The Warm Intro Machine with Marcus Bell

At 7:40 on a Tuesday, before most of the building has its coffee, Marcus Bell is reading a list of 63 promises. That is what he calls the request queue: every open ask from a portfolio founder, sorted by the date someone at the firm said “we’ll get on it.” The top row is a seed-stage founder in Austin who needs a first VP of Sales by October. The second is a Series B company that wants to meet three hospital systems before its board meeting. Marcus reads each row, moves two of them up, and writes one word next to a third: “Stale.”

The list is not a spreadsheet he maintains. An agent builds it overnight from the request form, the shared inbox, and the Slack channels where founders ask for help in less formal ways. It flags anything that has had no human touch in four business days. Marcus treats that flag the way a restaurant manager treats a ticket that has sat on the pass too long. “A founder asked us for something,” he told me. “Every day we don’t answer, we’re teaching them not to ask.”

I spent parts of three weeks with Marcus and his team in Menlo Park and San Francisco. The platform side of a venture firm is easy to describe and hard to run: help the companies you invest in hire people, find customers, meet each other, and get answers. Behind that sits a second job most founders never see, the operations behind deal intake and investor reporting. Marcus has both.

Platform Operations at Ashgrove is 14 people. They support the firm’s talent, go-to-market, and events teams, and they run the portfolio request desk directly. In July the desk logged 1,912 requests. About 41% were recruiting asks, 27% were customer or partner introductions, 18% were event logistics, and the rest were the long tail: a founder who needs a good immigration lawyer, a CFO asking which audit firms other portfolio companies use, a team that wants feedback on a pricing page.

When Marcus joined, the desk ran on a shared inbox and memory. Requests lived in whichever partner’s email they landed in. A founder could ask three people for the same introduction and get three different answers, or none. His first project was not an agent. It was a rule: every request gets one record, one owner, and one promised date, and the founder sees all three within a business day.

The agents arrived about eighteen months later, once the record was clean enough to trust. Today an intake agent reads every new request, classifies it, checks whether the same founder asked for something similar in the last 90 days, and proposes an owner. It writes the acknowledgment to the founder, which a coordinator reviews in batches twice a day. Median time to first response fell from 2.6 business days to 4.1 hours. The share of requests closed inside the promised date went from 58% to 89%.

The Tuesday request review in Menlo Park. Marcus Bell, second from left, reads the stale rows aloud before anything else.

If you ask anyone on the team what they actually make, they say introductions. A good one connects two people who both want the meeting, with enough context that the first call starts in the middle instead of at the beginning. A bad one costs the firm a favor with an executive it may need again next quarter. The team runs every introduction double opt-in: the person being introduced agrees before the founder ever sees their name.

Agents now do most of the preparation. For a recruiting request, a sourcing agent searches the firm’s talent network, which holds about 31,000 executives and operators who have opted in to hear about roles. It ranks candidates against the founder’s brief, removes anyone who joined a new company in the last year, and checks whether another portfolio company already has an open conversation with them. Then it drafts the outreach and the forwardable blurb, the short paragraph about the company that the candidate reads first.

Odalys Ferrer runs the talent network operations. Her team of four used to handle about 220 recruiting introductions a month. They now handle about 610, without new headcount. What changed is where their time goes. They spend almost nothing on search and formatting, and a lot more on the calls that decide whether a match is real: asking a candidate what they want next, or telling a founder that the brief describes two jobs, not one.

Customer introductions follow the same pattern with a different list. An agent matches a portfolio company’s target accounts against the firm’s network of corporate buyers, drafts the ask, and logs which executives have already received an introduction that month. Marcus set a hard cap: no buyer gets more than two introductions from the firm in 30 days, no matter how good the fit. The agent enforces the cap. A person has to approve any exception, and in July there were three.

The second half of Marcus’s job is less visible. Pitches reach the firm through every channel that exists: the website, cold email, partner referrals, demo days, LinkedIn messages, and the occasional printed deck in the mail. In an average week the intake operation sees about 1,450 new companies. Before the current system, an analyst could spend a full day matching a referral to a company the firm had already met under a different name.

Kwabena Osei built most of the intake pipeline. An agent reads each submission, extracts the company, founders, sector, stage, and location, and checks the firm’s CRM for prior contact. It catches renames, pivots, and second companies from repeat founders. Then it routes the record to the right investment team with a one-paragraph summary. The duplicate rate in the CRM dropped from about 11% of new records to under 2%.

There is one line Kwabena’s agents do not cross. They never decline a founder. They can sort, summarize, and route, and they can draft a reply, but a pass always comes from a person on the investment team. Marcus was firm about this from the start. “A founder will remember how we said no for longer than they’ll remember most of our yeses,” he said. “That message should come from someone who read the deck.”

Every quarter, the firm owes its own investors a clear account of the portfolio. That means collecting metrics from hundreds of companies that each report in their own format, on their own schedule, with their own definition of revenue. Hana Whitlock runs that process. When she started, the quarterly collection took her team of three about seven weeks, most of it spent sending reminder emails and reformatting spreadsheets.

Now a collection agent sends each company a request with last quarter’s numbers prefilled, follows up on a fixed schedule, and reads whatever comes back: a spreadsheet, a PDF board deck, or a paragraph in an email. It maps each figure to the firm’s template and flags anything that moved more than 25% from the prior quarter or disagrees with what the company said in its last board meeting. The collection now closes in 19 business days.

Hana keeps a list she calls “the arguments”: every number where the company’s figure and the agent’s reading disagree. Last quarter it had 74 rows. About two-thirds were definitions, such as bookings reported as revenue. The rest were real changes that the company had not yet explained. Each one gets a human conversation and a written note. No number reaches an investor report until someone on her team has signed it.

Marcus keeps a short document titled “Not for agents.” It has six lines. Agents do not decline founders. Agents do not send an introduction without a human read. Agents do not contact a candidate who has asked the firm to stop. Agents do not write anything that sounds like investment advice to a portfolio company. Agents do not reply to a founder who is in distress. Agents do not decide who sits next to whom at a dinner.

The last line sounds like a joke, and the team laughs when it comes up. It is not a joke. The firm runs about 140 events a year, from 12-person dinners to summits with more than a thousand guests. Agents handle invitations, dietary notes, travel, badges, and the follow-up survey. A person still does the seating chart for every dinner, because the value of the evening is two people who should know each other sitting close enough to talk.

He tests new automation the same way every time. Before an agent takes on a task, it runs against the last 30 days of real requests in a shadow mode, and someone on the team grades its output against what a person actually did. The bar is not perfection. The bar is that a person reviewing the agent’s work makes fewer corrections than they would on a new coordinator’s work in their second month.

The team hires from odd places. Odalys came from executive search. Kwabena was a data engineer at a logistics company. Hana spent six years in fund accounting. The common trait, Marcus says, is a specific kind of care: people who keep lists without being asked, and who notice when an item on a list has not moved.

The interview has one exercise that candidates remember. Marcus gives them a real, anonymized week of portfolio requests, about 50 rows, and the agent’s proposed handling for each. They have 40 minutes to mark which proposals they would approve, which they would change, and which they would take away from the agent entirely. He does not grade the number of corrections. He grades the reasons. A candidate who flags a technically correct introduction because the founder asked for the same person last month goes to the next round.

The team has added two people in the last year while request volume grew by about 70%. Marcus does not present that as a headcount win. He presents it as time. Each coordinator now spends about 11 hours a week on what he calls “the second call,” the follow-up with a founder after an introduction to ask whether it helped. Before the agents, the team made those calls for fewer than one request in ten.

On my last morning with the team, the queue had 58 rows and one of them was marked stale. It was a founder in Atlanta who had asked for help finding a head of clinical operations. The agent had found eleven candidates and drafted eleven blurbs. Odalys had held them, because the founder’s brief mentioned a regulatory filing that would change the role in two weeks, and she wanted to wait for it.

Marcus read the note, agreed with the hold, and then called the founder himself to explain the delay. The call took four minutes. He told the founder the date the introductions would go out and wrote it on the record. Then he moved on to the next row. It was the smallest thing I watched him do in three weeks, and I think it is the center of how he runs the team. The agents keep the promises organized. A person keeps the promise.