For insights & market research teams
The research assistant that knows what you already found.
Your team is sitting on years of decks, transcripts and tracker waves that nobody can search. Deckhand is an extra pair of hands for that archive: it indexes everything and puts an agentic researcher on top — so every new question starts from everything you've already learned, with every claim cited to the slide it came from.
No credit card required · your data stays yours
Has willingness to pay for premium moved since the 2024 repositioning?
What your research says
Stated WTP for the premium tier rose from 34% to 41% between tracker waves 5 and 7. W7 · S12
Qual work attributes the shift to packaging, not price framing. Focus groups '25 · S8
What public sources say
Category-level spend on premium variants grew 6.2% year on year. ONS · 2026
Peer-reviewed work finds stated WTP overstates realised WTP in FMCG. OpenAlex
The problem
Your research archive is write-only.
Every study your team has ever run produced a deck, and every deck went into a folder no one opens again. So new research quietly repeats old research — and the institutional memory walks out the door with whoever ran the study.
Years
of decks, transcripts, tracker waves and field notes — effectively unsearchable the day they're filed.
Twice
is how often you pay for a finding when nobody remembers wave 4 already answered the question.
Zero
citations in the average "I think we looked at this once" answer. Institutional memory shouldn't run on vibes.
How it works
From a folder of decks to a researcher who has read them all.
STEP 1
Upload your archive
Drop in your PowerPoint decks. Deckhand recovers what naive extraction destroys — exact chart series, crosstab columns with bases and significance letters, framework arrows, speaker notes — and anchors every fact to its slide.
STEP 2
Ask across everything
Search the whole corpus by keyword or by meaning, or ask a question in a sentence and get an AI-written answer where every claim cites the deck and slide it came from — and the unanswered part is named, not papered over.
STEP 3
Commission the researcher
For questions that need more than one search, an agentic researcher plans an investigation, works through your corpus, optionally consults official statistics and published research, and returns a cited brief.
The product
Built for evidence, not summaries.
Ingestion that respects your data.
Most tools flatten a deck into text and lose exactly the things a researcher cares about. Deckhand keeps the numbers as numbers.
- Exact chart series data, not a caption's approximation
- Crosstabs with per-column bases and significance letters intact
- Framework arrows, speaker notes and reviewer comments preserved
- Every fact anchored to its slide, so the agent can cite it
| Segment | Total | A · 18–34 | B · 35–54 | C · 55+ |
|---|---|---|---|---|
| Definitely / probably would | 41 | 52BC | 38 | 29 |
| Might or might not | 33 | 28 | 36A | 35 |
| Probably / definitely not | 26 | 20 | 26 | 36AB |
| Base (unweighted) | 2,014 | 612 | 846 | 556 |
Answers with receipts.
Ask a question in plain language. Hybrid retrieval — exact full-text and semantic search, fused — finds the slides; the model writes an answer where every claim carries its citation.
- Keyword mode: every word must appear on the slide, with document, slide and study context on every hit
- AI insight mode: a written answer, claim by claim, cited to deck and slide
- The unanswered part of your question is named explicitly
- Suggested next questions drawn from what the search noticed — an unstated basis, two sources that disagree
A researcher you can watch work.
The agentic researcher doesn't hand you a black-box answer. It shows its trace as it goes — what it read, what it searched, what it asked outside — before you read what it concluded.
- Reads your study catalogue first, then searches the corpus several ways
- Your findings and public sources reported in two columns that never merge
- The gap between them stated, not smoothed over
- Follow-ups name the source that would settle them — "worth investigating" becomes "worth asking the ONS"
- 00:02Read study catalogue — 34 studies, 6 tracker waves
- 00:11Corpus search: "willingness to pay premium" — 14 slides
- 00:19Corpus search (semantic): price sensitivity, repositioning — 9 slides
- 00:24Contradiction check: W5 vs W7 basis definitions
- 00:31External · ONS: category spend, premium variants
- 00:38External · OpenAlex: stated vs realised WTP
- 00:47Writing brief — 11 corpus citations, 3 external
Security & control
Confidential by architecture, not by promise.
A research question is confidential before any answer is. Deckhand is built so that reaching outside your walls is a deliberate, audited choice — and reaching across tenants isn't possible at all.
Isolation enforced by the database
Every tenant-scoped table sits behind Postgres row-level security with an unbound-means-deny policy. The application connects as a role that cannot bypass it — cross-tenant access is a property of the connection, not of anything a request could influence.
External research is off by default
The researcher consults outside sources — official statistics, published research, company filings — only when your subscription opts in. Each source is separately switchable, so an NDA that forbids naming your category to a search vendor can still allow ONS and OpenAlex.
Corpus first, enforced in code
Even with external research on, the agent must establish what your corpus already knows before any external tool will run. That ordering is enforced in code, not requested in a prompt — and every external query lands in the audit trail.
Spend you can cap
Monthly AI-spend and storage limits per subscription, checked before work is queued and again when it runs. Every tool the researcher holds is read-only. Roles separate researchers, authors and admins cleanly.
Get started
Give it the deck you're proudest of.
Start a free trial, upload a handful of decks, and ask the question you already know the answer to. Watch it come back cited to the slide.
Free trial includes 25 decks and full search. No credit card required.