Theus and Microsoft 365 Copilot
Copilot is broad work AI across Microsoft 365. Theus does a narrower job: it keeps research evidence intact, from the source page to synthetic testing.
Reviewed October 2026. Statements about other products reflect the cited sources as of that date.
What Copilot does well
Works across Microsoft 365
Email, files, meetings and chats, with Researcher and Analyst agents for multi-step work.
Retrieval that respects permissions
People only get results from content they can already access.
Agents that work together
Copilot Studio agents can delegate to each other and connect to more systems.
Copilot and Theus side by side
Different tools for different jobs. The aim isn’t to replace Copilot.
| Capability | CopilotWork AI across Microsoft 365 | TheusGoverned research evidence |
|---|---|---|
| Everyday work | Built for email, meetings, documents and agents | Not the goal. Theus is for research teams. |
| Finding documents | Strong across permitted Microsoft 365 content | Searches the Research Base built from your research |
| Unit of knowledge | Text passages retrieved from files | Facts with method, sample, statistics and page |
| Charts and tables | Copilot’s Retrieval API doesn’t retrieve from images or charts, and reads table text only in Word and PowerPoint files | Read and stored as facts with their values |
| Contradictions | Can compare passages when asked | Stay visible as part of the answer |
| Gaps in the evidence | Not a documented focus | Named, and Fill returns cited candidate facts for review |
| Consumer response testing | Can answer in character when prompted | Moderated Panel grounded in your Research Base, labeled as hypotheses |
| Asking why | Answers can reference the files they drew on | Supported reasoning is designed to resolve to Fact ID and page |
Where research needs more
Copilot handles “what did last year’s texture study say?” well. “What drives repeat purchase across all our studies, and where do they disagree?” needs each finding to keep its method, sample, statistics and page.
The study found significant differences between products.
Product A (7.2 ± 0.3) scored significantly higher than Product B (6.1 ± 0.4) on sweetness intensity (p < 0.05, Tukey HSD, n = 120 consumers, 9-point intensity scale).
The Theus fact keeps its Fact ID and source page. Illustrative example.
How Theus works on the same evidence
Research Base
Reports, tables and charts become facts that keep their statistics and page.
Expert
Dr. Theodore Sinclair answers across studies, shows contradictions and gaps, and says when the evidence can’t support a conclusion.
Panel
Dr. Evelyn Reed moderates a synthetic discussion grounded in your Research Base. Persona-only behavior is labeled, and results are hypotheses to test.
Need full synthetic category panels instead?
See ForethoughtDr. Evelyn Reed and Dr. Theodore Sinclair are fictional personas.
Why prompting a persona isn’t enough
You can ask Copilot to answer as a 45-year-old shopper. You get the model’s picture of that person, not that person.
Results vary by domain
Studies couldn’t reproduce known consumer effects such as the endowment effect and the sunk cost fallacy.
Sarstedt et al., Psychology & Marketing (2024) (opens in a new tab)
About a third replicated
GPT-3.5 reproduced 37.5% of the original results in the eight Many Labs 2 studies that could be analyzed.
Park et al., Behavior Research Methods (2024) (opens in a new tab)
Answers cluster around the average
Synthetic users varied less than people and missed meaningful subgroups.
Nielsen Norman Group, AI simulations studies (opens in a new tab)
That’s why Theus grounds panelists in your Research Base and treats every Panel result as a hypothesis to test with people.
When Copilot is the right choice
Summaries
Reports, SOPs and submissions you already have.
Meetings
Notes and follow-ups.
Updates
Stakeholder drafts from project material.
Workflow
Forms, approvals and scheduling.
Many teams run both: Copilot for the work around the research, Theus for the evidence itself.