Playbook Fundraising

Playbook: Fundraising

All rules here are best practice, not law. Override any of them when the campaign calls for it; note the best practice once and move on.

Use when: the angle depends on the prospect having money to spend or a mandate to grow — either a TAM filtered to funded companies, or a first line that names their round.

Do not use when: you want headcount growth or open roles as the signal (playbook-hiring-surge). If you want the round as a pure list filter and no copy at all, you only need lane A in §3.

One-line output: funding_line = "Saw you raised $52M in the Series B." — the whole sentence lives in the variable. §2's downstream gate explains why that matters more than it looks.

1. Trigger and scope

Three separate questions get called "fundraising", and they route to different sources. Do not mix them up: one is a solved free problem and one is not solved at all.

• Lane A, ever raised. "Only show me companies with venture backing." A search filter applied when the list is built. Free, reliable, no per-row work.
• Lane B, the copy line. "Write a first line about their round." A per-row lookup producing funding_line. This is the main deliverable and the tested path.
• Lane C, raised in the last 30 days. "Who closed this month?" The people-database funding indexes lag 3 to 4 months, measured. Lane C has no proven end-to-end path and is documented honestly below rather than faked.

Not covered: valuation claims, investor-name personalization, and anything reported only as a rumor.

2. Output contract

Inputs required per row

Output fields

Abstain value: "". funding_confidence is low whenever the line is empty.

Coverage expectation

8/10 rows usable (80%), where "usable" counts a correct abstain on a company that genuinely has no recent round.

Read the second number too: only 2/10 rows produced a non-empty line, because the test set was adversarial — three bootstrapped companies and three whose last round is years old. On a list already filtered through lane A with a 365-day funding window, every row has a funding event by construction, so the non-empty rate is governed by the equity-stage filter and the 12-month window rather than by source coverage. That rate was not measured. Measure it on your first real campaign before letting copy depend on the variable being present.

Copy-fit rules

Two fields, two different jobs. Keep them straight.

• funding_clause is what the model writes: lowercase first letter, no trailing period, under 80 characters, 5th-grade reading level, no em dashes. It must read grammatically inside Saw <funding_clause>.
• funding_line is what gets pushed: deterministic code wraps the clause into "Saw " + funding_clause + ".". When the clause is "", the line is "" too. The model never writes this field, a formula does, so the wrapping cannot be forgotten or hallucinated.
• Neither field ever names a month, a season, a year, or a date, and neither names total funding, valuation, or investors. §7 explains why, with measured error rates.

Downstream gate — read this before wiring the email body

⚠️ Spintax is a RANDOM chooser, not a conditional.

{Saw {{funding_line}}. |} picks one of its branches at random on every row, with no knowledge of whether the variable is populated. On rows that abstain it renders Saw . about half the time, and on rows that DO have a line it throws the personalization away about half the time.

Spintax can never be used as an if-populated gate for any variable, in this playbook or any other. This generalizes past fundraising and is probably the single most useful thing in this file.

There are exactly two mechanisms that work. Pick one per campaign:

1. Default: the self-erasing variable. Put the entire sentence in the variable — which is why funding_line carries the leading "Saw" and the trailing period. The body line is just {{funding_line}} on its own line. An empty variable renders nothing, so an abstaining row simply loses that sentence and the rest of the email is untouched. QA step: after upload, preview one abstaining lead and confirm the paragraph collapses cleanly without leaving a double blank line.
2. Split at upload time. Partition the list into populated and empty, and upload two campaigns or two variants — one with the funding sentence hard-coded into the body, one without. Use this whenever the funding angle changes the whole email rather than one line, and whenever the non-empty rate is low enough that a single body would read oddly for most recipients.

Only exclude abstaining rows entirely when the campaign angle is the round.

The usual word-level spintax block for wording variety is unrelated and still required. Spintax is for wording variety, never for conditional content.

3. Source chain (cost-tagged)

Lane A, ever raised / raised within 6 to 12 months (list building)

Equity stage set: Pre seed, Seed, Series unknown, Series A through Series E-J, Angel, Corporate round, Convertible note, Equity crowdfunding.

Lane B, the copy line (per row) — THE TESTED PATH

⛔ There is no step 5. If the lookup returned no company record at all, the row abstains and the line is empty, which is the correct outcome for a signal you cannot evidence. Do not add a paid company-enrichment fallback: it costs roughly 8 credits per company and does not carry funding fields anyway.

Lane C, raised in the last 30 days — NOT SOLVED

Rejected alternatives

• A people-database funding-date filter for lane C. Measured: a 90-day window over the entire US Software Development market returned total_count = 1. The index lags 3 to 4 months.
• Semantic "entity search" APIs for lane C. 0 of 20 returned companies had a round inside the requested window. It returned public companies, companies its own payload labelled Unfunded, and one labelled Acqui-Hired. Entity search ranks on static attributes and has no event-date awareness, so a date-bounded objective degrades into "companies that look fundable".
• total_funding as a copy fact. One company came back at $550.9M where the press reported $452M, because these totals fold in debt facilities and secondary sales.
• raised_at as a copy fact. Off by 15 days for one company, one month for another, two months for a third.

4. Verification

VERDICT: PASS 8/10 (80% usable) | claims 8/8 (100% correct) | chain = company lookup by domain → domain-equality guard → equity-stage + 12-month filter → model line writer | p50 3.0s/row | ~$0.22/1k.

This is a claim-bearing playbook, so it carries two bars: 70% usable AND 90% claim correctness.

The 90% is coverage correctness, not coverage — scored as true claims / rows where a claim was emitted. A company that never raised, correctly abstained on, is a correct outcome and leaves the denominator entirely. It is not a miss. The bar binds on truthfulness when the signal exists, because a fabricated round is the one failure nobody can fix after it sends.

Both misses above were coverage misses (no record for one domain; a two-month date error pushed one round outside the window), not false claims — which is why claim correctness is 8/8.

⚠️ Lane A's end-to-end list quality was not graded. Lane C is untested end to end and must not be described as working.