Marketers aren’t short on data. We’re short on direction.

Ask what “performed best” and someone says, “the dog meme.”

Cue more dog memes. ;-)

In a world of faster cycles, non-linear buying journeys and more signals than human sensemaking can manage, the promise of insight-driven decision-making can feel more like noise than guidance. What we need is a repeatable way to turn signals into decisions. That’s an insight engine.

It’s not a tool you buy or plug in, but a human-led, AI-powered system you create, a skill you build and a process you run every week. It’s made of repeatable processes, shared language, and a productive partnership between humans and AI.

Think radar and captain: AI (the radar) scans constantly and whispers, “Look here.” Humans (the captain) choose the course. When you wire a few key inputs, ask sharper questions, and close the loop from finding → decision → shipped change, data stops being noise and starts steering the business.

AI tools don’t just help us make sense of what happened. They help us understand why it happened and what to do next. By revealing hidden patterns, surfacing unmet needs, and connecting disparate signals, they help turn data into direction. And when marketers pair this with smart questions and sharp instincts, data stops being overwhelming and starts becoming transformative.

At the centre of this is a new role for marketing: the orchestrator of insight. Marketers curate inputs, frame better questions, and turn patterns into direction for product, creative, sales, and service. AI is the always-on radar and co-analyst, surfacing patterns, trends, and human behaviours you’d otherwise miss. But machines don’t replace judgment. They amplify it. When humans ask better questions, AI reveals deeper answers.

What Counts As An Insight?

Let’s get crisp. An insight is:

Non-obvious: It changes how you see the problem.
Reliable: It holds up across data slices and time windows.
Actionable: It points to a concrete decision or experiment.

Here’s a one-line test you can use: if it doesn’t change what we ship next, it’s not an insight. Let’s bring back the dog-meme example:

Observation: “Dog-meme posts get more likes.”

Insight: “Prospects with low prior category knowledge advance with step-by-step creative that reduces effort; aspirational headlines stall them.”

The latter is non-obvious, consistent across channels and tells your creative team what to do next. It’s an insight.

The Anatomy Of An Insight Engine

An insight engine is not a tool; it’s a stack of data, models, and rituals that repeatedly convert raw signals into decisions.

Govern (Trust Layer)Define shared definitions (‘What is a “qualified lead”?’) and publish consent and retention rules.
Collect (Signals Layer) Unify messy inputs (search trends, site analytics, email events, paid media logs, CRM fields, support tickets, sales notes, call transcripts) into a single accessible layer. Standardize IDs, timestamps, and consent flags. Build the habit of tagging events with the same taxonomy across teams. Watch Out: Adding sources faster than you add questions.
Interpret (Meaning Layer) Use models to detect segments, moments, and motivations; not just metrics. Topic modelling on transcripts, clustering on behaviours, causal uplift models for treatments, outlier detection for edge opportunities. The goal: pattern literacy, not dashboard decoration. Watch Out: Correlation Theatre
Orchestrate (Action Layer) Translate insight into experiments, briefs, and frontline enablement. Spin up creative variants, adjust offer logic, arm sales with “if you hear X, ask Y,” and update onboarding scripts. Ship changes and watch impact roll back into the signals layer. Loop closed. Watch Out: Ending meetings with screenshots, not assignments.

From Metrics To Meaning

Dashboards answer what happened.

Insight engines answer why, so what, and now what.

As an MBA marketing instructor, I’ve been coaching aspiring marketers to do this for years. Now, the insight engine can help us by pivoting from performance reporting to true behavioural understanding:

Segments: Who responds to what (and who never will).
Moments: When intent spikes and wanes, by channel and context.
Motivations: Which jobs-to-be-done, anxieties, and desired outcomes drive action.
Mechanisms: What actually causes lift—not correlation theatre.

Example: instead of “email open rate dropped,” your engine surfaces: “Opens fell 12% among new evaluators from partner referrals after day 5, coinciding with a shift from ‘how-to’ to ‘what’s new’ content. Switching the day-5 email to a ‘first-wins’ checklist restored engagement and increased trial completions by 9%.” That is direction (i.e., now what).

From Manager Of Marketing To Orchestrator Of Insights

This is where the marketer’s role evolves. You’re not the person who “owns the channel” or “pulls the report.” You take on the role of orchestrator of insight. You curate inputs. You decide the questions. You bridge the worlds of models and messaging, turning patterns into briefs that the creative team can use now and talking points that the sales team can carry into calls tomorrow. You don’t worship the dashboard; you run the process. You are the real differentiator, the multiplier, the captain – not the tech.

Orchestrator of Insights: Curate inputs. Frame questions. Turn patterns into decisions others can ship.

What does that process look like? Picture a Weekly Insight Review that feels more like a newsroom than a status meeting. The first ten minutes are table-setting: what changed, where, and for whom. Then you move into the heart of it; the patterns. Not a flood of numbers, but a handful of contrasted scenes: a conversation that went sideways and one that flew, a cohort that stalled and a cohort that sprinted, a phrase that unlocked trust and a phrase that triggered doubt. You end with decisions: what we will do differently because of what we learned, who will do it, and when we’ll know if it worked. People leave with assignments, not screenshots.

It starts with asking AI better questions:

Never ask AI to “find everything interesting.” AI rewards specificity. Here are seven “must-answer” campaign questions. Each is written to trigger a concrete decision the moment you have the answer.

1.Who are we moving this week, and at what moment are they deciding? If answered, we will change: segment inclusion/exclusion, geo/time windows, frequency caps.

Prompt Suggestion: Analyze last week’s data to pinpoint the single microsegment we can most move now and the exact decision moment; return segment include/exclude rules, geo/time window, and recommended frequency cap.

2. What single belief must shift for them to take the next step? If answered, we will change: headline/value prop, message hierarchy, creative concept.

Prompt Suggestion: From interviews, chats, and drop-off paths, extract the one belief blocking the next step and rewrite it as a before→after statement; return headline/value prop and message hierarchy.

3. Which anxiety or friction blocks that step, and what proof removes it? If answered, we will change: proof points (demo, testimonial, guarantee), social proof placement, UX friction (steps, form fields).

Prompt Suggestion: Diagnose the top anxiety/friction at [step] and match it to the minimal proof that removes it; return proof type + placement and the specific UX change (steps/fields) to ship.

4. Which message frame (problem, outcome, step-by-step, objection-handling) most reduces time-to-first-action? If answered, we will change: copy style, asset type (checklist vs. promo), length and structure of landing page/email.

Prompt Suggestion: Model past creatives to find which frame (problem, outcome, step-by-step, objection) minimizes time-to-first-action for [persona]; return copy style, asset type, and page/email length.

5. Where and when does intent spike for our core persona (channel × format × timing)? If answered, we will change: channel mix and budget split, bid schedules, creative format (short video, carousel, longform).

Prompt Suggestion: Surface channel×format×time combinations where intent spikes for [persona]; return budget split, bid schedule, and recommended creative format to exploit the spike.

6. What’s the smallest next step that meaningfully predicts revenue/retention, and what CTA/offer best drives it? If answered, we will change: CTA wording and placement, offer design (trial, sample, calendar nudge), success metric for optimization.

Prompt Suggestion: Identify the smallest next step that best predicts revenue/retention for [product] and the CTA/offer that most increases it; return CTA wording/placement, offer design, and the optimization metric.

7. What mechanism are we testing now, and what is the threshold to scale or stop? If answered, we will change: experiment design (cells, sample size), guardrails, go/no-go decision and rollout plan.

Prompt Suggestion: State the mechanism under test (e.g., belief shift, friction removal) and propose an experiment with cells and sample size; set scale/stop thresholds with guardrails and return go/no-go + rollout plan.

Wire your data just enough to answer these questions. You don’t need twenty sources; pick three that matter most. Then set a cadence (i.e., every Thursday afternoon) to tell the story of what you learned, what you’re going to do about it, and how you’ll know if it worked.

Pro Tip: Name great insights after people to make learning social and sticky. Capture it in a one-page “Insight of the Week” (Context, Finding, Evidence, Decision, Next Test).

The AI Unlock: How AI Turns Data into Direction

AI turns raw signals into clear next moves by running a tight performance loop:

1) AI scans the noise and spots the pattern

Feed it behaviour data (clicks, usage), voice-of-customer (calls, chats, reviews), and outcomes (conversion, retention). Models cluster themes, flag anomalies, and reveal “moments that matter” you’d miss by eye. AI spots the patterns that humans miss and connects dots we wouldn’t even think to connect.

2) AI explains the “why,” not just the “what”

Instead of “opens dropped on day 5,” AI links evidence to human motives: “New evaluators express fear of ‘breaking something’ right before activation.” That’s a mechanism you can act on.

3) AI translates the pattern into a decision

Turn the finding into one specific change with an owner and a metric: rename “Tutorial” to “First Wins,” add a 90-second “you can’t mess this up” walkthrough, and swap the day-5 email from “news” to “next step.”

4) Ship and learn

Publish the change, measure the effect (activation up/down), feed results back to the model, and ask the next, sharper question.

Remember, AI is your radar, not your captain. It surfaces patterns; you choose the course.

Humanology in Practice

At Jan Kelley, we’ve embedded Human-led, AI-powered insight engines into our JKAI platform, empowering strategy and creative teams with dynamic, data-informed direction.

For example, while working with a national home services brand, our system detected that customers who engaged with how-to video content were far more likely to convert during off-peak months. This unexpected insight allowed us to restructure both content sequencing and media spend, resulting in a 27% lift in seasonal engagement.

What made the difference wasn’t just the AI detection, it was our strategists’ ability to explore “why,” validate it with real customers, and translate it into creative decisions. In our Humanology model, insight doesn’t end with an output. It begins with human interpretation.

What To Do Now – And Next

This Week:

Write 7 key questions you want to answer in every campaign; make each question change a decision if answered.
Pick just 3 data sources (behaviour, voice of customer, outcomes); wire them together with a simple shared taxonomy.
Ask AI one focused prompt: “What anxieties stall people between demo and first success?” Use the answer to ship one small change.
Stand up a 60-minute Weekly Insight Review; end with two decisions, each with an owner, date, and metric.
Publish a one-page “Insight of the Week” (Context, Finding, Evidence, Decision, Next test).

This Month:

Backfill 3–6 months of transcripts, reviews, and call notes; cluster for themes and “switching moments.”
Map “moments that matter” across the journey and pair each with a message that reduces effort or risk.
Create a living Decision Log that tracks recommended actions, adoption, and outcomes.
Define guardrails: consent and retention policy, bias checks, explainability, and a human override.
Archive two dashboards that don’t feed decisions.

Always:

Start with better questions; never ask AI to “find everything interesting.”
Translate patterns into narratives others can use this week (creative briefs, sales talk tracks, onboarding tweaks).
Measure capability, not just outcomes: time-to-insight, decision adoption rate, and experiment velocity.
Name great insights after people to make learning social and sticky.
Close the loop: insight → decision → shipped change → observed effect → next question.

Closing Thoughts

Insight is no longer accidental. It’s intentional, iterative, and systemic. With AI doing the heavy analytical lifting and humans orchestrating the inquiry and interpretation, marketers are finally equipped to not just react to the market but to see around corners.

AI will keep getting faster at reading, clustering, and predicting. Our job is to keep getting better at asking, framing, and deciding. Be the orchestrator who turns signals into stories, stories into steps, and steps into tangible outcomes the business can feel.

Write your questions. Wire your basics. Tell your weekly story. Change one thing you can ship this week. Do that again next week. Direction doesn’t appear out of nowhere; it emerges from disciplined learning.

So here’s the real question: Are you still waiting for the next big insight to drop? Or are you building the system to find it?

Want to explore how Jan Kelley helps businesses build Human-led and AI-powered Insight Engines? Contact us to chat.

AI-Powered Insight Engines

More Humanology Insights

Never miss a Humanology Insights post