Case Study · Research
Case Study: Consulting Firm Cuts Research Time 65% with Cited Briefings
How a 15-person strategy boutique shipped the AI research assistant workflow — sub-query planning, web search, and footnoted Markdown briefings for client work.
Edited by The AIKnowHub team · Editorial teamResults
Median memo research time
2 hr
Cost per briefing
$1.20
Partner acceptance rate
78%
Hallucinated claims caught
0 shipped
Background
Northbridge Strategy produces 20–30 research memos monthly for PE and corporate clients. Associates burned half a day on search, skim, and synthesis before writing.
Pipeline
From Build an AI Research Assistant:
Research question
→ Planner: 4-8 sub-queries
→ Tavily per sub-query (top 5)
→ Fetch top 3 pages per sub-query
→ Haiku: rank source relevance
→ Sonnet: cited briefing + Limits section
→ Associate verification (required)
→ Notion export
Quality gates
- Minimum 8 unique source URLs
- No claim without inline citation
- Limits section lists uncovered angles
- Partner review before client send
Results (10 weeks)
| Metric | Before | After |
|---|---|---|
| Median research time | 6 hr | 2 hr |
| Memos / associate / week | 1.2 | 2.4 |
| Partner rework rate | 41% | 22% |
| Cost per memo | — | ~$1.20 |
Takeaway
Research automation works when citations are mandatory and humans verify before client delivery. Speed without sources is liability in consulting. Blueprint: workflow guide.
Lessons learned
- 01
Planner step with 4-8 sub-queries beat one-shot search — coverage improved immediately.
- 02
Limits of this briefing section was mandatory — partners trusted honesty over completeness theater.
- 03
Fetch full page text for top 3 URLs only — fetching everything burned tokens and time.
- 04
Associates still spent 45 min verifying citations — AI shortened search, not accountability.
- 05
Sonnet for synthesis, Haiku for sub-query ranking — predictable cost per memo.
Frequently asked questions
Related on AIKnowHub
Workflow
Build an AI Research Assistant
A research agent that takes a question, searches the web, reads sources, and produces a cited briefing.
Concept
Browser Agents Explained
AI that controls a web browser — clicking, typing, scraping, and filling forms. Use cases, risks, and how it differs from API-based automation.
Concept
Human-in-the-Loop Automation Patterns
When to require human approval in AI automations — approval gates, review queues, selective auto-send, and the metrics that tell you when to loosen the loop.
Comparison
Cheapest AI APIs
Per-million-token cost comparison across providers, with notes on where the cheap models are actually good enough.
Workflow
Build a Competitive Intel Monitor
A weekly agent that tracks competitors — pricing, features, launches, hiring, and positioning — and delivers a cited briefing to Slack or email.
Prompt
Executive Research Brief
Distill complex research findings into a one-page brief for decision-makers.