Skip to content
Syntasa

Customer story · Technology & eCommerce

$5 million a year saved on display advertising, with algorithmic retargeting.

Lenovo used Syntasa to replace rules-based retargeting with an algorithmic model that scores every visitor's likelihood to buy, saving $5 million a year in display advertising.

Talk with our teamDownload the case study

Results

Spend on the visitors most likely to buy.

  • Display savings$5Msaved on display advertising every year
  • Scaled globally$20Mprojected savings across email, search and affiliates, reinvested in marketing
  • Test size1M+visitors in the 30-day validation, at over 95% confidence
  • Head-to-head test3 monthsof rules-based against algorithmic audiences

Context

Which of today's visitors will buy tomorrow?

Lenovo is a global technology leader. At the time of this work it was the world's largest PC vendor and the third-largest smartphone maker, with a goal of reaching $12 billion a year in online sales within three years.

Lenovo's executives asked the Global Business Intelligence team to find ways to grow online sales. One answer was to identify the visitors most likely to buy, and reach them with the content that leads to a purchase fastest.

The team framed it as one business question: can we identify today's visitors who will account for 80% of future purchasers?

The constraint

Retargeting built on rules nobody could trust.

  1. Heuristic segmentsDisplay retargeting ran on rules built from human assumptions, such as treating everyone who visited a laptop page as the same.
  2. Most of the data ignoredEach customer browses with a different intent, and rules used almost none of the other data Lenovo held on each visitor.
  3. Rules that collidedTwo rules could run on top of each other, and the team couldn't be sure which audience a person would land in.
  4. Manual and hard to sizeBuilding rules was slow, laborious work, and audience sizes were hard to control. Lenovo's BI lead called it a pretty messy affair.

Deployment

From rules to a score for every visitor.

Lenovo tested its rules-based retargeting against Syntasa's algorithmic retargeting model for three months, on data for US visitors to lenovo.com.

  1. Step 1: Model every visitor

    All of the data, from purchasers and non-purchasers alike, went into a logistic regression. The team scored 100 variables and coefficients and built confidence intervals.

  2. Step 2: Test head to head

    An A/B test compared what the rules-based and algorithmic segments delivered. The results were validated over 30 days with more than a million visitors, at over 95% confidence.

  3. Step 3: Segment by value

    Every visitor now carries a purchase-likelihood score, so the team builds high-value and low-value segments and sizes audiences to fit the campaign.

“Syntasa has been instrumental for attribution modeling, by taking complex datasets from various sources and making them actionable through customized machine learning algorithms.”

Tushar Mukherjee, Head of Global e-Commerce Analytics, Lenovo

Outcome

What changed for the business.

  • $5 million a year saved

    Targeting the high-value visitors cut waste from display advertising, saving $5 million every year.

  • Purchase intent you can predict

    The Global Business Intelligence team can now predict how likely each web visitor is to buy.

  • Audiences sized on demand

    High-value and low-value segments make it simple to grow or shrink an audience, instead of rewriting rules.

  • Scaling across global marketing

    Lenovo is taking the propensity model to email, search, affiliates and other programs worldwide, with projected savings of $20 million reinvested in marketing.

“Syntasa's platform puts the business critical problem at the center. The pipeline, scheduling, and predictive capabilities are fit for real marketing use cases, but are flexible enough to prep difficult data and apply special transformations. With Syntasa, I am able to run experiments and focus on the insights, without spending hours coding a solution.”

Meagan Gentry, former Data Scientist, Lenovo

FAQ

Questions teams ask first.

Resources

Hear it from the Lenovo team.

Case study · PDF · 4 pages

Read the full Lenovo story.

The full story, with the architecture, how it works, a timeline and the detailed results.

Download the case study

Related.

Your constraint won't be our first.

Tell us what you are working under. We will bring the closest story and the engineers who delivered it.

Talk with our teamDownload the case studyA working session with our engineers, on your data and constraints.