Profile twelve competitors by Wednesday
At the start of a strategy engagement, the manager asks an analyst to profile twelve competitors: revenue, ownership, footprint, recent moves, positioning. The analyst opens dozens of tabs, reads annual reports and news, copies figures into a spreadsheet and pastes quotes into a document. By Wednesday the profiles are done, but half the sources are no longer recorded, two revenue figures disagree, and nobody is sure which year a number refers to.
When the partner asks where a figure came from, the analyst spends an hour finding it again.
Why desk research is slow and fragile
- Searching and reading take most of the time, and much of it is repeated across analysts.
- Sources are not recorded as facts are collected.
- Figures from different sources and years conflict, and this is noticed late.
- Findings are not organised against the questions the engagement needs to answer.
- Research from previous engagements is not reused.
Research also sits under time pressure it was not designed for. It happens in the opening days of an engagement, when the team is still setting up, and it feeds everything that follows. Errors made in a rush in week one travel into the final recommendations.
What that costs
| Issue | Effect |
|---|---|
| Analyst days on collection | Less time on analysis and synthesis |
| Missing sources | Figures cannot be defended to the client |
| Conflicting figures | Credibility risk in front of the client |
| No reuse | Same research done again on the next engagement |
A figure in a client deck that cannot be traced to a source is a liability. Clients do check, and they remember when a number turned out to be wrong.
How we build a research assistant
- The analyst sets out the research plan: questions, entities such as competitors or markets, and time period.
- The assistant searches the web, company filings such as Companies House records, and any subscribed databases you have access to through their APIs.
- A language model reads the sources and extracts findings against each question, with a citation and link for every claim.
- Conflicting figures are flagged side by side with their sources and dates.
- Findings are organised into a structured output, profiles, tables or briefing notes, in your templates.
- The analyst checks each finding against its source, edits and approves it. Unverified findings are marked as such.
- Approved research is stored for reuse on future engagements, with dates so staleness is clear.
Models can misread sources or state things confidently that the source does not support. That is why every claim carries a citation and an analyst verifies it before it goes near a client.
We usually start by building the research plan template with your analysts and trying it on a recent engagement's research questions, so they can compare the output with what they produced by hand. That shows where the sources are strong and where they need supplementing.
What research looks like afterwards
The analyst starts with a draft set of profiles, each fact linked to its source, conflicts already flagged. Their day goes on checking, filling gaps and working out what it means for the client.
When a partner asks where a figure came from, the answer is a click.
Is this your analysts' week?
- Desk research takes days per workstream.
- Sources are lost between collection and the deck.
- Conflicting figures are found late.
- Research is repeated across engagements.
- Analysts spend more time collecting than analysing.