Opti'General information
The project, the organisation, the donor, the area, the period: these lines go on the report cover, copied as they are, never reworded.
📊 From dataset to report, in a single chain
Generate your MEAL reports
from your own data
Import your survey data, from KoboToolbox or an Excel file: Opti' checks its quality, analyses it with the disaggregations the sector uses, then writes a narrative report in which every figure is calculated, never invented. You validate its readings before it writes, and you export it to Word, PowerPoint, PDF or HTML.
✓ Free account, no card required ✓ Figures calculated, never invented ✓ Word, PowerPoint, PDF, HTML
A report generated live, from the data of a KoboToolbox collection, during the webinar “SAFE AI and the AI Act: from theory to practice” (MEAL Afrique, 26 September 2026, in French). The video starts at the demonstration.
From data to report: nine steps, one chain
The Opti Gemba Report tool guides you in the order a MEAL analyst works. Every step keeps a record of what it did, so the final report can be reviewed and defended.
MEAL framework
Your indicators, matched to the questions that measure them. The report tracks their measured value, their baseline and their target.
Data
A data collection run in Opti', your KoboToolbox submissions through the direct connection, or an .xlsx dataset produced by another tool.
Quality
Missing required answers, broken constraints, interviews that are too short, duplicates, an enumerator to look at: counts, with no AI, listing the submissions concerned.
Cleaning
Corrections go into a separate, reversible layer. The raw data is never overwritten, and a log says what was changed.
Dashboard
Frequency tables and cross-tabulations, two-level disaggregation, counts shown under every table, seven chart types.
Analysis plan
Opti' proposes the sections and the questions; you tick, rename and move them. A matrix crosses your indicators with your disaggregations.
Writing
Before writing, Opti' puts its readings and the figures behind them to you. Then it writes from a register of checked values.
Exports
Word to rework, PowerPoint with native charts, PDF and HTML to share.
Analyse your KoboToolbox data
Connect your KoboToolbox account once: your projects' submissions arrive in Opti' for analysis, an interrupted import resumes where it stopped, and corrections made in Kobo are picked up. Your organisation does not allow API tokens? The Kobo .xlsx export can be imported too, like a dataset produced by ODK, SurveyCTO or a spreadsheet. The form does not need to have been built in Opti'.
The steps are detailed in connecting KoboToolbox to Opti', and the platform itself in what is KoboToolbox.
Data quality and cleaning, before any analysis
A report is only as good as its data. Opti's quality check guesses nothing: every finding is a count, with its figures and the list of submissions concerned, so you can go and look at them. It never says an answer is wrong; it says the answer does not follow the rule written in the form.
Missing answers
A missing required answer is only counted if the question was asked: a question hidden by skip logic is not an omission.
Constraints and types
The constraints and types of your form are replayed on the dataset. A dataset imported from another tool had never been checked against them.
Durations and enumerators
Interviews under a third of the median duration, inconsistent timestamps, an enumerator with more of them than the others: the rule is written in the finding.
Reversible cleaning
Variants of an open answer are grouped, the list of changes is reviewed before it is applied, and the raw data stays intact.
To go further on the five dimensions of quality and the DQA audit, see MEAL data quality.
Dashboard, disaggregation and charts
Each question gets the analysis mode it needs: counts, multiple choice, mean, or readable classes for a spread-out value. Disaggregation goes up to two levels, by sex, age, location, beneficiary status, type of assistance, disability or population group, and the number analysed is shown under every table. Charts, seven types to choose from, go to PowerPoint as native, editable charts; the whole analysis exports to Excel, with live formulas that reference the dataset.
You can also ask about your data in plain language, “how many households per village”, and the figure comes from the same calculation as the tables, never from an estimate by the model.
Findings, recommendations and narrative report
Figures are calculated, text is written, and the two never mix. That is what makes a generated report trustworthy.
A register of figures
Every calculated value gets an identifier; the text cites it, and a final check matches every number in the text to its source: confirmed, corrected or flagged.
Your readings first
Before writing, Opti' shows you what it is about to state, the figures behind it and the possible explanations. You validate, correct or discard, and add the context only the field knows.
The structure of an evaluation report
Executive summary and key figures, methodology and limitations, findings by theme, logframe indicator tracking, recommendations, full annexes.
Four tones
Operational for programme teams, analytical for an evaluation or a donor, synthetic for a decision-maker, or a factsheet to read in two minutes.
Every report ends with a human review block, and its cover points to the writing method, public, limitations included.
Word, PowerPoint, PDF or HTML
| Format | Use |
|---|---|
| Word | Rework the text, fold it into the annual report: real headings, the cover and the logo, the charts and the indicator table. |
| PowerPoint | Present to teams or to the donor: native charts that you edit in PowerPoint. |
| Share a fixed version: the cover and the contents on their own page, no figure split across two pages. | |
| HTML | Share a link or a file that opens anywhere, with figures in three views: chart, table, reading. |
Your data: what the model receives
To write, the model receives the question labels, the tables already calculated and, to sort the questions of the plan, up to four example values per column. Never the dataset row by row. The quality check uses no AI; grouping open answers, if you run it during cleaning, sends the model the wordings to match. A dataset with columns that identify people should be anonymised before generation.
Sensitive content is encrypted with AES-256-GCM, and no collected data trains a model. The details, the subprocessors and your rights are on ethical AI and data protection; a SAFE AI transparency sheet pre-filled for the report is on the method page.
Opti', Excel, Power BI or ChatGPT: which one for a MEAL report?
All four help produce a report. What changes is what is left to do by hand, and what is left to check.
| Opti' | Excel by hand | Power BI | ChatGPT | |
|---|---|---|---|---|
| The figures | Calculated on the whole dataset, matched to a register | Exact, table by table | Exact, once the model is built | Calculated on what you paste in, unchecked |
| Disaggregation | Two levels, counts shown | Pivot tables to build | To configure | If asked, with no guarantee |
| Quality check | Deterministic counts, submissions listed | Formulas to write | To model | No |
| The written report | Yes, after you validate the readings | No | No, dashboards | Yes, with no check of the figures |
| KoboToolbox data | Direct connection | Export to import | Connector to configure | File to paste |
| Exports | Word, PowerPoint, PDF, HTML | Workbook | Dashboards, PDF | Text to format |
Baseline, endline, PDM, DQA: your survey reports
Baseline and endline
The starting situation, then the same measure at the end of the project: running a baseline and an endline survey.
Post-distribution monitoring
Receipt, quality, satisfaction, complaints, disaggregated by gender and status: cash PDM, section by section.
Data quality assessment
Gaps between sources, monthly reports and indicators: DQA in a humanitarian project.
Focus groups and interviews
Qualitative summaries go into the report, next to the figures: running and analysing a focus group.
Seeing is free, taking it with you is paid
With a free account, you generate a report and read it in full, before you pay anything. To take it with you (Word, PowerPoint, PDF or HTML), two options:
- As a one-off: a single payment, by card, bank transfer or mobile money.
- With a plan Starter, Pro or Premium: everything you produce is yours to take, forms, analyses and reports.
Before the report, the form
Your survey has not started yet? The XLSForm generator for KoboToolbox writes the form from a description, or converts the one you have in Word, PDF or Excel. You can try it without an account, and the report then picks up its questions and indicators.
Frequently asked questions
What is a MEAL report generator?
A tool that starts from your survey data and produces the report: quality check, disaggregated analysis, charts, findings, recommendations and written text. Opti's follows the workflow of a MEAL analyst, from the data to the final document, and has you validate its readings before it writes.
Can the AI invent figures in the report?
No. Figures are calculated on your answers, not produced by the language model. The calculation fills a register where every value carries an identifier; the text cites these identifiers, and every number in the final text is matched to the register: confirmed, corrected, or flagged as unverifiable. You are given the count of all three at the end. See also: the writing method, check by check.
What data can I import?
The submissions of your KoboToolbox projects, through the direct connection or their .xlsx export, a data collection run in Opti', or an .xlsx dataset produced by another tool (ODK, SurveyCTO, a spreadsheet). A dataset collected elsewhere is enough: the form does not need to have been built in Opti'. See also: connecting KoboToolbox to Opti'.
Is my data sent to an AI model?
To write, the model receives the question labels, the tables already calculated and, to sort the questions of the plan, up to four example values per column: never the dataset row by row. The quality check uses no AI. A dataset with columns that identify people should be anonymised before generation. Sensitive content is encrypted with AES-256-GCM, and no collected data trains a model. See also: ethical AI and data protection.
Which formats does the report export to?
Word, to rework it; PowerPoint, with native editable charts; PDF and HTML, to share it. The analysis itself exports too, to Excel with live formulas, or to PowerPoint, one slide per table.
Can the report go to the donor as it is?
It must be reviewed first. Every exported report ends with a validation block: the person who reviewed it attests to it, with their name and role, and the cover states that it was written with the help of a model, with a link to the method. A review is often done by several people: as many reviewers as needed can be listed.
How long does a generation take?
Several minutes: a report runs several dozen analyses, and a progress panel shows the findings as they come. An interrupted generation, lost connection or closed tab, resumes where it stopped, without redoing or paying again for the analyses already done.
How much does a report cost?
With a free account, you generate a report and read it in full, before you pay anything. To export it, you buy it as a one-off, or you choose a Starter, Pro or Premium plan, with which everything you produce is yours to take. Prices are on the pricing page. See also: pricing page.
How is it different from ChatGPT?
A general assistant writes well, but it calculates on what you paste in, and a number can be plausible and wrong without anything flagging it. Opti' calculates on the whole dataset, keeps what is calculated apart from what is written, puts its readings to you before writing, and matches every figure in the text to its source. See also: AI for MEAL.
Your next report starts with your data
Create your free account: the Report tool opens, ready for your dataset.
Generate my first report →