What we have studied, and what it changed.
We study the fields our work draws on: AI that works from sources, decision analysis, evidence review in law, and the standards for decision aids. Each note says what we asked, what we found and what we built because of it.
Decidient among AI tools that work from sources
AI tools that answer from documents use retrieval-augmented generation (RAG). We studied the founding paper, the main kinds of RAG product and the newest research. Each part of what Decidient does appears somewhere in that field. The whole chain, from a described decision to a checked tool with a signed record, is what we bring together.
Six ways AI answers fall short →The fields we draw on
Decision-analysis software handles weights and trade-offs, but people must bring their own options and facts. Patient decision aids are careful and neutral, but each takes months to write by hand. Comparison websites reach many people, but often rank by who pays. Decidient combines the care of a decision aid with the reach of a website, built by a production-line in days.
Lessons from legal evidence review
For twenty years, courts have accepted computer-assisted document review only when it shows how much it missed. Reviewers sample what was set aside and report a miss rate. Decidient applies the same discipline to research: it keeps a dated copy of every page it reads, logs every search, and samples what it set aside to measure what it may have missed.
The questions behind each design
We compared the questions Decidient asks when it designs a production-line with the setup questions of leading document-AI, research and decision-modelling products. Theirs focus on technical settings. Ours follow decision analysis: the problem, the goals, the options, the consequences and the trade-offs. We kept what decision analysis needs and left the technical settings to the system.
When sources disagree
Two weather forecasts can both be reasonable. So can two price forecasts. We treat every fact as a claim made by a source at a time. Disagreements are kept and shown, never hidden. Work stops only when the disagreement would change the top recommendation. Then the person sees both values and their sources.
What good decision aids require
Decision aids in medicine have a published international standard and many randomised trials behind them. Good aids are balanced, show where their information comes from and say when it was produced. Chance is given as a number first. Every Decidient tool is checked against criteria drawn from that standard before it goes live.
How decision support works.
Eleven short pieces on the method behind every Decidient tool: how we frame a decision, how we compare the options, and what you get at the end.
Your decision
How the decision works: map the choices, rules and timing.
What matters to you: turn preferences into clear priorities.
Rules, constraints and trade-offs: separate must-haves from things you can trade.
Our approach
Compare the options: see every option side by side, on the same scale.
Test the assumptions: see what happens when your priorities shift.
Stress-test the result: check whether your top option really holds up.
Methods and sources: the science under the hood.
Your decision kit
Your best matches: see the full shortlist, not just one name.
Why they rank: see exactly what is driving your result.
What could change the answer: know what would move a different option to the top.
Be ready to decide: everything you need to make the final call.
Selected references
- Lewis, P. et al. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks.
- Keeney, R. and Raiffa, H. (1976). Decisions with Multiple Objectives.
- Hammond, J., Keeney, R. and Raiffa, H. (1999). Smart Choices.
- Grossman, M. and Cormack, G. on technology-assisted review; the Sedona Conference on validation.
- Volk, R. et al. (2026). International Patient Decision Aid Standards, version 5.0.
- Stacey, D. et al. (2024). Decision aids for people facing health treatment or screening decisions. Cochrane.
- Bates, J. and Granger, C. (1969). The combination of forecasts.
- Lahdelma, R., Hokkanen, J. and Salminen, P. (1998). SMAA: stochastic multiobjective acceptability analysis.