> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mavera.io/llms.txt
> Use this file to discover all available pages before exploring further.
# Bulk Response Analysis
> Pull SurveyMonkey bulk responses and analyze with Mavera for statistical patterns, sentiment trends, and recommendations
### Scenario
Your SurveyMonkey survey collected 1,200 responses across 20 questions — multiple choice, ratings, open-ended. Manually analyzing this would take days. This job pulls all responses via the bulk endpoint, structures them for analysis, then sends the entire dataset to Mave Agent with the instruction: "Analyze responses. Identify statistical patterns, sentiment trends, and actionable recommendations." The result is an AI-generated research report that would normally require a dedicated analyst.
**Flow:** SurveyMonkey `GET /v3/surveys/{id}/responses/bulk` → Aggregate and structure → Mave `POST /api/v1/mave/chat`: "Analyze responses. Identify statistical patterns, sentiment trends, recommendations." → Research report
### Architecture
```mermaid theme={"dark"}
flowchart LR
A["GET /v3/surveys/{id}/responses/bulk"] --> B["Parse all answer types"] --> C["Build statistical summary"] --> D["Mave Agent: Analyze patterns"] --> E["AI-generated research report"]
```
### Code
```python Python theme={"dark"}
import os, json, requests, time
from collections import Counter, defaultdict
SM = os.environ["SURVEYMONKEY_TOKEN"]
MV = os.environ["MAVERA_API_KEY"]
SM_BASE = "https://api.surveymonkey.com/v3"
MB = "https://app.mavera.io/api/v1"
SM_H = {"Authorization": f"Bearer {SM}", "Content-Type": "application/json"}
MV_H = {"Authorization": f"Bearer {MV}", "Content-Type": "application/json"}
SURVEY_ID = os.environ.get("SURVEY_ID", "your_survey_id")
# 1. Get survey structure
survey = requests.get(f"{SM_BASE}/surveys/{SURVEY_ID}/details",
headers=SM_H).json()
print(f"Survey: {survey.get('title', 'Untitled')} ({survey.get('response_count', 0)} responses)")
# Build question/answer lookup
questions = {}
choice_labels = {}
for page in survey.get("pages", []):
for q in page.get("questions", []):
qid = q["id"]
questions[qid] = {
"title": q.get("headings", [{}])[0].get("heading", ""),
"type": q.get("family", ""),
"subtype": q.get("subtype", ""),
}
for row in q.get("answers", {}).get("rows", []):
choice_labels[row["id"]] = row.get("text", "")
for choice in q.get("answers", {}).get("choices", []):
choice_labels[choice["id"]] = choice.get("text", "")
for col in q.get("answers", {}).get("columns", []):
choice_labels[col["id"]] = col.get("text", col.get("label", ""))
# 2. Pull all responses (paginated)
all_responses = []
page_num = 1
while True:
r = requests.get(f"{SM_BASE}/surveys/{SURVEY_ID}/responses/bulk",
headers=SM_H, params={"page": page_num, "per_page": 100})
if r.status_code == 429:
retry = int(r.headers.get("X-Ratelimit-App-Global-Day-Reset", 60))
print(f"Rate limited — daily limit. Retry in {retry}s")
time.sleep(min(retry, 60))
continue
r.raise_for_status()
data = r.json()
batch = data.get("data", [])
all_responses.extend(batch)
total = data.get("total", 0)
if len(all_responses) >= total or not batch:
break
page_num += 1
time.sleep(0.6)
print(f"Fetched {len(all_responses)} responses")
# 3. Aggregate answers by question
qa_data = defaultdict(list)
for resp in all_responses:
for page_data in resp.get("pages", []):
for q_data in page_data.get("questions", []):
qid = q_data["id"]
for ans in q_data.get("answers", []):
if "text" in ans:
qa_data[qid].append(ans["text"])
elif "choice_id" in ans:
label = choice_labels.get(ans["choice_id"], ans["choice_id"])
qa_data[qid].append(label)
if "row_id" in ans:
row = choice_labels.get(ans["row_id"], "")
qa_data[qid][-1] = f"{row}: {label}"
# 4. Build analysis summary
summary_parts = []
for qid, answers in qa_data.items():
q_info = questions.get(qid, {})
title = q_info.get("title", qid)
q_type = q_info.get("type", "")
if q_type in ("single_choice", "multiple_choice", "matrix"):
counts = Counter(answers).most_common(10)
dist = ", ".join(f"{label}: {count} ({count/len(answers)*100:.0f}%)"
for label, count in counts)
summary_parts.append(f"**{title}** (n={len(answers)})\n {dist}")
elif q_type == "open_ended":
sample = "; ".join(answers[:15])[:500]
summary_parts.append(f"**{title}** (n={len(answers)}, open-ended)\n Samples: {sample}")
else:
try:
nums = [float(a) for a in answers if a.replace(".", "").replace("-", "").isdigit()]
if nums:
avg = sum(nums) / len(nums)
summary_parts.append(f"**{title}** (n={len(nums)}, avg={avg:.1f})")
except ValueError:
summary_parts.append(f"**{title}** (n={len(answers)})")
summary = "\n\n".join(summary_parts)
# 5. Mave analysis
analysis = requests.post(f"{MB}/mave/chat", headers=MV_H, json={
"message": f"""Analyze {len(all_responses)} survey responses from "{survey.get('title', 'Survey')}".
QUESTION-BY-QUESTION DATA:
{summary}
Tasks:
1) Statistical patterns: Correlations between questions, unexpected distributions
2) Sentiment trends: Overall and per-question sentiment analysis
3) Key findings: Top 5 insights ranked by impact
4) Segment differences: Any visible subgroups in the data
5) Red flags: Responses that indicate problems or risks
6) Recommendations: 5 actionable next steps based on the data
7) Suggested follow-up questions: What should you ask next?"""
}).json()
print(f"\n{'='*60}")
print(f"ANALYSIS: {survey.get('title', 'Survey')} ({len(all_responses)} responses)")
print(f"{'='*60}")
print(analysis.get("content", "")[:3000])
print(f"\nSources: {len(analysis.get('sources', []))}")
```
```javascript JavaScript theme={"dark"}
const SM = process.env.SURVEYMONKEY_TOKEN;
const MV = process.env.MAVERA_API_KEY;
const SM_BASE = "https://api.surveymonkey.com/v3";
const MB = "https://app.mavera.io/api/v1";
const smH = { Authorization: `Bearer ${SM}`, "Content-Type": "application/json" };
const mvH = { Authorization: `Bearer ${MV}`, "Content-Type": "application/json" };
const SURVEY_ID = process.env.SURVEY_ID || "your_survey_id";
// 1. Survey structure
const survey = await fetch(`${SM_BASE}/surveys/${SURVEY_ID}/details`,
{ headers: smH }).then(r => r.json());
console.log(`Survey: ${survey.title} (${survey.response_count} responses)`);
const questions = {};
const choiceLabels = {};
for (const page of survey.pages || []) {
for (const q of page.questions || []) {
questions[q.id] = { title: q.headings?.[0]?.heading || "", type: q.family || "",
subtype: q.subtype || "" };
for (const r of q.answers?.rows || []) choiceLabels[r.id] = r.text || "";
for (const c of q.answers?.choices || []) choiceLabels[c.id] = c.text || "";
for (const c of q.answers?.columns || []) choiceLabels[c.id] = c.text || c.label || "";
}
}
// 2. Pull responses
const allResponses = [];
let pageNum = 1;
while (true) {
let res = await fetch(
`${SM_BASE}/surveys/${SURVEY_ID}/responses/bulk?page=${pageNum}&per_page=100`,
{ headers: smH });
if (res.status === 429) {
const retry = parseInt(res.headers.get("X-Ratelimit-App-Global-Day-Reset") || "60");
console.log(`Rate limited. Waiting ${Math.min(retry, 60)}s`);
await new Promise(r => setTimeout(r, Math.min(retry, 60) * 1000));
continue;
}
const data = await res.json();
allResponses.push(...(data.data || []));
if (allResponses.length >= (data.total || 0) || !(data.data || []).length) break;
pageNum++;
await new Promise(r => setTimeout(r, 600));
}
console.log(`Fetched ${allResponses.length} responses`);
// 3. Aggregate
const qaData = {};
for (const resp of allResponses) {
for (const pg of resp.pages || []) {
for (const q of pg.questions || []) {
(qaData[q.id] ??= []);
for (const ans of q.answers || []) {
if (ans.text) qaData[q.id].push(ans.text);
else if (ans.choice_id) {
let label = choiceLabels[ans.choice_id] || ans.choice_id;
if (ans.row_id) label = `${choiceLabels[ans.row_id] || ""}: ${label}`;
qaData[q.id].push(label);
}
}
}
}
}
// 4. Build summary
const summaryParts = [];
for (const [qid, answers] of Object.entries(qaData)) {
const q = questions[qid] || {};
const title = q.title || qid;
if (["single_choice", "multiple_choice", "matrix"].includes(q.type)) {
const counts = {};
answers.forEach(a => { counts[a] = (counts[a] || 0) + 1; });
const dist = Object.entries(counts).sort(([, a], [, b]) => b - a).slice(0, 10)
.map(([l, c]) => `${l}: ${c} (${(c / answers.length * 100).toFixed(0)}%)`).join(", ");
summaryParts.push(`**${title}** (n=${answers.length})\n ${dist}`);
} else if (q.type === "open_ended") {
summaryParts.push(`**${title}** (n=${answers.length}, open-ended)\n Samples: ${answers.slice(0, 15).join("; ").slice(0, 500)}`);
} else {
const nums = answers.map(Number).filter(n => !isNaN(n));
if (nums.length) summaryParts.push(`**${title}** (n=${nums.length}, avg=${(nums.reduce((s, n) => s + n, 0) / nums.length).toFixed(1)})`);
else summaryParts.push(`**${title}** (n=${answers.length})`);
}
}
// 5. Mave analysis
const analysis = await fetch(`${MB}/mave/chat`, { method: "POST", headers: mvH,
body: JSON.stringify({
message: `Analyze ${allResponses.length} responses from "${survey.title}".
QUESTION DATA:
${summaryParts.join("\n\n")}
Tasks: 1) Statistical patterns 2) Sentiment trends 3) Top 5 insights 4) Segment differences 5) Red flags 6) 5 recommendations 7) Follow-up questions`,
}),
}).then(r => r.json());
console.log(`\n${"=".repeat(60)}`);
console.log(`ANALYSIS: ${survey.title} (${allResponses.length} responses)`);
console.log(`${"=".repeat(60)}`);
console.log((analysis.content || "").slice(0, 3000));
```
### Example Output
```text theme={"dark"}
============================================================
ANALYSIS: Q1 Customer Satisfaction Survey (1,247 responses)
============================================================
## Key Findings
### 1. Bimodal Satisfaction (Critical)
Overall satisfaction shows two peaks: 8-9/10 (43%) and 3-4/10 (22%).
The middle is thin — customers either love you or are frustrated.
No "passive middle" suggests polarizing experiences.
### 2. Onboarding is the Breakpoint
92% of dissatisfied respondents (≤5/10) cite onboarding as their
primary pain. Satisfaction jumps from 4.1 → 8.3 after 30-day mark.
Intervention window: days 3-14.
### 3. Feature Satisfaction ≠ Retention Intent
Reporting module scores 8.2/10 in satisfaction but is cited by 38%
of "likely to churn" respondents as "not enough." High expectations,
not low quality, drives churn risk.
### 4. Support Channel Preference Shift
Under-35 respondents (34% of base) prefer chat (67%) over email (12%).
Over-45 prefer email (58%) over chat (21%). Your support channel
allocation doesn't match.
### 5. Open-Ended Red Flag
"We're evaluating alternatives" appears in 14% of detractor comments.
Cross-reference with CRM to identify at-risk accounts.
## Recommendations
1. Launch 14-day onboarding drip with check-in at day 3, 7, 14
2. Shift 30% of support capacity from email to chat
3. Create "power user" track for reporting module
4. Run churn risk analysis on detractor accounts
5. Deploy NPS follow-up survey targeting the "3-4" cohort
Sources: 0
```
### Error Handling
Private apps are capped at 500 requests per day. The bulk endpoint returns 100 responses per page, so a 2,000-response survey needs 20 calls. Cache responses locally after the first pull.
Matrix questions nest rows and columns. The code concatenates "Row: Column" labels. For complex matrices (10×10), the summary may be verbose — truncate to top 5 combinations.
If `choice_labels` doesn't find a match, the raw choice ID is used. This happens with custom "Other" options. Pre-populate the lookup from the survey detail endpoint.
Surveys with 20+ questions and 1,000+ responses generate large summaries. The code caps open-ended samples at 15 and text at 500 chars. For very large surveys, analyze in sections.