Forward-deployed engineers for buy-side research desks
Know what changed, and why it matters to your book.
We sit with your research team, then build private agents in your own cloud. They read the broker notes, filings, transcripts and alt data you already pay for, and tell each analyst what moved on their names before the open.
No new platform to learn. Results arrive in the inbox, notes and spreadsheets your team already uses.
8-K at 05:42: the CFO is leaving. Guidance is not changed in the filing. Last quarter's call, where the CFO set out the margin plan, is linked.
Estimate
Company B long
Broker note moved Q3 tracking from in line to above consensus. Spend per customer slipped for a second report running, so that is the leg to watch.
Data
Company C short
Card panel arrived with 18% fewer merchants than usual. Signal refresh paused until the vendor confirms.
Left out: a news-feed report of an upgrade on Company D. The broker's own note doesn't show it.
35 other names checked. Nothing material since yesterday.
Synthetic example. Company names are placeholders.
Earnings calls, read against last quarter.
Each transcript is lined up with the prior one from the same company. Numbers that moved, words that softened and lines that quietly disappeared are pulled out and tied to your position, minutes after the call ends.
Research agents · overnight runyour cloud
Filings
Broker notes
Alt data
Expert calls
News
Revised upCompany B · tracking above consensus
Tone shiftCompany A · outlook language softened
Sources disagreeCard panel vs. expert call
CompliancePossible MNPI held back
Brief ready3 of 38 names need your eyes
Company A · Q3 prepared remarks vs. Q2ready 11 min after the call
“We expect demand to remain strong resilient through the second half, with some variability by region.”
Softer word. Regional variability is new this quarter.
“We are reaffirming updating our full-year outlook to reflect current trends.”
The guidance numbers held. The verb changed.
“Pricing continues to be a tailwind for the business.”
Dropped. This line was in the last four sets of remarks.
“We are reviewing our cost structure across the organization.”
New. First mention of costs in six quarters.
Broker note · Company B · Q3 tracking
Tracking moved from in line to above consensus
Revised up
Against the same broker's last note
Metric
Prior
Now
Cons.
Revenue growth Y/Y
+6.1%
+7.4%
+6.3%
Active customers
+2.0%
+2.2%
+2.1%
Spend per customer
+4.0%
+3.6%
+4.2%
Wording
“demand remains steady”→“demand is accelerating into the quarter”
“no change to our outlook”→“upside to our prior estimate”
For the book
Supports the long. Spend per customer slipped for the second note in a row; watch it into the print.
The broker's panel was re-based this month, so Y/Y uses restated history. Two quotes linked to source.
Expert calls, checked while they happen.
We connect to the note-taker your analysts already use, such as Granola, and read the transcript as it's written. Claims are checked against your model, your data and earlier calls.
Expert call · former regional sales lead14 min in
Expert
“Pricing held through the spring, but we started seeing a lot more discounting in June.”
EarlierContradicts a March call with another expert, who said “no discounting this year.”
Expert
“I'd say fifteen, maybe twenty percent of our accounts were testing the new entrant.”
DataHigher than the panel. Panel share for the entrant is about 11%. Worth asking how accounts are counted.
Expert
MNPIPossible MNPI. Held back. The segment is kept out of the shared notes and compliance has the original. Nothing from it reaches the summary.
AskFollow-up while they're on the line: “Did the June discounting carry into July, or was it end of quarter?”
Compliance sees it first
Language that suggests material non-public information is flagged the moment it's said and kept out of shared notes, following your firm's policy.
Checked against what you know
New numbers are compared with your model and licensed data. Contradictions with earlier calls are shown with the source.
Notes filed before you hang up
Main takes, model-relevant numbers and verbatim quotes, filed to the right company in your notes system.
Alt data you already license, checked each time it lands.
Funds pay for card panels, receipts, app and web traffic. Analysts open a fraction of it. Agents take in each release, check vendors against each other and rerun the analyses your team relies on.
Triangulation
Several vendors on the same KPI, adjusted for each panel's known bias. Disagreements are flagged ahead of the print.
Panel health
Shrinking panels, merchant drop-outs and rebases caught before they show up as a false signal.
Point-in-time
Each series is stored as it was known on the day, so backtests can't peek at the future.
Dataset trials
A new vendor is tested against reported numbers before you sign the contract.
Company B · quarter-to-date KPI growth, Y/Ysynthetic
Quarter-to-date KPI growth by source
Source
Read
Chart
Card panel
+4.1%
Receipt panel
+3.4%
App activity
+5.0%
Bias-adjusted blend
+4.0%
Consensus
+2.8%
Sources agree: tracking about 120bp above consensus. Panel health normal.
Cohort retention, rebuilt with each data dropsynthetic
Retention by acquisition cohort, rebuilt with every data drop.
Months since first purchase →← Acquisition cohort
Higher retentionLower retentionNewest month of data
Cohort retention
Newer cohorts are holding on longer
Share of each acquisition cohort still active, by months since first purchase.
Q1 '25
Q2 '25
Q3 '25
Q4 '25
Q1 '26
What the agent flags: month-3 retention is up about 4 points across the last four cohorts, so growth is coming from customers who stay.
View data tableShare after a new entrant
Who is the entrant taking share from?
Share of category spend by month, from a transaction panel.
Incumbent A
Incumbent B
New entrant
Other
What the agent flags: the entrant reached about 11% share within a year, taking roughly 1.6 times as much from Incumbent B as from Incumbent A.
View data table
All examples on this page use synthetic data. No client, vendor or company data is shown.
Where your data goes.
your environment
Into your own cloud account, and no further than you approve. Each line below is a term in our client agreement.
Runs in your environment. We build and run everything in your cloud or on-prem. No copies on our systems unless you approve one in writing for a specific task.
Model calls on your terms. Only to providers you approve, under terms that forbid training, with zero retention where the provider offers it. Or to models hosted inside your account.
Never used for training. Your data and outputs are never used to train or improve any model, ours or anyone else's. The one exception is a model you ask us to tune that stays in your account.
Never shared. Nothing from your fund is used for another client, in any form, including anonymized.
Yours at the end. You own what we build for you, with a permanent license to any reusable parts of ours inside it. Your data is returned, then deleted, and we certify the deletion in writing.
How data moves in a typical deployment. *Where the provider offers it.
Forward-deployed: we build at your desk.
The engineers you meet on the first call are the ones who sit with your analysts and write the code. The diagnostic and pilot are fixed fee, agreed before we start.
Week 0
Sit with the desk
We spend a few days with your analysts and PMs, see what they read each morning, and pick one workflow worth automating.
Weeks 1–2
Ship a first version
A working tool on your coverage, in your environment, with your security team involved from day one.
Weeks 3–8
Tune it at the desk
Daily feedback and tests against your own past examples, until the team uses it without being asked.
After
Hand over or keep going
You get the code, documentation and a walkthrough for your team. Or we stay on a monthly retainer and build the next one.
Tell us what your team reads every morning.
You'll talk to the engineers who would build it. We'll say plainly whether we can help, and where we'd start.