MTA, MMM & Lift Studies: The Triangulation Approach

· Last updated · 13 min read

No single method captures the full picture. MTA provides tactical, touchpoint-level insights but misses untrackable channels. MMM captures all channels including brand and offline but lacks granularity. Incrementality testing proves causation but can't run everywhere at once. Triangulation uses all three: MTA for daily optimization, MMM for budget allocation, incrementality to calibrate both. When all three agree, you have high confidence. When they diverge, you've found something worth investigating.

Why One Method Isn't Enough

Every measurement approach has blind spots:

MTA
  • ×Can't see untrackable touches (dark funnel)
  • ×Misses iOS users (ATT opt-out)
  • ×Limited by cookie windows
  • ×Doesn't prove causation
  • Granular, real-time, actionable
MMM
  • ×Too slow for tactical decisions
  • ×Channel-level only, not creative-level
  • ×Requires 2–3 years of data
  • ×Sensitive to external factors
  • Privacy-safe, captures all channels
INCREMENTALITY
  • ×Can only test one thing at a time
  • ×Requires significant sample size
  • ×Expensive — the holdout is lost revenue
  • ×Point-in-time, results may not hold
  • Proves causation definitively

Relying on any single method is like navigating with one eye closed. Triangulation gives you depth perception.

The Three Pillars of Measurement

1. Multi-Touch Attribution (MTA)

What it does: Tracks individual user journeys across touchpoints and assigns credit based on the path to conversion.

Best for:
- Daily/weekly campaign optimization
- Creative and audience testing
- Understanding customer journeys
- Real-time performance monitoring

Limitations:
- Only sees trackable touchpoints
- Correlation, not causation
- Privacy restrictions shrinking visibility

MTA VIEW OF A CUSTOMER JOURNEY

Touch 1
Google Ad
20%
Touch 2
Blog Post
15%
Touch 3
Retargeting
25%
Touch 4
Email
40%
Touch 5
Purchase
$100 AOV

Tracked journey value: $100. But MTA missed the podcast mention that sparked the initial interest, the friend recommendation during the consideration phase, and the LinkedIn post that built trust.

2. Marketing Mix Modeling (MMM)

What it does: Uses econometric analysis of aggregate data (spend, impressions, revenue) to estimate channel contribution over time.

Best for:
- Annual/quarterly budget allocation
- Understanding brand and offline channels
- Long-term strategic planning
- Privacy-compliant measurement

Limitations:
- 4-8 week refresh cycles
- Channel-level granularity only
- Requires significant historical data
- Sensitive to market changes

Channel Revenue contribution Confidence
Paid Search 28% ± 4% High
Paid Social 22% ± 5% Medium
TV / Video 18% ± 6% Medium
Brand / Organic 20% ± 8% Lower
Email 8% ± 2% High
Display 4% ± 3% Medium

Includes offline, untracked, and long-term brand effects that MTA misses.

3. Incrementality Testing (Lift Studies)

What it does: Runs controlled experiments (holdout groups, geo-tests) to measure the true causal impact of marketing.

Best for:
- Validating MTA and MMM
- High-stakes budget decisions
- Testing new channels
- Challenging assumptions

Limitations:
- Point-in-time measurement
- Expensive (holdout = foregone revenue)
- Can only test one thing at a time
- Requires sufficient scale

INCREMENTALITY TEST · META

Question: what's the true incremental value of Meta? · 50/50 geo split · 4 weeks

TREATMENT (META RUNNING)
$500K
revenue
CONTROL (META PAUSED)
$380K
revenue
  • Incremental lift: $120K (31.6%)
  • Meta spend: $100K
  • True iROAS: 1.2×
MTA-REPORTED
2.8×
+133% overcredited
META PLATFORM
3.5×
+192% overcredited

How Triangulation Works

Triangulation doesn't just run all three methods—it connects them into a unified system.

The Triangulation Framework

MEASUREMENT TRIANGULATION

INCREMENTALITY
Ground truth · causation
calibrates
MTA
Tactical
what to optimise
MMM
Strategic
where to invest
↔ cross-validate ↔

Each method informs the others:

Relationship How It Works
Incrementality → MTA Calibration factors adjust MTA overcounting
Incrementality → MMM Validates MMM coefficients against experiments
MTA → MMM Provides granular input for MMM to model
MMM → MTA Reveals channels MTA can't see
MTA ↔ MMM Divergence reveals blind spots

Real-World Triangulation Example

Here's how a $5M/year marketing team might use triangulation:

REAL-WORLD TRIANGULATION · A WORKED EXAMPLE

  1. 1
    Baseline with MTA

    MTA shows: Paid Search 35%, Meta 28%, Display 18%, Email 12%, Organic/Direct 7%.

    Red flag: only 7% organic? Brand is strong — something's being miscredited.

  2. 2
    MMM for the strategic view

    MMM shows: Paid Search 25%, Meta 18%, Display 8%, Email 10%, Organic/Brand 30%, Seasonality 9%.

    MTA was overcrediting paid channels and undercrediting brand by ~4×.

  3. 3
    Incrementality to validate
    • Meta test: MTA said 28%, MMM said 18%, geo-holdout proves 14% incremental lift. MTA overcredits Meta by ~2×.
    • Display test: MTA said 18%, MMM said 8%, geo-holdout proves 3% incremental lift. Display is mostly retargeting existing intent.
  4. 4
    Calibrated view
    ChannelMTA rawCalibrationAdjusted
    Paid Search35%× 0.8028%
    Meta28%× 0.5014%
    Display18%× 0.204%
    Email12%× 1.0012%
    Organic / Brand7%× 6.0042%

    Budget decision: shift 15% from Display to Paid Search. Display is retargeting, not driving.

Triangulating ROAS across three methods

When MTA, MMM, and incrementality agree, you have signal. When they don't, you have a research question.

Paid Search

MTA
4.2x
MMM
2.0x
Incrementality
1.9x
Calibrated →
2.0x

Paid Social

MTA
1.2x
MMM
2.8x
Incrementality
2.5x
Calibrated →
2.5x

Email

MTA
3.8x
MMM
3.1x
Incrementality
Calibrated →
3.1x

Display

MTA
2.4x
MMM
1.8x
Incrementality
0.8x
Calibrated →
0.8x

Podcast

MTA
MMM
4.8x
Incrementality
5.2x
Calibrated →
5.0x

When incrementality is available, it wins. It's the only Rung-2 method — experimental evidence beats observational estimates. Where incrementality is unavailable (Email, in this example), MMM is the next-best estimate. The calibrated column is what gets used for budget decisions.

Illustrative ROAS · Calibration logic: prefer Incrementality > MMM > MTA when methods disagree

When Methods Agree vs. Disagree

When All Three Agree: High Confidence

METHODS AGREE · HIGH CONFIDENCE

Question: how valuable is our podcast sponsorship?

  • MTA: Can't track — podcast sits in the dark funnel.
  • MMM: 12% ± 4% of revenue.
  • Incrementality: 11% lift when paused for 6 weeks.
  • Survey: 15% say "discovered us via podcast."

Conclusion: podcast drives ~12–15% of business. Three methods converge — maintain or increase the spend.

When Methods Disagree: Investigate

METHODS DISAGREE · INVESTIGATE

Question: how valuable is Meta prospecting?

  • MTA: 22% of conversions — high value.
  • MMM: 15% of revenue — medium value.
  • Incrementality: 5% lift — low value.
  • Why MTA overcredits: users see a Meta ad, then search the brand. MTA credits Meta — but they would have converted anyway.
  • Why MMM overcredits: Meta spend correlates with peak season. Hard to separate the Meta effect from the seasonality.
  • Why incrementality wins: only 5% is truly incremental. Most "Meta conversions" are brand searches in disguise.

Action: reduce Meta prospecting by 30%, monitor. Keep brand search strong.

Disagreement is valuable: When methods diverge, you've found something worth investigating. Don't average the numbers—understand why they differ.

Implementing Triangulation

Phase 1: Foundation (MTA)

Start with multi-touch attribution as your tactical layer:

Action Timeline
Implement cross-channel tracking Month 1-2
Choose attribution model Month 2
Connect to ad platforms Month 2-3
Build reporting dashboards Month 3
Train team on MTA insights Month 3-4

Output: Daily/weekly actionable metrics for campaign optimization.

Phase 2: Strategy (MMM)

Add marketing mix modeling for strategic view:

Action Timeline
Aggregate 2+ years of data Month 4
Build or buy MMM Month 4-6
Initial model calibration Month 6-7
Quarterly refresh process Ongoing

Output: Quarterly strategic budget recommendations.

Phase 3: Validation (Incrementality)

Add incrementality testing to calibrate and validate:

Action Timeline
Design test framework Month 6
Test top channel (highest spend) Month 7-8
Apply calibration to MTA Month 8
Validate MMM predictions Month 9
Ongoing test calendar Quarterly

Output: Calibration factors for MTA, validation for MMM.

Ongoing: The Triangulation Loop

QUARTERLY TRIANGULATION REVIEW

  1. Week 1
    Review MTA trends
    Major shifts in attribution? New channels or campaigns to evaluate? Anomalies to investigate?
  2. Week 2
    Review MMM outputs
    How does MMM allocation differ from MTA? What's MMM seeing that MTA can't? Update calibration if incrementality tests ran.
  3. Week 3
    Plan incrementality tests
    Which channel has the highest uncertainty? Any major budget decisions coming? Set test design and timeline.
  4. Week 4
    Cross-validate and decide
    Where do methods agree/disagree? Make budget recommendations and document learnings.

Which Method for Which Decision

Decision Primary Method Supporting Methods
"Which Meta ad set should I pause?" MTA
"Should I increase podcast spend?" MMM Incrementality, Survey
"Is our brand campaign working?" MMM + Incrementality
"How do I allocate next quarter's budget?" MMM Incrementality calibration
"Is this new channel worth testing?" Incrementality
"Why did conversions drop last week?" MTA Platform data
"What's our true Meta ROAS?" Incrementality MTA (adjusted)
"Where is budget being wasted?" All three Compare and investigate

Common Triangulation Mistakes

1. Averaging the Numbers

❌ Wrong:
MTA says 25%, MMM says 15%, Incrementality says 10%
"Let's just say it's 17%"

✓ Right:
Investigate WHY they differ.
Incrementality is causal—weight it highest.
MTA at 25% suggests overcounting—find the cause.

2. Testing Only What Looks Good

❌ Wrong:
"Our brand campaign looks great in MTA.
 No need to test it."

✓ Right:
Test your best-performing channels.
If they're really that good, tests will confirm.
If not, you've found overcounting early.

3. One-Time Calibration

❌ Wrong:
"We tested Meta in Q1, got a 0.6x calibration factor.
 We'll use that forever."

✓ Right:
Calibration factors drift over time.
Re-test major channels annually at minimum.
Algorithm changes, creative changes, audience changes.

4. Ignoring Confidence Intervals

❌ Wrong:
"MMM says Search is 23% of revenue. Done."

✓ Right:
"MMM says Search is 23% ± 6% of revenue (17-29% range).
 Given this uncertainty, here's our confidence level..."

Meta's MTA calibration factor over time

Multiplier applied to Meta-reported attribution to match incrementality-tested truth. The factor moves — you have to keep testing.

Pre-2021
Baseline
0.7x
MTA captured most signal; CAPI not yet widespread.
Apr 2021
ATT rolls out
0.5x
iOS 14.5 launches App Tracking Transparency. Meta loses cross-app signal.
Q3 2022
Modeled conv.
0.55x
Meta starts modeling ATT-blocked conversions; signal partially recovers.
Q1 2024
CAPI mature
0.62x
Server-side CAPI integration recovers more iOS conversions.
Today
Current
0.55x
Meta keeps optimizing; calibration drifts as platform behaviour changes.

A calibration factor is a snapshot, not a constant. Every time the platform changes how it attributes (modeled conversions, signal-loss recovery, AI optimization), the factor moves. Re-running incrementality tests every 6–12 months is the only way to keep MTA honest.

Illustrative trajectory · Real factors vary by vertical and account; the shape of the curve is consistent across published case studies

Triangulation Maturity Model

Level Characteristics Typical Company
Level 0: Platform Only Trust Google/Meta numbers Most small businesses
Level 1: MTA Cross-channel attribution Growth-stage startups
Level 2: MTA + MMM Strategic and tactical views Mid-market ($5M+ spend)
Level 3: Full Triangulation All three, integrated Enterprise ($20M+ spend)
Level 4: Automated Triangulation Real-time calibration, continuous testing Industry leaders

Most companies should target Level 2-3. Full triangulation isn't overhead—it's insurance against expensive allocation mistakes.

Summary

No single measurement method tells the whole truth:

Method Strength Blind Spot
MTA Tactical, real-time, granular Untrackable, not causal
MMM Strategic, privacy-safe, complete Slow, not granular
Incrementality Causal, definitive Point-in-time, expensive

The triangulation approach:

  1. MTA for daily optimization — What to change today
  2. MMM for budget allocation — Where to invest next quarter
  3. Incrementality for validation — Calibrate and confirm

When methods agree: High confidence, act decisively.

When methods disagree: Investigate—you've found something important.

The goal: A unified measurement framework where each method compensates for the others' weaknesses, continuously calibrated through experimentation.

Further Reading

Foundational Concepts:
- What is Multi-Touch Attribution? — The tactical layer
- MTA vs MMM: When to Use Each — Detailed comparison

Practical Implementation:
- iOS Tracking & Attribution — Why triangulation is necessary
- Why Platform Reports Don't Match — The problem triangulation solves

Key Takeaways

  • MTA is tactical (what to optimize), MMM is strategic (where to invest), incrementality is validation (what actually works)
  • Each method has blind spots—triangulation covers them
  • Use incrementality tests to calibrate MTA and MMM
  • Agreement across methods = high confidence; disagreement = investigate
  • Triangulation is how sophisticated marketers handle privacy restrictions
Why can't I just use multi-touch attribution?
MTA only measures what it can track. With iOS privacy changes, cookie restrictions, and dark funnel influences (podcasts, word of mouth), MTA misses 30-50% of the picture. It also can't prove causation—just because a touchpoint was on the path doesn't mean it influenced the decision.
Why can't I just use marketing mix modeling?
MMM is powerful for strategic decisions but too slow and aggregated for daily optimization. It can tell you 'Meta drives 25% of revenue' but not 'this creative is outperforming.' It also requires 2-3 years of data and can be thrown off by external factors like competitors or economic changes.
How do I know if my MTA is accurate?
Run incrementality tests on channels where MTA claims high value. If MTA says Meta drives 30% of conversions, run a geo holdout test on Meta. If the incrementality test shows 15% lift, your MTA is over-crediting by 2x. Use this to create a calibration factor.
How often should I run incrementality tests?
Major channels should be tested annually at minimum. High-spend or volatile channels (Meta, display retargeting) benefit from quarterly testing. Always test when making major budget changes—validate before committing.
What if my three methods disagree?
That's valuable information, not a problem. Disagreement often reveals measurement bias, changing channel dynamics, or external factors. Investigate: Which method's assumptions are being violated? What changed? The investigation often reveals more than the numbers.
Holly Henderson
Holly Henderson

Co-Founder, mbuzz

Holly Henderson is Co-Founder of mbuzz. With 10+ years in marketing including roles at Westpac, Avon, and Forebrite, she's obsessed with making measurement actually useful.

Harvard Extension School Forebrite Westpac Avon

How mature is your marketing measurement?

The free Measurement Maturity Assessment shows where you stand, where you're exposed, and what to fix first. 10 questions, 3 minutes.

Take the Assessment

Ready to try server-side attribution?

Set up in 10 minutes. Free up to 30K records/month.