To analyse public submissions, log every submission, read a sample to build a list of themes, code each submission against those themes, check that two people code alike, and then decide on each proposal or group of proposals. The Supreme Court accepts grouping similar views into themes, as long as those who made submissions can understand that their input was considered. The 2026 Finance Bill drew memoranda from over 100,000 respondents, so this is now a real problem for committees and counties.
Why does analysis matter?
Collecting views proves little, because the test is what you did with them. Nyabuto found no evidence that public inputs were considered. The 2016 guidelines ask officials to carefully collect and collate input in a systematic and retrievable manner, and devote a section to collection and analysis.
Step 1: log everything
Give each submission a reference number on the day it arrives. Record the date, the channel, who sent it, the ward, and whether it is an individual or an organisation. Keep the original exactly as it arrived, whether that's a stamped letter, an email or notes from a forum. Nothing should be added or removed later.
Step 2: separate the points
One memorandum may make twenty points, and a forum note may hold many speakers. Break them down, so that each proposal can get its own decision. Link each point to its source, so that anyone can go back to the original.
Step 3: build the themes
Read a sample of about a hundred submissions across channels, and draft a short list of themes, with a clear definition of each. Your themes should follow the document, such as clause numbers, plus a few for issues that arise. Add a code for "other", and review it regularly, since new themes often hide there.
Step 4: code consistently
Have two people code the same sample separately, then compare. If they disagree often, your definitions are too loose. Agreement can be measured, and the measure can go in your report. Train the team until the definitions hold.
Step 5: count carefully
Counts show how widely a view is held, but a count isn't a vote. A forum attended by one group, or a campaign of identical emails, can inflate a view. Report how many people held each view and, separately, how many distinct arguments you found. Don't weight forum views as if they were a sample, and don't describe them as what "the public" thinks.
Step 6: decide on each point
Mark each proposal as accepted, partly accepted, rejected or out of scope, with a reason and the name of the person responsible. Though the Supreme Court found no express duty to give reasons to each person, it expects reasonable measures, and a clear reason is the simplest proof. A reason should be specific: "noted" says nothing.
Can AI help?
AI can suggest themes and find passages quickly, and can help with very large volumes. It can also make mistakes, drop minority views and invent patterns. Treat it as a draft: a person should confirm each extract before it enters the record, and the record should say how AI was used. Never let a model decide on a citizen's view.
How should you handle languages?
Submissions may arrive in Kiswahili, Sheng or a local language. Keep the original, and put a translation beside it. Check a sample of translations, because tone and meaning can shift. A view recorded only in translation can be challenged.
How should you handle forum notes?
Rapporteurs' notes shorten what people say, often in the officials' favour. Where you can, record audio, with consent, and check notes against it. Mark which words are the speaker's and which are a summary.
How do you report the results?
Use a table of themes, with counts, example quotes and the decision on each. The report sample gives a format. State your method in a short paragraph: how many submissions, who coded them, and how you checked.
How long does it take?
That depends on volume, and on deadlines set by law. Plan the analysis as part of the exercise, with named people. It's often the step that gets squeezed, so set time aside before the decision is due.
What should you keep?
Keep the log, the coding, the versions of the theme list, and the decisions. If the exercise is challenged, those files show how you got from thousands of views to a final document.
A worked example
Take an invented county consultation with 3,000 submissions. Of these, 2,200 are identical emails, 640 are forum comments and 160 are memoranda. The team logs them all, and groups the emails by exact wording into four groups. It breaks the memoranda into 900 separate points, and reads 100 items to build 14 themes.
Two officers code the same 100 items, and agree on 74. They tighten three definitions, and try again, reaching 91. They then code everything, and report that 2,840 people raised 6 main arguments on fees, and that 120 people raised access for persons with disabilities. They decide on each group, and say why. The small group's view leads to a change, which a simple count would have missed.
Read next
- Public participation system for Kenya: the TI pack page for this work
- Public participation software: all the pages in this section
- Handling identical submissions
- Deciding on every view
- Public participation report sample
- Takwimu Intelligence (TI) home page
- Public participation in Kenya: the guide to this library
- The TI platform and how we check our numbers
- Code open answers and measure agreement
- Build a codebook
Where TI fits
On Takwimu Intelligence (TI), views arrive from every channel into one list, passages drawn out by software count only once a person confirms them, and two coders' agreement is measured.
To see it on your own consultation, book a walk-through.