What to automate in a fund's referral workflow: a boundary map for fourteen tasks, the four no tool should touch, and a payback test with referral income set to zero

Short answer

Automate the clerical layer of a fund's referral workflow: merging company lists, stamping the terms version on outbound notes, internal reminders, status checks and first drafts built from fields the fund already holds. Keep judgment with a person: reading fit, choosing who asks the CEO, and rewriting anything a founder will read. Never automate the ask itself, a send without a recorded yes, anything that touches company records, or the reward disclosure. Justify each automation by platform hours saved with referral income set to zero; at pilot volumes, a spreadsheet usually wins.

Visual briefFull-size diagram
A four-column boundary map sorting referral tasks into automate, draft and approve, human only, and never; the never column holds the ask to the CEO, auto-sends, company records and the disclosure text, with an Illustrative payback of 3.4 quarters at 40 companies.
Illustrative. Task placement follows the boundary map on this page; the payback figure comes from its zero-revenue hours table.

Platform attention is the scarce input, and automation can spend it faster

For a venture platform team, the founder benefit of a data licensing referral is narrow and real: a CEO learns whether the company's own records could be licensed for AI training and evaluation, and gets an introduction to SourceX only if they want one. The constraint is attention on both sides. Platform hours go into screening and briefing; founder attention goes into every message the fund sends.

Automation saves the first and can burn the second. A template that once went to three founders a quarter can go to thirty in an afternoon, and its mistakes scale with it. The program sets edges no tool can see past: SourceX's screening usually marks an introduction as not a fit when the decision maker has not been told about it, and credit is settled from dated records, written introduction evidence among them, never link clicks, with the earliest valid introduction in the 90-day attribution window credited after the company has submitted a verified application.

The automation boundary map

Four boundaries, applied to fourteen tasks. "Automate" means the tool acts and a person spot-checks. "Draft, a person approves" means nothing reaches a founder or SourceX until a named person has edited and approved it. "Human only" means a person does the work and the tool, at most, records the date. "Never" means no tool does it under any configuration.

TaskStageBoundaryWhy it sits thereWhat the tool may touch
Merge company lists from the CRM, portfolio sheets and LP reports into one row per companyFindAutomateMechanical; errors show up as visible duplicatesNames, domains, relationship-owner field
Pre-fill headcount, founding year and location from data the fund already holdsScreenDraft, a person approvesFund data shows current headcount, not peak, and rarely separates contractorsFields the fund already holds
Flag dated events such as a system migration, sale or wind-downScreenHuman onlyUsually learned in a board seat or founder conversation, whose owner decides if it may be usedA person enters the event and its source
Choose who asks the CEOAskHuman onlyRelationship scores built from email and calendar activity measure contact, not permissionRead-only suggestions at most
Ask the CEO whether they want the conversationAskNeverThe program depends on the decision maker knowing about and wanting the introductionNothing
Send to a founder because a score or status changedAskNeverAn automatic trigger is an introduction nobody agreed toNothing
Record the CEO's answerAskHuman onlyConsent is what the CEO said, entered by the person who heard itThe timestamp on that person's entry
Draft the introduction from the row's fieldsIntroduceDraft, a person approvesSaves typing; the sender still writes the company's own reasonRow fields, never confidential notes
Write or reword the reward disclosureIntroduceNeverFixed text, written once from the terms and inserted unchanged; a paraphrasing tool can misstate payer, base or the minimumInsertion of the fixed text only
Submit the introduction through the partner linkIntroduceHuman onlyThe partner who signed the Channel Partner Agreement submits, after checking the recorded yesNothing
Request or move files, samples, exports or loginsAnyNeverPartners never handle company records or credentials, by any routeNothing
Check each introduction's statusTrackAutomateSourceX's portal shows the stage and connects to AI assistants over MCP for questions about your own referrals, link and current termsThe partner's own portal data
Follow up with a founder who has not repliedTrackDraft, a person approvesEach nudge spends founder attention; one follow-up, then stopThe draft only
Stamp the terms version and check date on outbound notesAllAutomateTerms change; Version 2026-10-22.2, since replaced by version 2026-11-04 was current when checked on 2026-10-26A comparison of the version line

The four "never" rows reduce to four sentences a platform lead can put on a wall. No tool asks a CEO. No tool sends to a founder on its own trigger. No tool touches company records or logins. No tool writes or rewords the disclosure. The disclosure rule has an outside anchor as well: the FTC's endorsement FAQ treats a financial relationship as a material connection and asks for a disclosure that is clear and hard to miss, so the line a founder reads should be the one a person wrote and checked, every time.

Status checks are the cleanest automation here, because the data is the partner's own; the stage view is described in where a partner sees the current stage, and introduction records without unnecessary personal data keeps the fund's log lean. Relationship tools that rank "who knows this CEO best" from CRM activity, as in the Affinity MCP guide, stay in the suggestion column: the strongest contact is not always the person entitled to ask.

Hours at each stage with referral income set to zero (Illustrative)

Automation has to pay for itself in platform hours alone. The default reward on a company's first qualifying transaction, the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap, is payable only once the Acquisition Partner has given final written acceptance of the complete, validated dataset, the handoff is complete, its payment has cleared, the company itself has been paid, SourceX finance has approved and the partner's tax and payment details are in. The fund controls none of that, so the base case books no referral income; a tool that only makes sense if rewards arrive is a bet.

Illustrative assumptions, in minutes per company screened: merging lists takes 6 by hand and 1.2 of spot-checking once automated; pre-filling fit fields takes 9 by hand and 3 of review; checking the terms version takes 1.8 by hand and none once automated; a status check takes 3 by hand and 0.6 automated. One company in five reaches a drafted introduction, which takes 18 minutes by hand and 12 to rewrite a machine draft. Net hours saved equal manual hours, minus the hours left after automation, minus upkeep.

TaskBuild hours, onceUpkeep hours per quarterNet hours saved per quarter at 40 companiesNet hours saved per quarter at 10 companies
Merge and deduplicate company lists60.52.70.3
Pre-fill fit fields for a person to approve101.03.00.0
Stamp the terms version on outbound notes20.50.7-0.2
Status checks through the partner portal30.31.30.1
Draft introductions for the sender to rewrite60.50.3-0.3
Total272.88.0-0.1

At 40 companies a quarter, 27 build hours come back in about 3.4 quarters (27 divided by 8.0), roughly ten months. At 10 companies a quarter, upkeep swallows the savings and the build never pays back. That is the base case's stop example: a fund running a ten-company pilot should build nothing, and revisit only if its own minutes turn the table positive. Drafting is the weakest row at both volumes, as the map predicts: the costly part of an introduction is the company's reason, and no draft writes that.

Founder time is left out of the savings column on purpose, so no automation gets credit for time it shifts onto a founder. Platform hours per qualified introduction are modeled stage by stage in whether data referral work is worth a platform team's time.

When to expand, pause or switch the automation off

Set these readings before launch and check them each quarter. The thresholds are starting points for a fund to adjust, not published benchmarks.

Reading at the quarterly reviewExpandPauseSwitch off and return to manual
Net hours savedAbove plan two quarters runningWithin a quarter of planZero or negative
Errors that reached a founder (wrong name, stale terms, duplicate send)NoneOne, caught and fixedAny repeat
Founder opt-outs or complaints after an automated draft went outNoneOneMore than one
Pre-filled fit fields overturned by the reviewerFewer than 1 in 101 in 10 to 1 in 4More than 1 in 4

Opt-outs deserve their own rule. For commercial email, the FTC's CAN-SPAM guide requires an opt-out to be honored within 10 business days and the opt-out mechanism to keep working for at least 30 days after the message is sent. Whether a particular platform note is commercial is a question for counsel (CAN-SPAM and one-to-one B2B email sets out the issues), but one suppression list across every program the platform runs, honored inside that window, removes the question from day-to-day work.

A switch-off case (Illustrative). A platform team wires its CRM to send the introduction template whenever a company's fit score crosses a threshold. In week one, three founders receive it without having agreed to anything, one of them a CEO whose relationship belongs to a partner who was never told. The fix is not a better threshold: turn the trigger off, have each relationship owner apologize personally, mark the rows "CEO not asked" and confirm nothing was submitted to SourceX. No hours saved could justify it, because it crossed a "never" line on its first run.

What stays with a person when a company looks like a fit

The fit read is human work because the published tests need interpretation. Each introduced company has to have reached 50 or more full-time employees at its peak, contractors left out of the count, and needs an operating history on paper, rights to license and an authorized contact: the CEO, who authorizes a licensing discussion and signs. For a staffing firm, only its own full-time employees count, not placed workers on assignment at clients; describe both numbers and introduce the firm anyway if its records are strong. Companies in the U.S., Canada and the EU qualify, as do companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 70% of the data is in English. Funding status never settles eligibility either way.

No tool on the fund's side can resolve whether the records match SourceX's current data needs, whether client contracts restrict them, or whether SourceX already knows the company. Partners never ask for files, samples or logins, so automated fields hold metadata at most: what exists, roughly how much, over which years, in which system, owned by whom. Where a portfolio company's own business is fixing automations, its failed-run records are a separate topic, covered in failed automation reviews.

Where this map can be wrong

  • Headcount data. Pre-filled figures are usually today's headcount from the fund's records, not peak full-time headcount, and they rarely exclude contractors. A reviewer overturning many pre-fills is the signal.
  • Credit. SourceX treats any earlier lead, account or email correspondence with a company as an existing relationship, and checks that in records the fund cannot see, so a clean dedupe on the fund's side says nothing about credit. Click logs supplement, but never replace, the dated written introduction.
  • Volume. The payback rests on Illustrative minutes and volumes; uneven screening or messier fund data moves the break-even.
  • Demand. SourceX checks relevance to its current data needs during review, and those needs change. A workflow tuned for volume can outlast the demand it was built for, so count completed reviews, not only sends.
  • Tool behavior. This page makes no claim about any CRM's or assistant's features beyond SourceX's own portal description. Check your vendor's documentation before relying on a capability.

Methods note. The boundary map applies SourceX's referral terms (Version 2026-10-22.2, since replaced by version 2026-11-04), its who-qualifies page and its partner site, plus the FTC's CAN-SPAM and endorsement guidance, all checked on 2026-10-26. The hours table uses stated planning assumptions; nothing in it was measured. Not measured: founder time, error rates in real deployments, reward income, and how often SourceX's review agrees with a fund's fit read.

Next step

Refer a company. No cap on rewards. Default reward on a company's first qualifying transaction: the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap; further qualifying transactions first agreed within 12 months of the first Closing pay 15%. Eligibility and payment conditions apply, including the USD 100,000 Minimum Transaction, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.

Once the four "never" rows are written into the team's process and a CEO has said yes in person, request a partner link, compare the company with the criteria set out on who qualifies and read the full program terms that the disclosure line summarizes. The portfolio-wide view of the program is on the venture platform teams page.

Automating a referral workflow and assessing SourceX as a company are different pieces of work. Readers doing the second can request investor materials through the SourceX investor overview.

This article is for information only and is not an offer to sell or a solicitation of an offer to buy securities. Any offering by SourceX is made only to verified accredited investors through its offering documents.

  1. Step 1Share your linkSend your personal link to a company you know.
  2. Step 2Company appliesThe company applies itself at /apply.
  3. Step 3Company signs and is paidThe company signs, delivers its data and is paid by SourceX under its agreement.
  4. Step 4You get your rewardthe greater of $50,000 plus 5% and 15% of the actual supplier payout, with no cap. A reward becomes payable after SourceX has paid the referred company under its signed agreement and the payout conditions in the program terms are met, including finance approval and your tax and payment details.

Common questions

Can an AI assistant run the whole referral workflow for a platform team?

It can help with the clerical parts: merging lists, drafting from fields the fund already holds and checking status in the partner portal. It should never ask a CEO, record consent on someone's behalf, touch company records or reword the reward disclosure.

Should a fund build automation before its first referral pilot?

Usually not. In the Illustrative model on this page, automation built for ten companies a quarter never pays back its build hours, while the same tooling at forty companies a quarter pays back in about 3.4 quarters. Run the pilot on a spreadsheet and decide afterwards.

Does an automated click log prove who gets referral credit?

No. The terms say a link click alone does not decide credit; attribution comes from dated records, including written introduction evidence. Keep the sent introduction itself, not just a tracking event.

Sources

  1. refer.sourcex.si/terms
  2. refer.sourcex.si/who-qualifies
  3. refer.sourcex.si
  4. ftc.gov/business-guidance/resources/can-spam-act-com…
  5. ftc.gov/business-guidance/resources/ftcs-endorsement…
  6. sourcex.si/investors

Free resources

Know a company with valuable proprietary data?

Become a referral partner, get your link and introduce the company’s CEO (in an owner-run company, usually the owner). Referral partners can join from most countries. We can't work with partners in countries or regions under U.S. sanctions or on U.S. restricted-party lists. The companies you introduce should be based in the U.S., Canada and the EU, or in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia with at least 70% of their data in English.

Refer a company →

I own a business

Explore licensing your company's data directly to SourceX for AI training and evaluation. Start a short assessment; no uploads needed.

Start an assessment