Figure — Reported time saved
up to 80%
of analyst time currently goes to data prep and manual reporting — much of it automatable
Handles cleaning, preparation and first-pass pattern detection, so people are freed to ask the right questions and interpret what the data means for the business.
Of analyst time currently goes to data prep and manual reporting — much of it automatable. Every figure on this page is traced to a named source below, with its method and its limits.
Read the evidence01
What these agents do
Analytics agents monitor metrics, detect anomalies, explain movements in plain language, compare cohorts, and produce forecasts. Assistive ones surface findings; autonomous ones refresh and publish forecasts on their own.
02
Genuinely good at
- Watching more metrics, more often, than any analyst team
- Turning a warehouse query into a readable narrative
- Surfacing an anomaly quickly enough to matter
- Consistent, repeatable scenario modelling
03
Genuinely bad at
- Sparse data, where the explanation will still sound certain
- Causation. It will find correlation and describe it fluently.
- Working without a well-defined semantic layer
- Knowing that an upstream pipeline broke rather than that demand fell
04
The risks that matter
- Reliability: a plausible explanation is more dangerous than an obviously wrong one.
- Decision risk: forecasts published without review get treated as facts.
- Access: warehouse credentials are the widest data access most organisations grant.
- Cost: query-generating agents can be expensive in ways that surface a month later.
05
How to evaluate one responsibly
- Ask what happens when the data is thin — does it abstain or does it answer?
- Test it against a movement you already understand.
- Confirm the credentials are read-only.
- Ask about query cost controls before, not after, the pilot.
Now compare what is actually declared.
The registry holds each agent’s stated facts — autonomy, oversight, compliance, residency, sustainability disclosure — with provenance on every field.
Compare analytics agents06
The evidence
- Independent research
- No commercial interest in the outcome. Method published.
- Analyst research
- Structured method, but the publisher sells advice in this market.
- Practitioner survey
- Self-reported. Good for direction, unreliable for magnitude.
- Vendor or customer material
- A claim about the publisher's own product. Not a finding.
Figures below are reproduced as published. We label who paid for the work and what the method can and cannot show. Where a number is self-reported or vendor-published, we say so next to the number rather than in a footnote.
4 of 4 sources
- 01Independent research
Access to an AI assistant improved data-analysis task performance, with the largest gains among less experienced analysts.
- Source
- National Bureau of Economic Research / academic replication — Generative AI and knowledge-work productivity literature2023–2025
- Method
- Randomised and quasi-experimental studies on structured analytical tasks.
- What it does not show
- Task-level studies. None measure whether the resulting decisions were better, which is the only outcome that matters here.
- 02Independent research
“When answers become cheap, the question becomes really important.”
- Source
- Cassie Kozyrkov — Public commentary, former Chief Decision Scientist, Google2023
- Method
- Expert opinion. Not a study.
- What it does not show
- Framing, not evidence. Included because it is the correct frame, not because it is measured.
- 03Analyst research
Analysts spend up to about 80% of their time on data preparation and manual reporting.
- Source
- Forrester — Data-preparation and analyst time-allocation research2022–2024
- Method
- Practitioner surveys and client data across data and analytics teams.
- What it does not show
- The 80% is an upper bound frequently quoted as an average. Preparation removed is not insight gained.
- 04Vendor or customer material
Some reporting workflows compress from three or four days to under four hours.
- Source
- Practitioner case reports — Published implementation write-ups2024–2025
- Method
- Single-team accounts, usually published alongside a tool recommendation.
- What it does not show
- Measures cycle time on a workflow chosen because it worked. No accounting for validation time added downstream.
We take no payment from any organisation named on this page, and no source is listed or omitted on commercial grounds. If a figure here is wrong or out of date, tell us and we will correct it with the date of the change.
Ideas in this domain
Continue