They wanted to serve more clients without giving up more evenings.
The market-data company wanted to serve more clients, but its partners were already spending 40 to 60 hours a week analyzing research by hand and didn't want growth to take more of their evenings away from their families.
The company sold research to hedge funds and private equity firms, using changes in brands' marketing emails to help clients understand discounts and campaign volume. To answer a question, a partner or subject-matter expert searched an archive of ~15 years of emails, checked the results, and wrote a response, with much of the useful information still sitting inside images.
They had 27 clients and ~$4M in annual revenue, along with a waitlist of funds that had wanted to buy for years. They'd tried hiring analysts to take on more work, but the experienced team was too busy serving existing clients to train them, and so the hires hadn't created the capacity they needed.
They came to us with another idea: build an AI simulator that would train those analysts faster. Before building it, we looked at why even their experienced people needed so much time to answer a question.
Even their best analysts were waiting on the same slow searches.
Over a week, the team logged its time while our observation tools examined the work inside the company's private environment. What we found was that research consumed much of the day because people kept waiting for database searches, trying again to find useful records, and reading information from campaign images.
A typical partner search took five to ten minutes, and a fully trained analyst would still inherit those same searches and manual steps. A simulator might shorten training, but it wouldn't remove the work that limited how many clients the company could serve.
We put that finding into a memo and a map of the workflow, then proposed rebuilding the data and answer process so the existing team's expertise could serve more clients.
The system had to protect the reputation behind every answer.
Moving delivery into an AI product raised another concern for the partners: they didn't want an incorrect answer reaching a fund, because one bad number could follow their name into the next client relationship in a business built on referrals.
So we proposed building the data foundation and proving the answers before exposing them to clients. If the client-facing version didn't meet the agreed bar, the same system could still speed up internal research, giving the partners a useful outcome without asking them to rely on an untested product.
We made the evidence usable before we automated the answer.
Reliable answers depended on usable evidence, and so the first three of the twelve delivery weeks went into extracting information from campaign images, reconciling duplicate feeds, and rebuilding the store around consistent records. We indexed those records so the system could retrieve the requested figures directly, with extraction matching the experts' readings 98.4% of the time on the evaluation sample.
Once that foundation was in place, we built a natural-language interface that connected each answer to its supporting records. Exact figures came from the data store, while the AI explained them and cited the evidence so a client could see where an answer came from.
Before involving clients, we tested the system against six months of real questions and asked partners to grade its answers against the responses their experts had already sent. Some question types still proved risky, and so we built a separate review step that routed those questions to a person.
That gave the company two ways to deliver the service: direct access for clients who wanted it and the existing email channel for everyone else. We also made custom videos answering clients' own recent questions so they could compare the new system with research they already trusted.
We missed a requirement and went back to solve it.
Our first version kept the best current view of the data, but clients also needed to recover a number exactly as it had been known at an earlier date. A later extraction or correction could change a historical value, which meant someone could unknowingly test a past investment decision against information they hadn't had at the time.
When we caught that gap after the engagement, we told the client before they raised it and built the versioned history at no extra charge. The correction preserved earlier values instead of silently overwriting them, allowing historical analysis to use the information available when the decision was made.
The full waitlist became clients, and the partners got their time back.
Once routine delivery no longer depended on partners researching every answer by hand, the company could take on its full waitlist. In ~3 months, it grew from 27 to 51 clients and booked ~$3.6M in new annual recurring revenue, with contracts signed and billing started.
That meant 24 additional clients without hiring new subject-matter experts, while each partner reclaimed 40 to 60 hours a week, even after allowing for the oversight the new system required.
Because those additional clients needed much less expert labor, the growth also reached earnings: annual revenue rose from ~$4M to ~$7.7M, while annual earnings before interest, taxes, depreciation, and amortization rose from ~$1.1M to ~$4.6M.
We also trained the team to add data providers, adjust which questions went to a person, and build new workflows so they could keep developing the service themselves. They later built their own client-facing interface on the foundation we delivered, without bringing us back to build it.
For a closer look at the research and answer process.
The technical companion explains how we turned campaign images into usable records, tested answers against past client questions, and kept higher-risk questions under human review. It also covers the historical-data correction, result measurements, and handover.
They came for faster training. We gave them room to grow.
The partners came to us for a faster way to train analysts, but measuring the work showed that even a skilled hire would inherit the same slow research process. We rebuilt that process so the company could serve its waitlist while partners spent less time researching answers by hand, then taught their team how to keep building on the system.
If you’re trying to grow your own business, start with your goal, look at what it'll take to achieve it, and then identify what's keeping you from taking action. Just remember that every business is different and that research that depended on partner time was just their bottleneck and it might not be yours.
If you’re a financial-services leader and want to grow your firm without growth depending on more partner time, we’ll identify the biggest bottleneck standing in your way, build the plan to remove it, and implement it in eight weeks* or less - guaranteed or we continue to work for free and you don’t pay.
Find my bottleneck* Eight-week delivery, or twelve weeks when substantial data foundations are required, as agreed before work begins. If a covered stage fails its agreed benchmarks, we provide up to four additional weeks of remedy work at no charge, then waive or refund that stage’s fee if it still fails. Successfully delivered stages remain payable. Agreed client-caused delays pause the clock. Delivery guarantee terms.