Analytics & Research
Power BI
Microsoft BI with Copilot in the tenant
Power BI is the default BI if you already pay Microsoft. Copilot needs a clean model. Do not skip the dataset owner.
List price: Pro ~$10/user/month. Premium/PPU and Copilot have separate packing. Confirm current packing.
Review status
Microsoft BI with Copilot in the tenant
Microsoft gravity. Pro is not Salesforce-expensive.
No four-job lab on this one yet. Placement is editorial. Confirm the price with the vendor.
Where it earns the seat
- • Often already licensed
- • Strong Microsoft gravity
- • Copilot in a tenant you administer
Where it fails
- • Model quality is the product
- • Premium features hide in SKUs
- • Not a research engine
Best for
BI copilots
Natural-language questions on a dataset someone owns.
Operator review · evidence before adjectives
Power BI evaluation plan
First decide whether the question belongs to the open web, a document corpus, a spreadsheet, or the governed warehouse. Products that blur those sources create persuasive wrong answers.
Know which job you are buying
External research
Find claims, sources, companies, and market evidence
Boundary: Every material claim needs a source that actually supports it
Enterprise search
Retrieve internal documents under existing permissions
Boundary: ACL inheritance and source freshness are the product
BI and semantic layer
Answer governed questions about company metrics
Boundary: Metric definitions belong upstream of the chat box
One-time analysis
Explore a clearly limited CSV file or workbook
Boundary: Repeatable calculation steps beat an unexplained chart
The four-run test
Make the demo survive real work
Use the same inputs with every finalist. Score the artifact, the audit trail, and the failure—not the presenter.
01
The sourced market brief
Feed it: A narrow market question requiring dates, competitor claims, and primary sources.
PassDistinguishes reporting date from event date, links primary evidence, and labels inference.
Where it faceplantsCites a search snippet, circular press coverage, or a source that does not contain the number.
02
The KPI trap
Feed it: “What was net retention last quarter?” with two similarly named metrics in the warehouse.
PassUses the certified metric, shows filters and period, and identifies the semantic definition.
Where it faceplantsInvents a query or chooses the convenient metric without surfacing ambiguity.
03
The hostile spreadsheet
Feed it: A CSV with nulls, mixed currencies, duplicate rows, and one outlier large enough to move the mean.
PassProfiles quality first, preserves a transformation log, and presents robust alternatives.
Where it faceplantsDraws a polished trend before checking units and duplicates.
04
The access boundary
Feed it: A question answerable only from an executive document the test user cannot open.
PassRefuses or returns only permitted fragments without revealing titles or snippets.
Where it faceplantsUses retrieval as a side channel around document permissions.
Score the operating result
| Metric | Measure | Red flag |
|---|---|---|
| Citation validity | Claims supported by the linked source, not merely accompanied by a link | Citation count is used as a quality measure |
| Query reproducibility | A second analyst can recreate the number from logged steps | Only a screenshot or prose answer survives |
| Metric fidelity | Agreement with certified definitions and filters | The assistant can silently author new metrics |
| Permission leakage | Unauthorized content exposed in answers, snippets, titles, or logs | Search and generation are tested separately |
AAT · Editorial reading
The Dials
Four readings. Not a score. Not a room.
Microsoft gravity. The dataset owner is the product.
Company size
How automatic
Style
Collaboration
Four independent meters for this product. They do not cross. Company size, how automatic, whether it invents a voice or follows yours, and how much of the team can work in it.
More Analytics & Research tools
More tools in this category. Confirm list prices on the vendor page.
Market intelligence search for operators who live in transcripts
AlphaSense is licensed content plus search: transcripts, filings, broker research. Perplexity is not a substitute for that corpus.
Work graph search across the tools you already pay for
Glean is internal search with permissions. Buy it when the wiki is a junk drawer. It is not a public research engine.
Cited answers instead of a link dump
Perplexity is the research seat with citations. Use it for current events and competitor pages. Still verify anything that can move money.
Document matrix for diligence and research
Hebbia is for packet-scale document questions — diligence, filings, memos. A chat window is the wrong UI for that job.
BI with an Einstein/Pulse copilot layer
Tableau is the visualization system of record for many orgs. Pulse/AI is a layer on trusted extracts — not a reason to skip a semantic model.
Chat with a spreadsheet or CSV
Julius is the analyst-in-a-chat for a file you upload. Good for ad-hoc. Not your warehouse.
Search-first analytics for business users
ThoughtSpot is search-over-the-warehouse for people who will not open a workbook. Still needs a governed semantic layer.
AI search with modes, not just one chat
You.com is another cited-search seat with research modes. Use it if Perplexity is blocked or you want ARI-style research runs.
No-code predictions on a spreadsheet
Akkio is “train a model on this CSV” without a data-science hire. Fine for lead scores. Not a data platform.