Consultation Summary (Remissammanställning)
Det här innehållet är inte tillgängligt på ditt språk än.
Below are two visualizations of the same task — summarizing consultation responses (remissvar) for a Swedish government inquiry (betänkande). The first shows how it is done manually. The second shows how the same process runs using an Intric assistant with an MCP integration connected to the Swedish government’s remiss portal.
Before Intric: Manual Consultation Review
Section titled “Before Intric: Manual Consultation Review”A civil servant downloads each consultation response individually, reads through every submission, manually categorizes opinions by proposal, and then compiles a continuous prose summary in a Word template. For a major inquiry this can involve 50–150 respondents and take several days.
Step 1 — Civil servant downloads the inquiry
Section titled “Step 1 — Civil servant downloads the inquiry”The process begins when a civil servant downloads the relevant government inquiry (betänkande) as a Word document.
What happens:
- The civil servant locates the inquiry on the government’s website or internal drive.
- The Word file contains the proposals (förslag) that will form the structure for the summary.
Step 2 — Civil servant navigates the remiss portal
Section titled “Step 2 — Civil servant navigates the remiss portal”The civil servant opens the government’s remiss portal to locate all submitted responses.
What happens:
- The civil servant searches for the inquiry by reference number or title.
- They identify the list of submitting organisations (remissinstanser).
Step 3 — Individual responses downloaded from the portal
Section titled “Step 3 — Individual responses downloaded from the portal”Each consultation response is downloaded separately as a PDF or Word file.
What happens:
- The civil servant opens each respondent’s entry and saves the file locally.
- For large inquiries this step alone can take several hours.
Step 4 — Civil servant reads and annotates each response
Section titled “Step 4 — Civil servant reads and annotates each response”The civil servant opens each file in their Office Suite and reads through the submission.
What happens:
- Relevant opinions are highlighted or copied into a working notes document.
- The civil servant tracks which respondent said what and which proposal it relates to.
Step 5 — Civil servant categorizes opinions by proposal
Section titled “Step 5 — Civil servant categorizes opinions by proposal”After reading all responses, the civil servant organizes the collected notes.
What happens:
- Notes are grouped under each proposal from the inquiry.
- The civil servant checks that no respondent or opinion has been missed.
Step 6 — Civil servant drafts the summary text
Section titled “Step 6 — Civil servant drafts the summary text”For each proposal, the civil servant writes a continuous prose paragraph that weaves together all positions.
What happens:
- Every named respondent and their position must appear explicitly.
- Minority views and specific quotes are included where relevant.
- The civil servant switches back and forth between the notes and the template.
Step 7 — Summary finalized and delivered
Section titled “Step 7 — Summary finalized and delivered”The completed summary document is reviewed, formatted, and delivered to management or published.
What happens:
- The civil servant reads through the full document to check for gaps.
- The finalized summary is sent to decision-makers or posted in the case management system.
With Intric: Automated Consultation Summary
Section titled “With Intric: Automated Consultation Summary”The civil servant uploads the Word file and starts the assistant. The assistant uses an MCP integration to connect directly to the government’s remiss portal, fetches all responses in batches, maps every opinion to the correct proposal, and delivers a fully attributed prose summary — including an opening paragraph and an “Other” section for cross-cutting comments.
Step 1 — Civil servant starts the plan by providing the ID of the proposal
Section titled “Step 1 — Civil servant starts the plan by providing the ID of the proposal”The civil servant opens the Intric assistant, provides the SOU number or reference ID for the government inquiry, and sends the instruction to begin the consultation summary workflow.
What happens:
- The civil servant pastes the inquiry ID (e.g. SOU 2023:72) and optionally attaches the Word file containing the proposals.
- The assistant acknowledges and begins executing the plan autonomously.
No manual downloading of responses needed. The assistant handles all subsequent steps via the MCP integration.
Step 2 — Language model identifies and maps proposals
Section titled “Step 2 — Language model identifies and maps proposals”The assistant reads the Word file and extracts every proposal from the inquiry.
What happens:
- The language model parses the document and produces a structured list of proposals.
- This structure becomes the skeleton for the final summary.
Output: A numbered proposal map that all subsequent steps use as a reference.
Step 3 — Intric requests batch 1 from the MCP server
Section titled “Step 3 — Intric requests batch 1 from the MCP server”The assistant calls the MCP integration to fetch the first batch of consultation responses (respondents 1–25) directly from the government remiss portal.
What happens:
- Intric sends an authenticated request to the MCP server.
- The MCP server queries the portal on behalf of the assistant and returns the responses.
No manual downloading. The integration retrieves responses programmatically.
Step 4 — Language model processes batch and maps opinions to proposals
Section titled “Step 4 — Language model processes batch and maps opinions to proposals”The language model reads batch 1 and maps each opinion to the corresponding proposal from Step 2.
What happens:
- Every opinion is attributed to the named respondent.
- Positions (support, opposition, reservation) are recorded for each proposal.
- Specific quotes that add value are flagged for inclusion.
Quality check: The model notes any opinions that do not fit an existing proposal, marking them for the “Other” section.
Step 5 — Intric fetches remaining batches via the MCP server
Section titled “Step 5 — Intric fetches remaining batches via the MCP server”The assistant repeats Step 3 and Step 4 for each remaining batch of respondents (26–50, 51–75, etc.) until all responses have been processed.
What happens:
- Each batch is fetched from the MCP server and processed by the language model in sequence.
- The opinion map grows with each batch.
Traceability: Every respondent name is recorded alongside their position throughout this phase.
Step 6 — Language model merges batches and writes the full summary
Section titled “Step 6 — Language model merges batches and writes the full summary”With all batches processed, the language model merges the full opinion map into the final consultation summary.
What happens:
- Under each proposal, a continuous prose paragraph is written that names every respondent and their position.
- Groups with the same view are summarized together, with all members named: “Gävle, Malmö, Stockholm, and Salem support the proposal to…”
- An opening summary paragraph is generated, followed by the per-proposal sections and an “Other” section.
Format: The output follows the established remissammanställning template exactly.
Step 7 — Intric performs a final completeness check
Section titled “Step 7 — Intric performs a final completeness check”Before delivering the summary, the assistant verifies that no respondent or opinion has been omitted.
What happens:
- The assistant cross-checks the list of all respondents fetched from the MCP server against those named in the summary.
- Any gaps are identified and filled before the response is returned to the civil servant.
This step is mandatory in the assistant prompt and cannot be skipped.
Step 8 — Summary delivered to the civil servant
Section titled “Step 8 — Summary delivered to the civil servant”The completed, fully attributed consultation summary is returned to the civil servant in the conversation window.
What happens:
- The civil servant receives a structured document ready for review.
- Minor adjustments can be made directly in the conversation before final export.
Result: A complete remissammanställning with an opening paragraph, per-proposal sections, and an “Other” section — all with full attribution.
Before vs. After
Section titled “Before vs. After”| Task | Manual | Time spent | Intric | Time spent |
|---|---|---|---|---|
| Download inquiry | Civil servant locates and downloads the betänkande as a Word file | ~10 min | Civil servant uploads the Word file to the assistant | ~2 min |
| Collect responses | Civil servant navigates portal and downloads each response individually | ~2–4 hrs | Automated via MCP integration | — |
| Read and categorize opinions | Civil servant reads each submission and manually maps opinions to proposals | ~2–3 days | Automated | — |
| Write summary text | Civil servant drafts continuous prose with full attribution for each proposal | ~1–2 days | Automated | — |
| Completeness check | Civil servant manually cross-checks that no respondent has been missed | ~2–3 hrs | Automated | — |
| Review and approve summary | Review and approve the final summary | ~1–2 hrs | Review and approve the final summary | ~1–2 hrs |
| Total time | Full process from start to finish | ~3–5 days | Staff time only | ~1–2 hrs |