Next experiment
Killed. Bid Scout / public-bid filter failed CRO’s filter: cold trust
- competing with free/cheap BidNet-class alerts. Do not run this test. Experiment 2 is doc-to-data.
Archive only. This was Experiment 1: RFP / tender scout, one week, one supplier niche. The question was whether operators would pay for filtering, not whether a bot can find RFPs. They would not, at a time EV Walter will repeat.
This was the first test of a pre-award signal business. It is closed. The remaining month slot is one ops loop — doc-to-data if a named warm partner exists. Do not run five experiments in parallel.
Hypothesis
Suppliers in one niche already miss or over-read public RFPs and tenders. A daily brief of opportunity | why fit | disqualifier is worth a paid pilot if and only if they treat the filter as the product.
If they want the bot to write the proposal, that is a different business — and not this test.
Setup (day 0)
- One supplier niche — capabilities you can write as a short fit checklist (what they do, where they work, what they will not bid).
- Watch list — the public bid / tender sources that niche actually uses. Keep it finite. Do not “scrape the whole internet.”
- Five firms — real suppliers in that niche who will receive the briefs. Not a waitlist. Not a Twitter audience.
- Brief format (locked):
- Opportunity — what it is, who the buyer is, due date / link
- Why fit — mapped to the checklist
- Disqualifier — the reason they should skip (or “none; still a maybe because …”)
- Cadence — daily on weekdays for one week. Empty days are allowed; do not pad with junk to look busy.
No invented volume targets. Success is judgment from the five firms, not a made-up open rate.
Days 1–5
Each day:
- Watch the sources.
- Enrich anything that matches the niche at a glance.
- Decide fit / skip using the checklist.
- Send only the survivors, each in the three-part format.
- Escalate close calls in a separate “maybe” line — do not hide uncertainty inside a fake yes.
Keep a log of: notices seen, notices dropped, briefs sent, and any firm reply (useful / noise / wrong disqualifier). Counts stay in the log; do not turn them into a market story.
What “they value filtering” means
A paid pilot is in play only if at least some of the five firms say, in substance:
- they would stop or reduce their own portal scanning if this kept running, and
- they would pay to keep the filter, not to get proposal drafts, and
- wrong disqualifiers and missed fits are correctable by tightening the checklist (the loop is tunable).
Praise without willingness to pay is a no. “Send more” without naming false positives is a no. “Have it write the bid” is a no for this experiment.
Kill / continue
| Outcome | Do |
|---|---|
| Firms ignore the briefs or treat them as a free newsletter | Kill this niche test. Do not expand sources. |
| Firms argue with the disqualifier line (good: they are using the filter) | Tighten the checklist; consider a second week on the same niche. |
| Firms ask to pay for the daily filter | Paid pilot. Keep the format. Do not add proposal-writing. |
| Firms only want more volume | You are not selling filtering. Stop. |
Post-mortem: the signal test died on cold trust + a free/cheap firehose. The remaining month slot is doc-to-data, not a sixth content idea. The kill list stays killed.
Not this week
- Cold-calling 50 people to “get feedback on AI.”
- Multi-niche coverage.
- Auto-submitting bids.
- Pairing wholesale real estate as the primary bot test (weaker leverage; see wholesale-re).
Sources
- ChatGPT Deep Research export — TBD from Google Drive
Research/Inbox. - CRO-approved experiment (now killed): one supplier niche, daily opportunity | why fit | disqualifier briefs, five firms, paid pilot only if they value filtering.
- Document-to-data Experiment 2