A Signal Is Not Binary
The difference between an event and a signal
An event is a raw fact: “Company X posted a job offer.” An intent signal is that event interpreted, filtered and put back in context for your specific offer: “Company X, a 150-person consulting firm in the Paris region, posted four senior developer roles at once, which points to fast growth and, possibly, a need for a project management tool.” If the term is new to you, start with our definition of an intent signal.
Between the raw event and a lead your sales rep can actually contact, there is configuration. It decides whether a piece of news becomes an opportunity or noise.
The same event means different things to different sellers
Take a job offer for a sales director at an 80-person company in Lyon.
- For a CRM vendor, it is a strong signal: the new director will review the team’s tools.
- For a recruitment firm, it is an even better one: the role is open, the need is immediate.
- For a health insurer, it is a weak signal: one hire barely changes its contract.
- For a prospecting agency in Lille that only works with mid-size enterprises, it is not a signal at all.
One event, four readings. That is why a signal is not “switched on”: it is configured for a given offer.
The Families of Settings
The available filters depend on the signal type, but they fall into a few broad families. Rodz produces over 100 types of intent signals, and each one exposes the filters that make sense for it.
The company scope
The foundation of every signal: which companies do you want to hear about?
- Industry: industries to include, industries to exclude (for example recruitment and staffing agencies when you track job offers).
- Company size: headcount ranges to include or exclude.
- Location: headquarters country, region or city, with optional exclusions.
- Company lists: domains or websites to exclude, such as your current customers or your competitors.
The event itself
These filters apply to the event, and they change from one signal to the next:
- Job offers: keywords in the job title, keywords to exclude, job location.
- Fundraising: minimum amount, funding stages, investors to exclude.
- Company registration: NAF code (French business activity code), legal form, location.
- Public tenders: keywords in the contract subject, CPV codes, amount, geography.
- Social mentions: the keywords or phrases to monitor.
The person to contact
A signal is about a company; your rep writes to a person. Persona targeting defines who to receive for each event:
- Target job titles (CEO, HR director, sales director, CFO…) and titles to exclude.
- Seniority level.
- Location of the person, with optional exclusions.
Pace and volume
Two simple settings have a direct effect on your day-to-day:
- Daily lead limit: the maximum number of leads the signal delivers per day. It protects both your team’s capacity and your credit budget.
- Deduplication period: after how many days the same company or person can become a lead again for this signal (never, after N days, or always).
Too Narrow or Too Broad: The Real Tuning
Most poorly configured signals fail in one of two directions.
Too broad. A “sales” job offer, anywhere in the country, any company size: you receive hundreds of leads, most of them outside your target. Reps sort instead of prospecting, and trust in the signal collapses within two weeks.
Too narrow. An exact job title, a single city, a tight headcount range, plus three exclusions: you receive almost nothing and wrongly conclude the signal “doesn’t work”. An overly restrictive keyword misses a large part of the market, because companies don’t word their job ads the way you imagine.
The right approach is to start slightly broad, read the first leads, then tighten. Every exclusion should answer a false positive you actually observed, not a theoretical worry.
Validating Your Keywords Before You Go Live
Keyword configuration looks straightforward until a homonym slips through. The word “orange” can mean a colour, a fruit, or a major telecom operator. The acronym “IMS” covers at least two meanings in a B2B context: Information Management System and AI Management System, which in English also surfaces as “AIMS”. A keyword that seems unambiguous in isolation often isn’t once it hits real job postings or company descriptions.
This is the step most teams skip, and it is where a lot of false positives are born.
The ambiguity problem
Before activating a keyword-based signal, run a quick check:
- Does the term have a second meaning in a different industry? If you sell to logistics companies and you filter on “flow”, you will also catch finance teams talking about cash flow.
- Does the acronym expand in more than one way? Write out every expansion you know. Then search the term on a job board and read the first ten results that come back. If more than one or two are off-target, the keyword needs tightening.
- Does the term change meaning across geographies? A word that is sector-specific in France may be generic in Belgium or Switzerland.
Using an LLM to catch what you miss
A fast and practical method: paste your keyword list into a general-purpose language model and ask it to list every industry, context or meaning in which each term commonly appears. Ask specifically for cases where the term could match a company or role outside your target. A prompt like “List every professional context in which the phrase [X] is used, including sectors I might not expect” tends to surface the edge cases you have not thought of.
This is not a substitute for reading actual leads, but it is a useful pre-flight check that takes five minutes and catches the obvious ambiguities before you spend credits on them.
A QA loop before production
Once you have a keyword set you are reasonably confident in, run a small batch before setting a high daily limit:
- Set the daily lead limit to 5 or 10.
- Let the signal run for two or three days.
- Read every lead that comes through and flag the ones that are off-target.
- For each off-target lead, ask: which keyword matched, and why? Then either exclude that term or add a negative keyword that would have filtered it out.
- Once the false-positive rate on this small batch is acceptable, raise the daily limit.
The logic here is the same as a staged rollout in software: catch errors while the cost of a mistake is still low. A signal running at 50 leads per day with a 30% false-positive rate burns through credits and rep time. Five minutes of QA at 10 leads per day prevents that entirely.
This matters most for keyword-heavy signals: job offers, public tenders, and social mentions. For signals that rely less on free-text matching (fundraising rounds, company registrations), the ambiguity risk is lower and the QA loop can be shorter.
Configuration Examples
SaaS vendor targeting scale-ups after a funding round
| Setting | Value |
|---|---|
| Signal | Fundraising |
| Minimum amount | €2M |
| Industry | Software, internet |
| Size | 20 to 500 employees |
| Location | France, Belgium, Switzerland |
| Persona | CTO, VP Engineering |
Generalist recruitment firm
| Setting | Value |
|---|---|
| Signal | Recruitment campaign |
| Excluded industries | Recruitment, staffing |
| Size | 50 to 2,000 employees |
| Location | Paris region |
| Persona | HR director, Talent Acquisition |
Accounting firm targeting new founders
| Setting | Value |
|---|---|
| Signal | Company registration |
| Legal form | SAS, SASU |
| Location | Your area and neighboring ones |
| Persona | Founder, managing director |
| Daily limit | Matched to the team’s calling capacity |
These examples are starting points, not recipes. The right configuration is the one that produces leads your reps actually want to contact.
Configuring With Method
- Start from your offer. Complete the sentence “I want to contact a company when it…”. The answer points to the signal type.
- Look at your best customers. Which events came before their purchase? What size, industry, contact?
- Configure one or two signals, slightly broad, with a reasonable daily limit.
- Validate your keywords before going live: check for ambiguous terms, run a QA batch at low volume, and tighten before raising the daily limit.
- Read the first leads and note the false positives: industry outside your target, wrong size, wrong contact.
- Tighten with targeted exclusions, then add more signals once the first ones are stable.
Timing matters as much as tuning: a well-configured lead contacted a week later loses most of its value. According to research on B2B buying windows, response speed is one of the strongest predictors of conversion. Rodz recommends acting within 48 hours of detection, the window where reply rates run 4 times cold-outbound levels.
In the Rodz app, configuration is guided step by step, with the filters specific to each signal type. To automate it, see the tutorial to configure your first signal via the API and the complete Rodz API reference. For prioritization, see the Balance scoring model.
Frequently Asked Questions
Do I need to configure everything myself?
No. Each signal type comes with sensible defaults, and the interface guides you step by step. Start with the essential filters (industry, size, location, persona), then refine based on the first leads you receive.
Can I change a signal along the way?
Yes, and it is recommended. A signal is tuned like an instrument: tighten the filters when too many leads are off target, widen them when volume is too low. Changes apply to the leads that follow.
How many filters should I turn on?
As few as possible to start. Every filter you add reduces volume. A good signal combines a few scope filters (industry, size, location) with clear persona targeting, then exclusions added as you observe false positives.
Am I billed for an off-target lead?
Rodz removes false positives before delivery and only bills qualified leads. A good configuration is still the best way to receive leads that are useful for your offer.
How do I know if a keyword is causing false positives?
Read the first batch of leads at low volume (5 to 10 per day) before scaling. For each off-target lead, trace which keyword matched and why. If a term is consistently pulling in the wrong companies, either remove it or add a negative keyword that excludes the unwanted context. An LLM check before you start, as described above, catches the most common ambiguities before you spend any credits.