Taking the busywork out of prospecting, without taking out the judgment.
I was hired to prospect. I ended up running the system the team prospects on.
How I got involved
Projects like this were not part of a BDR intern's job. My manager mentioned a stalled Apollo build in a team meeting, and I volunteered.
The workflow was meant to find the right contact inside every target account automatically, the slowest part of a rep's day at about ten minutes an account. It looked finished and produced nothing, with no error to explain why. To be useful I had to learn three things at once: Apollo, the Salesforce integration the RevOps manager had built and walked me through, and the compliance domain itself, because you can't automate a judgment you don't understand.
By hand, about ten minutes an account
Accounts worked one at a time, one to three contacts pulled from each.
The build, seconds of review
As many accounts as we choose to run, more qualified contacts than a week can absorb.
Deciding what the AI is allowed to decide
The bug was a field nothing wrote to. Fixing it was the easy half.
I rebuilt qualification in layers. Accounts qualify before they ever enter the workflow, against criteria our team and A-LIGN have built over years of knowing what a real prospect looks like. The model qualifies contacts inside those accounts, reading a person's live LinkedIn against our compliance criteria. Reps make the final call, seconds instead of ten minutes. The contact layer took the most iteration: too loose and we waste sends on the wrong people, too tight and we miss the real buyer at companies that structure the role differently. Every contact returns a written reason a rep can override.
We proved it on CMMC, the Department of Defense's standard for its contractors and one of A-LIGN's highest-demand lines. Emails through the workflow bounced at about 1%, against a 6% team average.
Then almost everything changed at once
CMMC was paused federally and the focus moved to SOC 2, where the platform isn't yet streamlined into the team's day to day: integrations have gaps, key people aren't always available, and the accounts in play became more valuable, not less. Keeping it moving meant working around all of it, bridging data in with CSV imports, getting creative with Apollo's filtering to keep running tests, and holding the scarce accounts for reps to work by hand until the output earned trust.
None of it was what I'd designed for.
What the build had to survive
A team of fourteen, not a workflow of one
Fourteen personal workflows is fourteen things to maintain and fourteen ways to drift. I built one workflow per audit line feeding a shared list reps filter themselves, so there's a single place to fix and a standard everyone runs the same way.
Two audit lines that don't share a buyer
What worked for CMMC didn't transfer. The buyer changes with the audit line, so the personas, the account sourcing, and the criteria the model reads against all had to be rebuilt per line rather than copied across. Treating them as one market was the fastest way to produce confident nonsense.
A quarter that doesn't pause for a rebuild
Reps had numbers to hit while I was still changing the thing they prospect on. Every revision had to land without disrupting live pipeline, so tests ran on batches rather than the whole list. How fast to push changes was a call I made with the managers, not one I made in the build.
Accounts and credits that both ran out
With CMMC paused, SOC 2 accounts got more valuable, not less, and every enrichment run spends from a pool the whole org shares. The question stopped being how many accounts we could process and became how many were worth it. Scarce accounts stayed with reps by hand until the output earned trust.
Reps who still need their own judgment
Hand fourteen people a system that says "qualified" and they stop building the instinct to question it. The review step stays, and every contact carries a written reason a rep can read and override, even though skipping both would be faster.
In production
Running today
- All fourteen reps run it day to day
- Trained beyond the team: reps in EMEA and the email marketing group
- Reply rate, meetings booked, and bounce tracked since the first run
- That data drives the prompt revisions and the persona rebuilds
Being extended
- Each audit line gets its own account sourcing and personas; the person who buys CMMC is not the person who buys SOC 2
- Account lists built upstream, so the model inherits a clean starting point instead of guessing
- Volume is not the limit anymore; the real decisions are how much to run and which accounts deserve it
- With the managers: when contacts reach people, what a standard play looks like, and how hard to lean on the tool while a mid-quarter number still has to be hit
The first broken version taught the lesson everything since has run on: judge a system by what it produces, not by whether it looks right.