asking the author. |
| Risk-based assurance | Quality and relevant technical owners | The reviewer can reconstruct why this approach was proportionate for the selected function. |
| Traceability | Validation or assurance lead | The chain remains understandable after a representative change. |
| Change management | Change owner, system owner and quality | The decision and follow-through can be reconstructed from maintained records. |
| Approvals and record controls | Record owner and quality | The sample can be attributed, read and understood under the applicable procedures. |
| Evidence and review readiness | Quality, process owner and training owner | An unfamiliar reviewer can locate and understand the relevant evidence and see whether the action was checked. |
Every completion signal in that table is something a reviewer observes. None of them is "a document exists".
6. Define the evidence of completion before work starts
For each action, write down what a reviewer will be able to look at when it is done. Two sources help. Each question in the assessment names the evidence to check, for example "A recent change with impact rationale, approval and closure record." And each practice description says what the practice looks like at the next level. If the evidence you name would not let a reviewer agree that the higher description is true, name different evidence.
7. Sequence the dependencies
Some actions are inputs to others. A verification usually comes before a change, because its result decides whether the change is needed. An action that relies on another team's schedule, such as a change board date or a supplier's release calendar, should say so. Record each dependency next to the action, so that a late date can be explained without a meeting.
8. Review the outcome
When an owner reports an action as complete, a second person looks at the evidence. That person records what was examined and what it showed. Anything unresolved stays open with a new date. An action is closed by this review, not by the owner's report.
9. Reassess with compatible versions
Repeat the assessment only when there is new evidence to answer from, and keep three things constant: the scope, the model version and the question bank version. The current model version is csa-maturity-2.0.0 and the question bank version is 2.0.0. If either has changed since your first profile, label any difference as a change in method, not as improvement.
Two further rules keep the plan and the profile from contaminating each other. Closing an action does not by itself justify a higher answer: reselect each answer against the wording of the practice descriptions. And changing an answer in the assessment closes nothing in your quality system.
The action register
The register has seven fields, split across two tables so it stays readable on a small screen. The ID column joins them.
Empty register to copy
| ID | Action | Trigger | Owner role |
|---|---|---|---|
| A1 | |||
| A2 | |||
| A3 |
| ID | Evidence of completion | Dependency | Target date | Verification |
|---|---|---|---|---|
| A1 | ||||
| A2 | ||||
| A3 |
Field notes:
- Trigger is the answer or prompt that caused the action, with its question ID.
- Owner role holds the role in a shared example. In your own register add the person's name.
- Evidence of completion is what a reviewer will look at, not a statement that the work was done.
- Verification names who checks the evidence and what they do with the result.
Completed synthetic example
Synthetic sample. The profile and the register below are invented. The numbers come from running the answers through the published scoring model (model version csa-maturity-2.0.0). They do not describe any organization.
Take the worked example published with the assessment (54.8, in the band called Defined), which our tutorial on reproducing the 54.8 profile answer by answer sets out in full, and change two answers to "I do not know yet": supplier releases (Q12) and record protection (Q15). The profile becomes partial, with 16 of 18 numeric answers (numeric coverage 88.9%), no overall score or band, and a possible range of 48.4 to 60.4. Change management shows a possible range of 33.3 to 66.7, and approvals and record controls 33.3 to 83.4. Q15 raises a prompt to verify the evidence.
The profile proposes three primary actions: check how assurance records are protected (Q15), look at how supplier releases are handled (Q12, merged with the low answer on Q11) and run a small evidence-retrieval exercise (Q16, merged with Q18). The team keeps all three and writes the Q11 change out as its own action, because it depends on what the Q12 check finds.
| ID | Action | Trigger | Owner role |
|---|---|---|---|
| A1 | Verify how assurance records are protected | Q15 answered "I do not know yet"; verification prompt | Record owner and quality |
| A2 | Find one supplier release and its internal assessment | Q12 answered "I do not know yet" | Change owner, system owner and quality |
| A3 | Require an impact review before changes are implemented | Q11 answered 1: reviews "occur inconsistently or after implementation" | Change owner, system owner and quality |
| A4 | Run a small evidence-retrieval exercise for one system | Q16 answered 1; Q18 answered 1 | Quality, process owner and training owner |
| ID | Evidence of completion | Dependency | Target date | Verification |
|---|---|---|---|---|
| A1 | Record of a demonstration of the access controls and one retrieval or recovery check | None | Day 14 | Quality lead reads the record. The team then reselects the Q15 answer |
| A2 | One supplier release notice with its internal assessment and disposition, or a note that none was found | None | Day 21 | System owner confirms the example is typical. The team reselects Q12 |
| A3 | (a) Approved procedure change. (b) Three consecutive changes after its effective date, each with an impact rationale dated before implementation | A2, so that supplier releases enter the same review | (a) Day 45; (b) Day 85 | A reviewer who was not involved in the changes samples them and records what was checked |
| A4 | Retrieval log written by a reviewer who did not build the package, with the gaps listed | A1, so the reviewer knows which copy is authoritative | Day 60 | Each gap has an owner and a closure or an approved disposition |
Days are counted from the day this fictional team agreed the plan. A real register carries calendar dates set by each owner. Two features of the example deserve attention. A1 and A2 change nothing: they replace an unknown with a supported answer, which may turn out to be high or low. And A3 has two pieces of evidence with two dates, for the reason set out next.
Why a written procedure is not consistent application
Action A3 could be reported as complete on Day 45, when the revised procedure is approved. It would be premature. Compare what each piece of evidence shows.
| Evidence | What it shows | What it does not show |
|---|---|---|
| Approved procedure that requires an impact review before implementation | The expectation is defined and authorized | That any change has followed it |
| Training record for the change owners | The people concerned were told | That they apply it |
| Three consecutive changes after the effective date, each with a rationale dated before implementation | The practice was applied in those cases | That it will continue, or that each rationale was sound |
| A later sample taken by a reviewer who was not involved | Application has persisted and someone is checking | That the control works in every case |
The assessment's wording draws the same line. For change impact (Q11), the description scored 1 reads "Impact reviews occur inconsistently or after implementation." The description scored 2 reads "Changes have an appropriate pre-implementation impact review, responsible owner and approval." The subject of that sentence is changes, not a document. A procedure on its own supports neither description. Only records of changes do.
The same applies in every practice area. Writing an intended-use template does not show that owners review descriptions when workflows change. Publishing a traceability standard does not show that links survive a change. Plan the second piece of evidence from the start, and give it its own date.
The 30, 60 and 90 day frame, used as a frame
The assessment model includes a three-window planning frame. The wording below is the model's own.
| Window | Focus |
|---|---|
| Days 1 to 30 | Agree scope, verify unknowns, name owners and choose a few evidence-backed actions. |
| Days 31 to 60 | Implement the selected practice changes and retain reviewable examples. |
| Days 61 to 90 | Sample application, review unresolved issues and repeat the assessment using a comparable scope and version. |
The model attaches this note: "This is a planning template, not a promise that an organization becomes compliant or mature in 90 days. Set dates based on your actual risk, resources and change process."
Three cautions follow from it:
- The frame is not produced by your score. A low profile does not mean 90 days, and a high one does not mean fewer.
- A practice change that goes through formal change control may take longer than a window. Say so in the register and keep the evidence definition unchanged.
- An urgent issue found during a real review follows your established process immediately. It does not wait for a window.
What this plan does not establish
- It does not promise a higher score. A verification action can end in a low answer as easily as a high one, and that is a useful result.
- It does not establish compliance with any regulation. The profile is self-reported and the plan is a planning document.
- It does not replace your quality system. Where an action belongs in a formal process such as change control or corrective action, that process governs and the register only points to it.
- The synthetic example shows the mechanics. Its actions and dates are not recommendations for your team.
If you do not have a profile yet, you can create a starting profile for one defined scope. The assessment asks for the scope, takes you through seven practice areas and a review step, then asks for your first name, last name, work email and company before it shows the profile, which you can download as a PDF.
Related reading
- How to assess CSV/CSA maturity: agree the scope and collect a small evidence sample before you answer.
- Interpret a CSV/CSA maturity profile: a worked example: the full arithmetic of the 54.8 example that the synthetic register starts from.
- Why unknown answers and high scores need careful interpretation: what a reassessment can and cannot tell you.
This article describes assessment model csa-maturity-2.0.0.
