Next case →
Case study · 02

Smart Flighting

Convincing media teams that spending in fewer weeks would beat spreading the budget evenly. An interactive timing optimizer that replaced Excel, targeting a +1–1.5% net-revenue uplift from timing alone, without spending a euro more.

Role
Product lead
Context
Heineken · 2021 – 2026
Domain
Media planning
Approach
Response curves · saturation
Flighting plan 2025 · Meridia

Every campaign and touchpoint, across the year

Brands break into campaigns, campaigns into touchpoints. Each bar is when that channel is on air, so overlaps and gaps read at a glance before anything is booked.
Timeline List Assessment
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
ROI
Nordlys
1.38
Coastal Nordlys
1.42
Open TV
Paid TV
Online video
Social
Nordlys Winter
1.31
Open TV
Online video
Social
Verano
1.35
Verano Summer
1.44
Open TV
Paid TV
Online video
Verano Zero
1.28
Online video
Social
Brand, full year Campaign flight Touchpoint on air
Illustrative concept, not the actual product.Fictional brands and data.

The product in one screen. Every brand's campaigns and the touchpoints under them, laid out across the year, so overlaps, gaps and commitments read at a glance instead of across a dozen spreadsheets. Illustrative concept · fictional brands and data.

+1 – 1.5%
Net-revenue uplift from timing alone: same budget, better weeks.
Targeted · internal
Delivery
Excel → interactive toolkit
A manual, fragile spreadsheet process replaced by a fact-based planning tool teams actually use.
Behaviour change
Front-loaded → retimed
Planners moved from always-on habit to concentrating pressure in the weeks that still respond.
Approach
Saturation, made visible
Response and ad-stock curves surface the week each audience stops responding to more pressure.
01 · Overview

What Smart Flighting does

Allocation AI decides how much each brand and channel gets. Smart Flighting decides when to spend it, turning a static, usually front-loaded budget into a dynamic, week-by-week execution plan.

Media teams draft a flighting plan (a weekly laydown of campaigns per brand and touchpoint) and test it against AI response curves. The tool projects the uplift of each schedule, flags campaign overlap, and shows the week an audience becomes saturated by consecutive pressure.

Built under the "Win with Execution" initiative, it grew from an Excel proof of concept into an interactive toolkit that runs alongside the same allocation system the budget was set with.

02 · The problem

The model was never the hard part.

Media budgets were laid down statically, usually front-loaded, and planned in a spreadsheet, week by week, on experience. But audiences saturate: past a point, another week of pressure adds cost, not sales, because the message is already absorbed (ad-stock). The data science could model response and saturation per brand and touchpoint. That was not the hard part.

The hard part was a belief. Telling a media team that going dark in some weeks, and concentrating pressure in others, earns more from the same money runs against the instinct to be "always on." A plan with gaps looks like under-spending to the people who run it. The blocker was never the model; it was persuading planners to accept a schedule that felt wrong.

Nordlys Pilsner · Online video · Meridia

Where the next euro of pressure stops paying

Modeled response per week of advertising pressure. The curve bends because audiences saturate: past the point where the line flattens, another GRP costs full price and returns almost nothing.
Weeks past saturation
5 of 16
First 560 GRPs buy
79%
Next 450 GRPs buy
18%
SATURATION POINT pressure stops paying back same total pressure, retimed Front-loaded plan ~1000 GRPs/wk Retimed plan ~560 GRPs/wk MODELED NET-REVENUE RESPONSE WEEKLY PRESSURE (GRPs) →
Modeled response Retimed plan, at the frontier Front-loaded plan, past saturation
Illustrative concept, not the actual product.Fictional brands and data.

Why the plan goes dark some weeks. Every touchpoint saturates: the first ~560 GRPs buy most of the response, the next 450 buy almost none. Made legible so a planner can choose to move money off the flat part of the curve. Illustrative concept · fictional brands and data.

03 · My role & ownership

Accountable for whether planners actually changed the plan.

I led discovery and owned the move from a manual Excel toolkit to an interactive product. I was the standing bridge between the data scientists building the response and saturation curves and the media planners who had to change their laydowns.

I was accountable for whether planners actually built plans differently, not just whether a tool existed.

04 · Key decisions

Three product bets that earned belief

Each one traded the fastest route to an optimum for a plan media teams would actually run.

01
A simulator planners drive, not an optimizer that hands them a plan.
The impressive path was a one-click "here is your optimal schedule." I rejected it. A media team won't accept a plan that goes dark in weeks they always run, unless they can test their own laydown and watch the model show them the waste. So the tool lets them drive, and surfaces saturation as they go.
The tradeoffA slower route to the mathematically optimal plan, in exchange for one planners trust enough to use.
Driven, not handed
02
Made wasted pressure visible per week, not just a recommendation.
Rather than a black-box "better plan," the tool shows where each extra week of pressure stops paying back. Exposing the model's logic to scrutiny is the point: a planner who can see the saturation is a planner who will move the money.
The tradeoffThe model has to defend itself on screen, every week.
Visible over black-box
03
Shipped a lightweight tool to replace Excel, not a platform.
It started as an Excel proof of concept, itself the cheap validation that timing alone moved the number. I said no to over-building: get planners off spreadsheets and into a fact-based toolkit fast, prove the timing value, and leave deeper integration for later.
The tradeoffAn interim tool rather than the fully integrated end state.
Ship over build
05 · What shipped

A simulator planners drive, alongside the planning cycle.

An interactive simulator and optimizer under "Win with Execution." Planners build a full-year plan across brands, campaigns and touchpoints, then test any laydown against AI response curves, seeing projected uplift, campaign overlap, and where the audience saturates, week by week. It replaced a manual Excel process and runs alongside the yearly allocation system.

Response curves · Ad-stock & saturation modelling · Interactive simulator + optimizer · Allocation AI architecture

Nordlys | Coastal Nordlys

Summer campaign, Meridia, across TV and social.
Draft ▾ Compare scenarios
Scenario 1 Scenario 2 Scenario 3 Scenario 4

More weight on social

✦ Optimized
Edit scenario
Shift budget off saturated TV weeks into social, and concentrate the spend into fewer bursts.
Campaign KPIs
Cost & activation
Cost
€9.2M
GRPs (norm. 30s)
7.1K
Impressions
1.3B
AP adherence
84%
Flighting assessment
Sell-out ROI
1.38
Optimal timing
47%
Optimal start
45%
Overlaps
0
Volume
Net revenue
€6.8M
Gross profit
€3.0M
Incr. volume
4.6M hL
Touchpoint breakdown
Open TV ✦ Optimized
Cost
€4.1M
GRPs (norm. 30s)
3.2K
Impressions
0.6B
Sell-out ROI
1.42
Weeks on air
11
0 2.5M 5.0M 7.5M 10.0M 12.6M ▸ Jan 25 ▸ Feb 25 ▸ Mar 25 ▸ Apr 25 ▸ May 25 ▸ Jun 25 ▸ Jul 25 ▸ Aug 25 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 5.0M IMPRESSIONS WEEK NUMBER (ISO)
Responsive capacity (still pays) Impressions delivered per week
Illustrative concept, not the actual product.Fictional brands and data.

How a planner tests a call before committing. Each scenario is a saved what-if (here, more weight on social) scored on the same KPIs, with delivered impressions plotted against the responsive ceiling. Illustrative concept · fictional brands and data.

06 · Impact

Timing alone, worth 1–1.5% of net revenue.

The real proof wasn't the model's accuracy: it was the behaviour change. Planners moved from defaulting to front-loaded, always-on plans to testing laydowns against saturation and redistributing pressure to the weeks that still respond. Targets are labelled as targets.

Fewer, bigger events · Nordlys Pilsner · Meridia

Same budget, retimed

The always-on plan the team started with, next to the one Smart Flighting proposed. Identical spend. The model pulled pressure off the weeks that had stopped paying and out of the dead tail, into fewer, fuller bursts that stay under saturation.
Before · Front-loaded16 weeks on air
Saturation
12345678910111213141516
5 weeks pushed over saturation, then a long dead tail the audience barely registers.
After · Smart Flighting11 weeks on air
Saturation
12345678910111213141516
Every week capped under saturation, dark in the weeks that did not respond. No wasted pressure.
Planned pressure Above saturation (wasted) Saturation threshold
Total budget
€1.0M
unchanged, the constraint
Weeks past saturation
5 0
no pressure wasted
Modeled net revenue
+1.3%
same spend, better timing
Weeks above ROI benchmark
44 73%
pressure lands where it pays
Illustrative concept, not the actual product.Fictional brands and data.

The payoff, on the same money. The always-on plan next to the retimed one: identical spend, pulled off the wasted weeks into fewer, fuller bursts under saturation. Weeks past saturation fall to zero and modelled net revenue rises about 1.3%. Illustrative concept · fictional brands and data.

07 · The lesson

More is not better. Timing is. The same budget, spent when the audience can still absorb it, beats the same budget spread thin, and a media team believes that only when the tool shows them the wasted weeks.

08 · In context

The timing layer of a three-part system

Smart Flighting takes the budget Allocation AI sets and stretches it across the calendar; Promo Advisor fills that calendar with offers.

Keep exploring

That's the timing layer. See the rest.

Allocation AI secures the funding; Promo Advisor executes the offer, or head back to all work.

Tomasz Czarnecki © 2026 · czarnecki.ai · Product visuals are illustrative concepts