How a global advisory firm automated client research

A global leader in research and advisory.

A global research and advisory firm serving heads of in-house technology and operations centers, with two decades of proprietary benchmark data and reports.

Region

Global

Sector

Research and advisory

Annual revenue

> $10M/year

Employees

500+

Function in scope

Client research queries

Every question became a research request.

The firm’s research lived in two forms that were hard to use together, and every client question that crossed both waited on an analyst.

1. Two sources, one questionBenchmark data sat in one workbook. About 900 reports sat in a separate library.
2. Answers waited on analyst timeQuestions were sorted by effort. Simple ones took up to an hour, medium ones half a day to a full day, and deep ones three to five days.
3. Every answer rebuilt from scratchEach question meant pulling the number from the workbook, hunting down the right report and writing it up, then doing it all again for the next client.
4. Trust could not be compromisedMost of the library predated 2021, so old research could easily surface as current. And for a firm selling trusted data, one invented number was disqualifying.

A simple document chatbot was not enough for this research.

xA generic chatbot could not be trusted with this research. To get it right, the advisor had to pass four tests.

1. No invented factsEvery number had to come from the firm’s own data and every claim from a named report and page.
2. Recent research firstMost of the library predates 2021, so older material could be used only after telling the user it was older and asking if they wanted it.
3. Answers across both sourcesA single question had to draw on structured benchmarks and on research reports, including charts and tables inside slide decks.
4. Protect the proprietary dataOnly aggregate benchmarks could be exposed, never company-level records, and the advisor could never discuss competitors or other client companies.

One agent, two knowledge bases, every answer cited.

Lyzr built a conversational research advisor that answers from the firm’s own benchmark data and research reports, framed around each user’s own company.

Core Capabilities

• Benchmark queriesReads about 30 aggregate benchmark tables through a read-only database tool, with every query checked before it runs
• Research searchSearches 2021-onward research by default, and an older archive only after asking the user
• Company-aware answersFrames each answer around the user’s sector, size and locations, using a curated company profile
• Page-level citationsCites every figure to its benchmark and every claim to a report and page, and one click opens the source

Tracing every number back to its source.

If the data isn’t there, the advisor says so.

Read-only, approved tables only

Every query is checked by the database’s own parser, so the agent cannot read anything outside the approved aggregate tables.

Recency by design

Recent and archive research live in separate knowledge bases, and older material is used only after the user is told and agrees.

Scope limits

The advisor never names or discusses other advisory firms or other client companies.

Human in the loop on deep analysis

Deep-analysis questions are not left to the advisor alone, and an analyst stays in the loop.

Reversible changes

The firm’s team updates benchmarks, profiles and reports through an admin panel, and every change is logged with one-click undo.

What was once research, is now a repository.

Hours of research is now readily available as a time-saving resource.