Signal Detection

This guide walks through how to use the signals server to qualify accounts. It covers when to use detect_signal vs. individual detectors, how to tune set_signals_order, and how to interpret each signal.

When to use detect_signal

detect_signal is the recommended entry point. It dispatches every detector in the configured order and returns the union of fired signals. Dispatch is free — only the detectors that actually run are charged.

$
curl -X POST https://api.gtm-tools.sh/api/v0/detect_signal \
>
  -H "Authorization: Bearer $GTM_TOOLS_API_KEY" \
>
  -H "Content-Type: application/json" \
>
  -d '{"domain": "gymshark.com", "techs": ["zendesk.com", "intercom.com"]}'

Use detect_signal when:

• You’re qualifying a fresh list of domains and want every available signal.
• You want a single round-trip per company.
• You haven’t yet figured out which signals matter most for your ICP.

Use individual detectors when:

• You only care about one specific signal (e.g. you only sell to companies hiring SDRs).
• You want to short-circuit on the first hit and avoid running the rest.
• You’re A/B testing detector ordering.

Configure detector order

set_signals_order controls the order detectors run inside detect_signal. Front-load the cheapest, highest-precision signals so you can act on the first hit.

$
curl -X POST https://api.gtm-tools.sh/api/v0/set_signals_order \
>
  -H "Authorization: Bearer $GTM_TOOLS_API_KEY" \
>
  -H "Content-Type: application/json" \
>
  -d '{
>
    "order": [
>
      "signal_hiring_sales_rep_repost",
>
      "signal_trustpilot_negative_support_reviews",
>
      "signal_hiring_support",
>
      "signal_socials_spike",
>
      "signal_technologies_identified"
>
    ]
>
  }'

View the current order any time:

$
curl https://api.gtm-tools.sh/api/v0/get_signals_order \
>
  -H "Authorization: Bearer $GTM_TOOLS_API_KEY"

Signal interpretation

Hiring signals

Trustpilot signals

Negative-support reviews are uniquely actionable: the customer is pre-qualified, the pain is named, and you can quote the review in your outreach.

Marketing signals

Tech-stack signals

signal_technologies_identified is the only signal that needs configuration up-front. Pass techs as an array of identifiers to match against the website.

$
curl -X POST https://api.gtm-tools.sh/api/v0/signal_technologies_identified \
>
  -H "Authorization: Bearer $GTM_TOOLS_API_KEY" \
>
  -H "Content-Type: application/json" \
>
  -d '{
>
    "domain": "gymshark.com",
>
    "techs": ["zendesk.com", "intercom.com", "gorgias.com", "freshdesk.com"]
>
  }'

1
{
2
  "domain": "gymshark.com",
3
  "matched": ["zendesk.com"],
4
  "evidence": [{ "tech": "zendesk.com", "source": "html", "snippet": "..." }]
5
}

This is how you confirm a prospect is on the competitive stack you displace.

Scoring + ranking

The signals server returns raw fires; ranking is your job. A simple weighted score:

1
const WEIGHTS: Record<string, number> = {
2
  signal_hiring_sales_rep_repost: 5,
3
  signal_trustpilot_negative_support_reviews: 4,
4
  signal_hiring_support: 3,
5
  signal_socials_spike: 2,
6
  signal_technologies_identified: 2,
7
  signal_hiring_role: 1,
8
};
9
10
function score(detected: { name: string; fired: boolean }[]): number {
11
  return detected
12
    .filter(s => s.fired)
13
    .reduce((acc, s) => acc + (WEIGHTS[s.name] ?? 1), 0);
14
}

Then sort target accounts by score and prospect top-down.

Token budget

A full detect_signal run with all 9 priced detectors firing costs 45 tokens (9 × 5). In practice, most companies trigger 2–4 signals, putting the median around 15 tokens per account. That’s $0.15 per qualification.

Next steps