CASE STUDY · AI TRANSFORMATION
AI workflow automation for equity research
They had four times the requests and the same number of analysts. Eight weeks later we'd built a pipeline that automates the heavy reading and writing, so the team could focus on the parts that actually need an analyst.
5-7d→<1d
Report turnaround
~4×
Throughput capacity
8 wk
Idea to production
THE SITUATION
Four times the research volume in a year
The company writes equity research for institutional clients. It's methodical work, done by hand: read the filings, build the models, write the report. That was fine when volume was steady. Over the past year requests had grown three to four times and the team hadn't. Reports that used to take a week were stacking up. IPO coverage was the worst of it. A single prospectus runs 400 to 600 pages, and someone had to read all of it, pull out what mattered, and turn it into a structured report. They'd been using ChatGPT and Claude here and there, but nothing systematic. Engineering had moved their development work onto AI tools. The research process itself, the part generating revenue, was still entirely manual.
THE BRIEF
Scaling equity research without new hires
The CEO wanted more throughput without hiring proportionally. Quality couldn't slip. Clients expected the same scrutiny they were used to, and compliance left no room for shortcuts. The budget wasn't huge either. The question was how to get more reports out without the team working weekends.
THE WORK
Building a multi-agent research pipeline
We started by sitting with all four analysts, watching them actually work. That's the part most teams skip, and it's where the real picture shows up. We read their old reports, talked to people in other functions, and traced how a single report travelled from request to delivery. A lot of it didn't survive a simple question: why do you do it this way? Some steps had been there for years, more habit than need. And the reports weren't one job. An IPO meant chewing through a 600-page prospectus from cold; a company update was really about remembering what you'd said last time and what had moved since. Same name, two different problems. We built a multi-agent pipeline in n8n, mixing OpenAI and Anthropic models by task. One is better at parsing documents, another at planning and writing, so we let each do what it's good at instead of forcing one model to do everything. A separate agent went out to public sources for market context, comparable deals, and recent news — the background reading analysts used to do by hand. The prospectuses were the hard part. 600 pages is too much for any single pass, so we broke them into sections, summarised each chunk in parallel, ran sentiment across the lot, and stitched it back into a structured draft. Four two-week sprints. Every one had to put something in front of the team that worked on real reports, not a demo. By sprint three it was running on live work. Production by four.
THE OUTCOME
Freeing analysts for higher-value research
A first draft that used to take an experienced analyst two or three days now comes back in under 30 minutes. The pipeline does the reading and the first pass; the analysts pick it up from there — commentary, client framing, the final call. The work only they can do. Four analysts, four times the volume, no new hires. The backlog's gone, and they spend their days on analysis instead of data collection.
FAQs
Common questions
Using AI to take over the repetitive parts of research: reading documents, pulling data, drafting reports. The analysts still do the analysis, but the system removes the slog around it.
Manual workflows slowing you down?
Tell us what's eating your team's time. We'll give you an honest read on where AI can help.