Signal sourcing

Firmographics decide which accounts. Signals decide why now, and why-now is what makes a rep pick up the phone in the two weeks before a show. Read this before the first batch of connector calls — every trap below costs hours if you rediscover it live.

The signal hierarchy

Get as many as the connectors allow, in this order of value:

1. Tech-stack tags for the vendor's own product and for every named competitor. The single highest-value query in the pipeline. Own product → existing customer, routed to the expansion track. Competitor → displacement target, scored.
2. A dead or dying competitor. Search for this explicitly, every time, by name: shut down, acquired and sunset, end-of-life, no longer sold. Its install base must migrate and the deadline is not yours to create. Nothing else on this list converts like it.
3. Existing-customer flag — the vendor's tech tag OR the logo on the customers page. Two sources because both are incomplete; a logo page lags churn and a tech tag misses private deployments.
4. Scoops, news, funding rounds, exec hires — Tier 1 only. Rich, slow, and not worth pulling for 400 accounts.
5. Intent topics — if the account has them enabled. Many do not. Check rather than assume, and record the coverage rate in the method section.

Connector traps

Contact search with a title filter usually accepts only ONE company at a time. Pass eight company IDs together with a job-title list and you get zero rows back — not an error, not a warning, an empty result that looks like the accounts simply have no matching people. Two workarounds, both far better than fifty single-company calls:

• Filter by department (Information Technology, for example) plus management level. This does accept many company IDs at once.
• Filter by job function, which is more precise. Look the function IDs up first, then batch about eight company IDs per call with a page size of 100 and filter titles yourself afterwards.

Run both passes and merge. Department catches the executives; job function catches the hands-on owners who will actually be at the booth.

Look up enum values before filtering. Industries, departments, job functions, tech products, and metro regions are controlled vocabularies. Guessing returns either an error or — worse — a silently empty result. Note whether the field wants the id or the name; it differs field to field within the same connector.

Tech-product lookups usually need the vendor name first, then the products belonging to that vendor. One call per vendor. Do this for the vendor's own product and every competitor named on their site, before scoring.

Big results overflow the context window. A batched contact search returns on the order of 65,000 characters. When the harness saves an oversized result to a file, do not read it back in — write a small script that parses every saved result file and keep the raw payloads out of the conversation entirely. The pipeline is designed as separate steps precisely so this stays possible.

Know which calls cost credits. Search, lookup, and scoops are typically free; enrich and research calls are not. A "no credits" answer rules out every enrich-style call and every find-and-enrich tool, and it means no emails and no phone numbers anywhere in the deliverable. Confirm at the end that you spent none — an explicit "zero credits spent" line in the method section is worth writing.

Match rates and what to do with misses

Expect roughly 90% of accounts to resolve against a company-data connector. Carry the missing 10% through the whole pipeline with blank firmographics rather than dropping them: an unresolved row is usually a small private company, a regional division, or a name typed into a registration form by hand — and a named attendee from an unresolved company is still a named attendee in the room.

Never let a blank become a zero. Blank size, blank industry, and blank tech tags each need to read as unknown in the score and in the file, because the fix for unknown (ask a human, search the web) is different from the fix for low.

What to write down about coverage

Signals are worth exactly as much as a reader's confidence in them, and that confidence comes from stating the gaps. For each signal type, record:

• how many accounts you queried, and how many returned anything
• which competitors you had tech tags for and which you did not
• whether intent was available at all
• what you searched for and did not find — especially the dead-competitor search, since a null result there is a real finding

A plan that says "eleven accounts run a competitor that reached end-of-life, tech-tag coverage was 62%, so treat the absence of a tag as unknown" is trustworthy. The same plan without those numbers is a spreadsheet of guesses with good formatting.