A campaign can hit every deadline and still miss the customer.

Consider a homeowner comparing replacement heating systems. They request an estimate, return to the financing page and ask for a callback. The marketing team sees a lead. The ad platform records a conversion. The next mailing goes out on schedule.

The homeowner still has a question. Whether somebody answers it, with the right context, will matter more than whether the campaign launched on time.

That is the gap I want marketing to close.

Campaigns are useful. They organize investment, creative and offers. But a customer's next step depends on what has changed for them, and those changes keep happening between launch dates.

In Salesforce's 2026 State of Marketing research, 69% of marketers said they struggled to respond promptly because they could not access the context they needed. That is a practical constraint on growth, even when the business already has the information somewhere. ¹

Signal-Led Growth starts with a different operating question: given what we know now, what should the business do next for this customer?

Marketing has an understandable bias toward production. A new campaign is visible. More creative, more audiences and more sends can be counted. Improving a decision that prevents an unnecessary discount or gets an unanswered question to the right person is harder to put on a launch calendar.

Yet budgets make those decisions consequential. Gartner's 2026 survey found marketing budgets effectively flat at 7.8% of company revenue, with 56% of respondents saying they lacked the budget to deliver their strategy. The survey mostly represents large enterprises, but it frames a question every operator can ask: where will the next dollar make a difference? ²

Sometimes the answer is a new audience. Sometimes it is a stalled opportunity, a renewal at risk or an existing customer who would benefit from another service. Sometimes the business has already done enough and should stop paying to persuade somebody who has decided.

A growth system needs to make room for all of those answers.

I use Signal-Led Growth to describe a way of operating in which new evidence changes what a business does next.

Predictive models help assess fit, likely behavior and customer value. AI decisioning evaluates the available actions. Agents carry out approved work across the relevant channels, within goals, permissions and spending limits set by people.

The signal is new evidence. A Growth Play is the repeatable, governed response to that evidence. Campaigns can still provide the creative, offer and investment around it.

The customer relationship continues across all of those activities. A booking should change acquisition activity. A service problem should change the tone and timing of an expansion offer. A purchase should update the next conversation, wherever it happens.

A new mover record tells you that circumstances have changed. The age of a home adds context. A visit to a pricing page suggests interest. A requested callback tells you what the person wants next.

Those observations carry different weight.

Three signals do not turn uncertainty into certainty. They may be stale, linked to the wrong person or generated by the same interaction. The useful questions are where the evidence came from, how fresh it is, whether it refers to the right customer and how much it changes the decision.

Physical channels require the same discipline. USPS mail tracking provides handling and logical delivery information. Informed Delivery provides digital previews and interactions. Those events should not be described as proof that someone opened or read a physical mailpiece. ³

A personalized QR scan tells you that a code was used. It can connect an interaction to a mailed record, but someone else may have scanned it. The system should retain that uncertainty.

More data is valuable when it leads to a better decision. A larger collection of weak observations can just produce more confident mistakes.

Experienced marketers have worked with triggered journeys, predictive scores and next-best-action systems for years. Real-time decisioning and control groups are already documented in established platforms. Signal-Led Growth builds on that history. ⁴

The opportunity now is to reduce the manual work between understanding an opportunity and carrying out an appropriate response.

An agent can retrieve the latest estimate, check whether an appointment already exists, assemble an answer from approved information and initiate the next permitted step. Predictive models can help prioritize the work. Business rules can enforce eligibility, budget and contact limits.

The economics of model use have also changed. Stanford's 2025 AI Index reported a more than 280-fold fall in the cost of querying models at roughly GPT-3.5 benchmark performance between November 2022 and October 2024. That comparison covers model queries; integration, supervision and channel execution remain separate costs. Lower query costs help explain why applying models to smaller decisions has become more practical. ⁵

Execution still depends on reliable data, connected systems and controls. An agent can initiate a mailing quickly; it cannot remove postal transit time. It can request an appointment; it cannot create installation capacity that the business does not have.

The human team remains responsible for deciding what the system is trying to achieve and what it is allowed to do.

The kitchen table, the phone and the mailbox belong in this story. So do the website, the app and the digital checkout.

J.D. Power's 2026 U.S. Insurance Shopping Study found that 48% of new auto policies were purchased digitally. Even a considered purchase can close on a screen. A customer may still have asked questions or compared options elsewhere along the way. ⁶

That makes continuity the practical requirement. When a customer changes channels, the business should carry forward the relevant offer, questions and actions already taken.

Invoca's 2026 benchmark, based on more than 70 million calls, reported a 56% average human answer rate. The remainder should not be treated as lost buyers: some calls are short hangups, misdials or interactions handled through automation. The finding gives operators a reason to inspect what happens after a campaign generates a call. ⁷

Did the customer get the answer they needed? Did the conversation lead to a useful next step? Did the booking reach the system that controls the next advertisement?

Sending a response is only one part of execution. The other part is making sure the rest of the business knows what happened.

Here is an illustrative Growth Play for an open home-services estimate. Its objective is a qualified appointment, with the eventual completed job and contribution tracked as well. It is a proposed operating example, not a customer result.

The team sets the eligible audience, permitted contact channels, spending limits, service-area rules and available appointment capacity. The agent works within those boundaries.

The important part is the connection between evidence, judgment and execution. The Play describes what should happen when circumstances change, including when activity should stop.

The same logic applies to acquisition, retention and expansion. A prospective customer may warrant an initial introduction based on fit. A renewal may call for help resolving a service issue. A relevant additional product may be worth offering after the original need has been met.

A signal-led business still invests in creating demand. It does not need to wait for every prospect to announce an intention to buy. It does need a plan for what happens when the evidence changes.

A high likelihood of purchase does not tell you how much a marketing action will change the outcome.

A customer may already be ready to buy. An additional discount could reduce margin without increasing conversion. Another customer may need one specific question answered before they can decide. A third may have no current need at all.

The useful judgment is whether a particular intervention is likely to add enough value to justify its cost and the customer's attention.

That changes how channels compete for budget. Direct mail has different costs and timing from an email or a call. A more expensive intervention may make sense for one opportunity and be wasteful for another.

Waiting, suppressing redundant activity and asking a person to take over are valid decisions. They should be visible in the system's record, alongside the messages it sends.

A booking after a message is encouraging. It does not establish that the message caused the booking.

Matchback and attribution help connect activity with outcomes. A well-designed comparison addresses the additional effect of the intervention.

Before launching a Play, define the outcome, the comparison and the time needed to observe it. If the question is whether a new Play improves current performance, the comparison can be the existing process. If the question is whether an extra contact adds value, some eligible customers can be randomly held back from that contact.

Keep assignment consistent across the channels being tested. Choose the comparison size around the available population, normal conversion rate and improvement worth detecting. A standard holdout percentage is not a substitute for that judgment.

Then look beyond response. Were appointments qualified and attended? Did they become completed work or retained customers? What additional contribution remained after media, production, incentives and handling costs?

Where results are uncertain, say so. A small test can help identify a problem without being strong enough to support a lift claim.

This is where the final step, Learn, matters. Results should change the Play: which evidence gets more weight, which contacts are suppressed, which channels earn budget and which situations need human judgment. The next decision should benefit from what the business has learned.

Giving agents authority to execute makes the operating boundaries more important.

Every Play needs an owner. That person defines the objective, economics, permitted actions, escalation points and standard of evidence. Customer Growth Engineering is the discipline of building and improving those Plays. A marketing lead, an agency operator or a dedicated team can do that work.

The system also needs clear rules for data and contact. Use information for an appropriate, documented purpose. Preserve the distinction between an observed event and an inferred need. Respect the relevant preferences and permissions, and make it possible to review why an action was taken.

A signal alone creates no permission. A page visit should not silently become authorization for an automated sales call.

Good execution should make the experience more useful to the customer. Otherwise, faster decisions will simply create faster irritation.

The starting point can be modest: one recurring situation where customer context changes what the business should do.

Choose an outcome that matters. Identify the evidence available. Define the decisions, allowed actions and stop conditions. Give an agent a bounded role, connect the result back to the customer record, and test whether the intervention adds value.

That is the direction we are pursuing at Boostt AI: predictive intelligence, AI decisioning and agent execution across the customer journey, with people setting the goals and limits.

Campaigns will still launch. The operating advantage will come from what the business does between them.

When something changes for a customer, does your business recognize it, make a useful decision and carry that decision through? And can you show that the action was worth taking?