How to build a findings-to-recommendations matrix

What is a findings-to-recommendations matrix?

A findings-to-recommendations matrix is a structured table that connects evidence to a finding, the finding to an implication, and the implication to a recommendation with an owner and a next action. It is the working table behind the recommendations section, not the list of recommendations itself.

The chain it records is:

evidence → finding → implication → recommendation → owner → action

A report can have sound findings and still end with weak recommendations. This usually happens when recommendations are written far from the evidence: the team analyses interviews, documents or programme data, writes findings, and then adds recommendations near the end as a separate section. Some come out too broad, some name no owner, and some do not respond to any finding at all. In a monitoring and evaluation cycle, this is often the step where a report loses the trail it built earlier; the wider process is set out in monitoring and evaluation reporting for South African NPOs.

A good matrix lets the team answer, for every recommendation:

• Which finding supports it?
• What evidence supports that finding?
• What problem, gap or risk does the finding point to?
• Who needs to act, and what should change?
• How practical is it?
• Has it been reviewed, and by whom?

Some teams call it a recommendations matrix or a recommendation support table.

When does a report need one?

A report needs a findings-to-recommendations matrix when it has to move beyond description and its recommendations will be questioned. The more findings, reviewers and audiences a report has, the more the matrix earns its place.

It is most useful when:

• the report has many findings that cut across themes, places, groups or source types
• recommendations need to be justified to a funder, technical reviewer or government counterpart
• recommendations need owners, priorities or timeframes
• the report feeds an action plan, management response or implementation process
• several reviewers will comment on the recommendations

Typical uses include evaluation reports, situation analyses, programme and strategy reviews, policy reviews, consultation reports and annual M&E reports. A short internal update with one author and a few clear findings usually does not need one.

How is it different from an evidence table, source register or action plan?

Each tool in an evidence workflow answers a different question. The matrix answers one the others do not: what should happen because of these findings?

The matrix draws on the layers below it. Each finding should point to rows in a source-linked evidence table, which in turn point to sources. Public consultations need a related but different record: a consultation response matrix shows each comment and the response to it, while this matrix shows how findings became recommendations. The action plan comes after both and is owned by whoever implements.

What fields should the matrix include?

Include enough fields to make each recommendation traceable and reviewable, and no columns nobody will maintain. A short report can work with eight fields; a donor-funded or public-sector evaluation usually needs more.

How do you get from findings to recommendations?

Start from what the evidence has shown, not from what the team would like to recommend. Each recommendation should be the end of a chain the matrix can show.

1. Confirm the report questions.
2. Review the evidence table or synthesis outputs.
3. Write clear finding statements.
4. Check the source support for each finding.
5. Write the implication of each finding.
6. Draft the recommendation.
7. Test whether the recommendation is specific, realistic and linked to the finding.

A funder or technical reviewer may ask why a recommendation appears, why it is high priority, or why a particular actor is named. The matrix lets the team answer before the report is final. It is one step in a wider report writing workflow, which covers scoping, synthesis and sign-off.

Why add an implication step?

The implication explains why a finding matters, and it is where most of the judgement happens. Without it, a recommendation reads as a leap from evidence to action.

The implication also helps the team choose the right recommendation, because one finding can point to different problems. Inconsistent referral pathways could reflect unclear guidance, weak coordination between districts, poor staff induction, missing service maps, fragmented case management or a lack of monitoring data. Each of those leads to a different recommendation. The matrix makes the choice, and the reason for it, visible.

How should evidence strength shape a recommendation?

Match the force of a recommendation to the strength of the evidence behind it. A finding supported across several source types can carry a direct recommendation; a point raised by one stakeholder usually cannot.

This scale is a working convention, not a formal grading system. Where a formal approach is expected, guideline bodies use established methods; NICE's guideline manual, for example, summarises qualitative evidence using GRADE-CERQual. Either way, the matrix is where the team records the strength judgement next to the recommendation it limits.

How do you make a recommendation specific enough to act on?

Name who should act, what should change, how soon, and what it depends on. A recommendation without an owner is hard to review and harder to implement.

Reporting standards set the same bar. UNICEF's adapted UNEG evaluation report standards (updated 2017) say recommendations should be logically derived from the findings or conclusions, refer back to sources of evidence, identify the target group for each recommendation and make priorities for action clear. CDC's guide to evaluation reporting (2013) says recommendations should address specific findings and be feasible, realistic, actionable and tailored to the people who will use them.

Decision frameworks add the questions reviewers will ask about feasibility. GRADE's Evidence to Decision framework for health system and public health decisions weighs, alongside the evidence of effect, the priority of the problem, resource use, equity, acceptability and feasibility (Moberg et al., 2018). A matrix with columns for feasibility and dependencies prompts the team to answer those questions early.

The report team may only be able to suggest an actor type. That is fine: the aim is a recommendation a reviewer can test.

Recommendation types help here too. Tagging each row as policy, programme, service delivery, data and monitoring, funding or research lets the right people review the right rows, and shows imbalance: if every recommendation is about coordination while the findings point to staffing and data problems, the gap is visible.

How does the matrix support review?

The matrix lets a reviewer move backwards from any recommendation to its sources: recommendation → implication → finding → evidence summary → source IDs → source material. That route is what keeps a challenged recommendation in the report.

Use it to check whether:

• every recommendation responds to an actual finding
• every finding that needs a response has one
• recommendations duplicate or contradict each other
• the named actor is the right one
• high-priority recommendations are justified
• contradictory or minority evidence was considered
• sensitive findings have been handled with care

Reviewing the logic in the matrix first keeps the debate out of the final draft. The same rows then feed decision notes, validation workshops and action planning. In the UNICEF State of Palestine situation analysis, a recommendation matrix linked to coded evidence sat behind the draft, so each recommendation carried the material that supports it.

Use the same source IDs, finding IDs, theme labels and report sections as the source register, evidence table and quote bank. The matrix should not be a separate island of reasoning.

Where can AI help, and where must people decide?

AI can help group, compare and check matrix rows once the evidence base is structured. It should not decide what the report recommends.

AI can help with:

• grouping findings by theme and spotting repeated implications
• suggesting draft wording for a recommendation
• flagging recommendations that are vague, overlapping or duplicated
• checking that every recommendation has a linked finding
• turning approved matrix rows into draft report paragraphs

AI should not:

• invent evidence or create recommendations without source support
• decide final priorities
• set aside minority or contradictory evidence
• make political, legal, clinical, safeguarding or technical judgements
• replace the evaluation lead, subject specialist or report owner

Give the tool structured inputs: findings, evidence summaries, source IDs, evidence strength and draft implications. Treat what comes back as a draft for review. People decide whether each recommendation is valid, feasible and safe to include.

What tools work for a findings-to-recommendations matrix?

A spreadsheet is enough for most reports. The structure matters more than the tool: stable IDs, clear fields, a review status and links back to the evidence.

Move to linked tables only when the matrix needs them: separate tables for sources, evidence extracts, findings, recommendations, report sections and review comments. For organisations that report on the same programme every year, a source-linked evidence system, built and handed over keeps those tables connected between cycles. For a single smaller report, a clean spreadsheet with consistent IDs is usually the better choice.

Findings-to-recommendations matrix template

Start with eight columns for a smaller report. The rows below are illustrative.

For a larger report, use the expanded column set and the pre-report check:

Expanded matrix columns
Finding ID | Theme | Finding | Evidence summary | Source IDs | Evidence strength |
Gap or risk | Implication | Draft recommendation | Recommendation type | Owner |
Timeframe | Feasibility | Dependencies | Priority | Report section | Review status | Notes

Check before recommendations go into the report
[ ] Every recommendation links to at least one finding
[ ] Every finding links to evidence or source IDs
[ ] Recommendations are specific enough to act on
[ ] No recommendation claims more than its evidence supports
[ ] High-priority recommendations have a stated justification
[ ] Each recommendation names a responsible actor or actor type
[ ] Timeframes are realistic and dependencies are visible
[ ] Contradictory evidence has been considered
[ ] Sensitive findings are handled with care
[ ] Duplicates are merged or removed
[ ] Report wording matches the matrix
[ ] Review status is up to date

If a recommendation cannot pass these checks, rewrite it, soften it, merge it or remove it.

Common mistakes

• Writing recommendations after the report is drafted. Build the matrix while findings are being developed, so recommendations stay close to the evidence.
• Recommendations with no finding behind them. Every recommendation should respond to a finding, gap, risk or implication. Anything else belongs in a separate strategy discussion.
• Recommending from one quote. A single vivid quote rarely justifies a major recommendation. Check the wider evidence first.
• Skipping the implication. Without it, the reader sees the action but not why it follows.
• Vague wording. "Improve coordination", "strengthen capacity" and "increase awareness" say nothing about who acts or what changes.
• No owner. A recommendation nobody owns is hard to review and easy to ignore.
• Ignoring feasibility and contradictory evidence. Some evidence-based recommendations are unrealistic in context, and mixed evidence calls for a qualified recommendation.
• Too many recommendations. A long list is harder to act on. Use the matrix to group, merge and prioritise.
• Dropping the matrix once the draft exists. Keep it live through review comments, revisions and final checks.