Market Research

Traditional consumer research in financial services is expensive, slow, and often small in sample. Dialogue gives research and insight teams programmatic access to thousands of real, unprompted consumer opinions, structured for analysis. You can test hypotheses, explore emotional territories, and produce research reports grounded in genuine consumer voice — in hours, not weeks.

This guide walks through producing a consumer sentiment report on pension saving.

Step 1: Establish the top-line numbers

Start with aggregate statistics to set the quantitative context for your report.

$
GET /api/v1/stats

From the response, extract:

• Total corpus size (baseline sample)
• Emotional register distribution for pension_sipp comments
• Life stage distribution
• Which behavioural signals appear most frequently

This gives you the “headline findings” section of any research report — what proportion of the audience expresses anxiety versus pride versus cynicism, and which life stages are most represented.

Step 2: Segment the data by hypothesis

Quantitative breakdowns only tell you what. Segment filtering tells you how different groups differ from the overall picture.

Anxiety segment — what are they actually afraid of?

$
GET /api/v1/comments?product=pension_sipp&emotional_register=anxiety_fear&min_likes=5&limit=50

Cynicism segment — what’s driving distrust?

$
GET /api/v1/comments?product=pension_sipp&emotional_register=anger_cynicism&min_likes=5&limit=50

The relief segment — what does resolution look like?

$
GET /api/v1/comments?product=pension_sipp&emotional_register=peace_of_mind_relief&min_likes=5&limit=30

Read through each set and note the recurring themes, language patterns, and specific concerns. This is your qualitative thematic analysis — equivalent to a researcher coding focus group transcripts.

Step 3: Test specific research hypotheses

Semantic search lets you test hypotheses rapidly. Frame your hypothesis as a natural language query and see what the corpus returns.

Hypothesis: People don’t engage with their pension until a life event forces them to

$
POST /api/v1/search
$
{
>
  "query": "life event that made me start thinking about my pension",
>
  "limit": 15,
>
  "threshold": 0.4
>
}

Hypothesis: Auto-enrolment has created passive savers who don’t understand what they own

$
POST /api/v1/search
$
{
>
  "query": "automatically enrolled in pension but don't really understand it",
>
  "limit": 15,
>
  "threshold": 0.4
>
}

Hypothesis: The gender pension gap is a lived experience people talk about

$
POST /api/v1/search
$
{
>
  "query": "pension gap women career breaks savings disadvantage",
>
  "limit": 15,
>
  "threshold": 0.35
>
}

If a hypothesis returns strong, relevant, high-engagement comments, there’s evidence for it. If results are thin or low-similarity, the hypothesis may not reflect a real consumer concern.

Step 4: Find the representative quotes

Every research report needs verbatim evidence. Pull the highest-engagement quotes for each theme you’ve identified — these are the comments that resonated most broadly and make the strongest illustrative evidence.

$
GET /api/v1/quotes?product=pension_sipp&emotional_register=anxiety_fear&min_likes=30&limit=5

$
GET /api/v1/quotes?product=pension_sipp&life_stage=pre_retirement&min_likes=20&limit=5

$
GET /api/v1/quotes?product=pension_sipp&life_event=redundancy&min_likes=10&limit=5

Step 5: Produce the research narrative

Use the AI analyst to synthesise your findings into a structured research report.

$
POST /api/chat
$
{
>
  "messages": [
>
    {
>
      "role": "user",
>
      "content": "Produce a consumer sentiment research brief on pension saving in the UK. Structure it as: Executive Summary, Key Findings (with data), Emotional Landscape (by register), Life Stage Analysis (how different stages relate to pensions differently), Trigger Moments (what prompts engagement), Barriers (what prevents action), and Verbatim Evidence (five quotes per theme). Use only evidence from the corpus — cite actual comments and translate all tag values into plain English."
>
    }
>
  ]
>
}

For a follow-up deep dive:

$
POST /api/chat
$
{
>
  "messages": [
>
    { "role": "user", "content": "Produce a consumer sentiment research brief on pension saving..." },
>
    { "role": "assistant", "content": "[previous response]" },
>
    { "role": "user", "content": "Focus now on the pre-retirement segment specifically. What are their top three fears and what does the data suggest they most need from a financial provider at this life stage?" }
>
  ]
>
}

Using the MCP server

The MCP server is particularly powerful for research workflows in Claude Desktop — you can iterate through a research question conversationally rather than constructing individual API calls.

“I’m writing a consumer research report on UK attitudes to life insurance. Start by giving me the top-line numbers from the corpus — emotional register breakdown, life stage breakdown, and which life events are most commonly associated with life insurance comments.”

“Now look at the anxiety and grief segments specifically. What are the recurring themes? Give me the three most important insights with verbatim quotes supporting each one.”

“Finally, compare the language used by people who have taken out cover versus people who are still procrastinating. What’s different?”

Pagination for large data pulls

For full corpus analysis, use the offset parameter to paginate through all matching comments:

$
GET /api/v1/comments?product=pension_sipp&limit=100&offset=0
$
GET /api/v1/comments?product=pension_sipp&limit=100&offset=100
$
GET /api/v1/comments?product=pension_sipp&limit=100&offset=200

Feed the full set into your own analysis pipeline — topic modelling, sentiment analysis, or LLM-based thematic coding.

Tips

• Cross-tabulate signals with emotions — comments tagged protection_gap_described that are also anxiety_fear are your highest-priority research subjects. They represent people who are aware of the problem but haven’t acted. Filter for both and you have a ready-made segment profile.
• Use semantic search for competitive intelligence — search for competitor brand names or specific product names to understand how real consumers talk about them.
• Benchmark across time — as the corpus grows with new channels and sources, you can re-run the same queries and compare results to track shifts in consumer sentiment over time.