A report writing workflow that keeps the evidence visible

What is a report writing workflow?

A report writing workflow is the full chain from scoping the evidence to signing off the final text: scoping, source handling, synthesis, drafting, review and sign-off. Drafting is one stage of it, and it is rarely the stage where a report goes wrong.

Reports seldom fail because the prose is weak. They fail when the logic between evidence, findings, conclusions and recommendations is hard to follow. Caveats disappear, appendices carry the wrong material, or recommendations drift beyond what the evidence supports. For programme teams, this workflow is the drafting end of the wider monitoring and evaluation reporting workflow for NPOs.

The UK government's functional standard for analysis describes the same shape as a cycle: initial customer engagement and scoping, designing the analysis, conducting and checking it, delivering it, and sign-off (GovS 010: Analysis, version 2.2, 2025). A report workflow follows that cycle, with the trail back to source kept intact at every step.

What should you decide before anyone starts drafting?

Decide what the report has to help its audience decide or do, which questions it must answer, and what evidence is in scope. If those are still open, the report is not ready to be written, even if the template is.

CDC's Program Evaluation Framework (2024) treats this as a step of its own: before interpretation, the team determines the evidence needed to answer the evaluation questions, including what data will be collected, how, when and from whom. CDC's guide to evaluation reporting (2013) adds the reader's side: define the target audience and the action you want them to take, then shape the report around both.

Before drafting, write down:

• the decision or action the report must support, and who takes it
• the questions the report must answer
• the evidence in scope, and the source types that feed it
• the subgroups, places or time periods that matter
• the unit of analysis: a submission, an interview, a case, a site visit or a coded claim
• the level of traceability reviewers will expect
• who reviews each section, and who signs off

The unit of analysis deserves more attention than it usually gets. If unlike material is mixed too early, for example by counting one organisation's formal submission and one individual's survey comment as equal votes, later findings become muddy and hard to defend.

What controls should be in place at intake?

Give every source a stable ID, record it in a source list, and agree the codebook and file rules before analysis starts. These controls are what let a reviewer move from a sentence in the draft back to the page it came from.

At minimum, an evidence-heavy report needs:

• stable source IDs that never change once assigned
• a live source list with version and status fields
• quote or excerpt locators: page, paragraph or timestamp
• a controlled codebook or theme list
• memo notes that record analytic decisions
• named owners for review and sign-off

Gale and colleagues' account of the Framework Method (BMC Medical Research Methodology, 2013) treats charting data into a matrix and writing analytic memos as stages of the analysis itself, not as admin done afterwards. Intake controls deserve the same status.

An illustrative row from a report-ready source list:

Why keep raw sources, synthesis and draft text separate?

Each layer answers a different question, and mixing them makes claims impossible to check. Keep raw material, extracted evidence, synthesis and draft text distinct, but linked by IDs.

A simple layer model:

1. Raw source base: documents, transcripts and datasets as received.
2. Evidence layer: extracted records or coded excerpts, each with a source ID and locator.
3. Synthesis layer: evidence tables, framework matrices or theme memos.
4. Draft report sections.
5. Review comments and final revisions.

Writers draft from the synthesis layer, not from scattered folders. Reviewers move back down the layers when they need to verify a point. When notes, quotes, interpretation and polished prose sit in one document, nobody can tell which sentence is evidence and which is the writer's reading of it.

What belongs in the middle layer between evidence and the draft?

The middle layer is whatever structured, source-linked material the writer drafts from: evidence tables by question, framework matrices for comparison across cases, theme memos, or finding sheets per section. Choose the form that matches the job.

• Evidence tables suit reports organised by question. A source-linked evidence table keeps each claim beside its source and caveat.
• Framework matrices suit many qualitative inputs compared across themes, groups or places.
• Theme memos suit reports where an interpretation needs explaining, such as why two themes were merged.
• Finding sheets suit large reports with several writers, one sheet per section.

When interviews or case studies feed the report, coding comes first. The method is set out in turning interviews and case studies into report-ready findings.

A useful middle-layer entry carries more than a summary. It shows which source or subgroup the point came from, whether it was widespread, mixed or exceptional, which caveat applies, which excerpt verifies it, and which report section it feeds. An illustrative entry:

How do findings, conclusions and recommendations differ?

Findings state what the evidence shows. Conclusions say what those findings mean for the report's purpose. Recommendations say what should happen next. The three connect, but they are not interchangeable.

UNICEF's adapted UNEG evaluation report standards (updated 2017) give useful tests. Conclusions should be substantiated by the evidence, refer back to the sources in the findings and not introduce new evidence. Recommendations should be logically derived from the findings or conclusions and refer back to sources of evidence. CDC's reporting guide (2013) adds that recommendations should address specific findings and be feasible, realistic, actionable and tailored to the people who will use them.

Between a conclusion and a recommendation sits the step many reports skip: framing the conclusion against the decision. Ask what matters most now, what remains uncertain, and what the decision-maker has to weigh. A longer summary does not do this; a clear statement of priorities and trade-offs does. Decision frameworks make the weighing explicit. GRADE's Evidence to Decision framework for health system and public health decisions asks, beyond the evidence of effect, about the priority of the problem, resource use, equity, acceptability and feasibility (Moberg et al., 2018). A recommendation that has not faced those questions is easy to reject in review.

An illustrative chain:

• Finding: Across district interviews, transport costs and unclear referral pathways were the barriers to follow-up visits raised most often. Urban respondents reported shorter delays.
• Conclusion: Missed follow-up is partly a problem of cost and navigation, not only of service capacity.
• Recommendation: Prioritise referral-pathway guidance and transport support in districts with long travel distances, then monitor missed follow-up rates.

When a report has many findings, build this chain in a table before writing the recommendations section. The findings-to-recommendations matrix is that table.

How should the report be ordered, and what goes in appendices?

Put context, purpose and method before findings, and findings before conclusions and recommendations. Keep the narrative in the main body and move detailed evidence tables, codebooks and instruments to appendices.

The UNICEF-adapted UNEG standards say context, purpose and methodology would normally precede findings, followed by conclusions, lessons learned and recommendations, and that the executive summary should be able to inform decision-making. NICE's guideline manual (last updated October 2025) presents evidence tables usually in an appendix, with evidence summaries carrying the synthesised narrative. CDC's reporting guide (2013) gives the same advice: put data tables and instruments in an appendix so interested readers can reach them without slowing everyone else.

The same evidence trail can sit behind several outputs:

Where does AI fit in a report writing workflow?

AI helps once the evidence base has IDs, clear fields and a working synthesis structure: for retrieval, first-pass grouping, summarising and drafting from checked material. On weak structure it produces confusion faster. People still own the method, the interpretation and the recommendations.

The UK Department for Transport's evaluation of its Consultation Analysis Tool (published December 2025) is a useful public example of the pattern. The tool generates themes from consultation responses and shows users an audit trail for each stage of theme generation. A human-in-the-loop review follows the automated step: reviewers read a sample of responses and refine the themes before the analysis goes further.

Human review should stay in charge of:

• setting the method and the codebook
• deciding what is grouped and what stays separate
• judging whether a pattern is strong enough to report
• interpreting thin or conflicting evidence
• writing conclusions in proportion to the evidence
• making recommendations that are feasible, ethical and fit the context

If AI is used anywhere in the chain, record which source set was in scope, what the AI step did, who checked the output and how the approved output fed the draft. In the UNICEF State of Palestine situation analysis, AI helped organise and retrieve coded evidence, while the findings and recommendations were written and signed off by the lead consultant and specialists.

Teams that produce this kind of report every year often need the structure in place before the next cycle starts. That is the point of a source-linked evidence system, built and handed over: writers and reviewers work from the same trail, and it stays with the team.

What should you check before a report goes to review?

Run a short logic check on the draft before it goes to reviewers. It catches the gaps reviewers would otherwise find for you.

Report logic checklist

[ ] The report states the decision or audience need it serves
[ ] Each section has a distinct purpose and report question
[ ] Every finding links to source IDs and locators
[ ] Conclusions add no evidence that is absent from the findings
[ ] Every recommendation responds to a named finding or conclusion
[ ] Caveats and subgroup limits are stated where the claim is made
[ ] Detailed evidence tables sit in an appendix and match the text
[ ] Review comments are tracked in one log with owner and status
[ ] Any AI step is recorded, with the name of the person who checked it
[ ] A named person has signed off the final version

Common mistakes

• Treating drafting as the first step. Without agreed questions, evidence scope and review route, the draft becomes the place where every problem surfaces at once.
• Using document comments as the review system. Comments are useful, but they are a poor record of evidence questions, owners and sign-off. Keep one review log.
• Checking sources after the writing is finished. Late source checks lead to rewrites. Check each claim against its source as it is drafted.
• Letting conclusions introduce new evidence. If a conclusion needs evidence that the findings do not present, the finding is missing, not the conclusion.
• Writing recommendations at the end, away from the findings. Recommendations written last tend to reflect the team's views rather than the evidence.
• Summarising more instead of framing the decision. A shorter restatement of the synthesis does not tell a decision-maker what to weigh.
• Moving caveats to the appendix. A limitation belongs next to the claim it limits.