Real-time signals help only when the response fits the signal
AI-assisted research. This note was prepared with the assistance of AI tools. It may contain errors and is not professional advice. See our Legal Disclosures.
Agents can act on new information within minutes, which makes the old batch rhythm of marketing data look slow. Overnight syncs, weekly score updates, and monthly reports suited people who worked on similar schedules, but they leave an agent waiting on yesterday's picture.
Real-time signals close that gap. Consider a high-intent pattern, such as several people from one target account reading implementation documentation on the same afternoon. It could reach the account owner as a research brief before the end of the day.
That speed is useful. It also introduces risks that batch processing never had to manage, because a fast response to a weak or misread signal is still a bad response.
This post covers what counts as a signal and how to move from a stream of events to a sensible action. It then sets out the guardrails that keep speed from turning intrusive, and how to measure whether any of it helps.
What counts as a signal
Signals are events that suggest a change in a buyer's or customer's situation. The common sources fall into five groups:
- High-intent page visits: pricing, implementation guides, security pages, or comparison pages.
- Product usage changes: milestones reached, features adopted, or drops in usage.
- Direct engagement: replies, meeting bookings, and event registrations.
- Support activity: new tickets and the topics they raise.
- Public company events: funding, leadership changes, or expansion news, where appropriate to use.
Not every signal deserves a response. Most single events are noise. One pricing page visit might be a student, a competitor, or an existing customer checking an invoice. Much of the design work lies in deciding which signals and combinations carry enough meaning to act on.
From stream to action
A practical real-time setup has four stages. Keeping them separate makes each one easier to test and adjust:
- Collect: events flow from your website, product, and other systems into a central stream.
- Interpret: rules or an agent assess whether an event, or a combination of events, is meaningful.
- Decide: the system chooses a response, such as notifying a person, preparing a brief, adjusting a sequence, or doing nothing.
- Act and log: the response is carried out within the agent's permissions and recorded.
The signal data flow diagram shows these stages in context. The mechanics of triggers, queues, and retries behind them are covered in event-driven orchestration.
Interpretation is where agents add the most. Rules handle simple thresholds well. An agent can weigh a combination, such as a pricing visit plus a support ticket about an integration plus a new stakeholder in the CRM. It can then write a short explanation of why the pattern might matter, which is what makes the brief useful to its reader.
A worked example: one account, one afternoon
Consider a hypothetical software company tracking a target account. Over two hours, three visitors from the account's domain read the implementation guide, one views pricing, and a known contact registers for a technical webinar.
A rule-based system might fire three separate alerts, or none if each event falls below its threshold. An interpreting agent groups the events by account and notes that the activity spans technical and commercial pages. It checks the CRM, finds an open opportunity in an early stage, and prepares a brief for the account owner. The brief summarizes what was viewed, without naming individual visitors, and suggests two possible next steps.
The account owner decides what to do. The agent does not email anyone at the account, and it records the brief, the events it used, and the time it was delivered. If the owner reaches out, that action is recorded too, so the outcome can be compared with similar accounts later.
Guardrails for speed
Speed changes how a response feels to the person on the other end. These four guardrails keep real-time responses useful rather than unsettling.
Avoid the surveillance effect. Reacting too quickly and too specifically can feel intrusive. An email that arrives minutes after someone viewed a pricing page, referencing that page, can damage trust. Use signals to inform timing and priority for people, not to trigger instant automated messages that reveal tracking.
Throttle responses. Limit how often any account or contact can trigger an action in a period. A burst of activity should produce one brief, not ten.
Respect consent. Use only the tracking and signals that your consent framework allows, and apply opt-outs in real time, not at the next sync. Consent-aware personalization covers the design, and the rules for your regions are worth confirming with counsel.
Prefer human-facing actions. Real-time signals are often most valuable as alerts and briefs for sales and success teams, who can decide how to respond. Building trust with sales describes how to make those briefs worth reading.
What to measure
Speed is only worth paying for if it changes outcomes. The cleanest comparison is between signals routed in real time and comparable signals handled through the previous batch process. Where volume allows, hold back a share of qualifying accounts, such as 10%, and keep them on the batch path. The method in incrementality testing for agent-driven campaigns applies directly.
We recommend tracking four measures for both groups:
- Time to awareness: how long from the qualifying signal to a person seeing it.
- Action rate: how often the account owner acts on the brief.
- Downstream outcomes: meetings booked and opportunities progressed.
- Negative signals: unsubscribes, complaints, and replies that mention feeling tracked.
Watch the last measure as closely as the others. A gain in meetings that comes with a rise in complaints is not a gain you want to scale.
What changes at scale
A single high-intent pattern routed to a few account owners is easy to manage. Hundreds of signals per hour across many accounts are not. Alert fatigue sets in quickly, and owners start ignoring briefs that arrive too often.
Scale also brings infrastructure limits, such as rate limits on the CRM and costs for each interpretation call. The question we are studying is how to set signal thresholds that adapt to each owner's capacity. As volume grows, the stream should deliver fewer and better briefs, not more of them.
How to apply this: choose one high-intent signal combination and route it in real time to the account owner as a short brief, rather than an automated email. Compare outcomes against a held-back group on the batch path.
Related research
Connecting agents to your CRM without breaking it
The CRM is the system of record for customer relationships. Giving agents access safely takes a deliberate plan for permissions, writes, logging, and reversal.
The marketing data contract: schemas agents can depend on
When upstream data changes without warning, agents break quietly. A data contract defines what agents can rely on and who is responsible for keeping it true.
Retrieval for marketing agents: getting the right facts at the right time
Retrieval lets an agent look up what it needs instead of holding it all in its instructions. Done well, it raises accuracy and makes claims checkable.
Put this research to work
Marketix turns these patterns into software: agents, automated workflows, and a hub that keeps people in control. Sign up for early access.
