What it is
One vendor documents chat-based experiences in its BI tool that find content, answer questions about data, summarise reports, and help authors create report pages and write measures.
When a question relates to data in a governed model, the assistant uses that model to answer. The quality of the answer therefore depends on how well the model and its definitions are prepared.
Why it matters
Natural-language access widens who can use data. It also exposes weak definitions: the same vendor says that without preparation its assistant can misinterpret the data and return generic or inaccurate results.
Accuracy is not solved. On a research benchmark for turning questions into SQL, the best system scored about 83% against about 93% for people. That measures research systems and not any BI product, but it shows the size of the gap. DII has read no evidence of how widely enterprises use these features, and some are still in preview.
The problem it addresses
DII editorial view
Many people who need an answer from data cannot write a query or build a report, so they wait for someone who can. An assistant lets them ask in plain language. It answers from the governed model it is given, and a confident wrong answer reaches people less able to spot it.
Business and organisational relevance
DII editorial view
These assistants widen who can ask a question of data. That raises the stakes on definitions, because a wrong answer now reaches people who are less able to notice that it is wrong.
Practical use cases
- Asking a question of a governed model in plain language
- Summarising a report
- Helping an author write a measure or build a report page
Benefits and opportunities
- Wider access to analysis
- Faster authoring
Risks, limitations and dependencies
- Wrong answers stated confidently; the vendor itself warns of inaccurate results on an unprepared model
- On a research benchmark for turning questions into SQL, the best system scored 82.95% against 92.96% for people. This measures research systems, not any BI product
- Cost: the documented product requires paid capacity and bills by usage
Prerequisites for adoption
DII editorial view
- A governed model with clear names, descriptions and agreed metric definitions
- A rule on which answers a person must verify
- A view of where data is processed and what usage costs
Implementation considerations
- Prepare the model before switching it on. Clear names, descriptions and agreed metric definitions are what the assistant reads.
- Decide which answers need checking. A confident wrong answer is the main risk; agree where a person must verify.
- Check where data is processed, and the cost. One vendor documents usage-based billing and a setting that controls processing outside your region.
The DII Radar assessment
Catalogue status
Not yet assessed. DII has not published a Radar assessment for AI-Augmented Analytics. It is listed in the catalogue so that it can be explained. No relevance, direction, strategic impact or enterprise adoption is given for it, and it is not plotted on the Radar.
What to consider next
DII editorial view
- Prepare the model before switching the assistant on
- Decide which answers need checking
- Check where data is processed, and the cost
Evidence and references
The sources DII read for this explainer. Each entry gives what the record holds; nothing is added.
- Copilot for Power BI overview Evidence of what one vendor ships and of its own stated limits, not of adoption. View source for reference 1 (external site)
- BIRD-SQL: a big bench for large-scale database grounded text-to-SQL evaluation (leaderboard) A research benchmark, not a measure of any BI product. View source for reference 2 (external site)