Customer story · Google Cloud
Replenit unlocks 360% quarterly growth with its AI-native decision engine on Google Cloud.
Replenit uses Google Cloud to anticipate individual shopper needs, cutting costs by 90% and driving 360% quarterly growth. This report walks through what was built, on which products, and what it changed.
360%
business growth in one quarter
with Google Cloud
95%
faster pipeline processing
with Gemini platform
90%
lower pipeline costs
using Gemma on Cloud TPUs
4 hrs
to ship a new production pipeline, instead of 3 days
Google Cloud customer story
About this report
The problem
Broad customer segments and rigid automation rules blast shoppers with irrelevant, mistimed messages. Replenit built an engine that decides for each one instead.
Messages that miss the moment desensitise shoppers, lower conversion rates and waste marketing budgets. Replenit built an AI decision engine that enriches limited retail data into structured memory for the customer, the product and the brand, applies the skill matched to the KPI you are targeting, and owns the workflow end to end: replenishment, cross-sell, engagement, churn and winback.
Exhibit 1
Predicting and handing off, against deciding and owning
What matters
Legacy marketing tools
Replenit decision engine
How a shopper is seen
As a spreadsheet average, inside a broad segment
Through structured memory for the customer, the product and the brand
What drives the next move
Predictable, rigid automation rules
The skill matched to the KPI you are targeting
Who owns the workflow
One system predicts and hands off to another
The engine decides the next move for every shopper and owns the workflow end to end
What the shopper gets
Irrelevant, mistimed messages
The right information at exactly the right moment
The platform
Replenit’s small team needed less platform work and more product work, plus the data governance to win enterprise clients.
In its early stages Replenit ran on a third-party cloud provider. Managing the platform took heavy manual effort and drew hours away from product development. The architecture also struggled with large, multi-layered data operations, and testing and iterating on model experiments created significant engineering overhead. The team chose Google Cloud for its intuitive setup, the data governance needed to win enterprise clients, and dedicated startup resources.
Exhibit 2
Before and after the move
What changed
On a third-party cloud
On Google Cloud
Platform effort
Heavy manual effort, hours drawn away from product development
An intuitive setup that optimises engineering hours
Large, multi-layered data
The architecture struggled to handle it efficiently
BigQuery gathers, normalises and enriches it into unified records
Model experiments
Testing and iterating created significant engineering overhead
New production-ready pipelines ship in as little as four hours instead of three days
Google for Startups Cloud Program
Through the program the founder team gained technical guidance, workspace resources and shared expertise, which let the startup move fast and migrate its workloads with confidence.
“Google Cloud gives us the security infrastructure we need to protect sensitive data at scale. At the same time, the Google for Startups Cloud Program gave us the kind of technical and commercial support and resources that are hard to find as an early-stage founder.”
The engine
Replenit sits on top of a brand’s existing technology stack and decides which shopper to contact, when, and with what context. Marketing teams no longer build campaigns by hand.
Exhibit 3
From raw data to a ready-to-use instruction
Brand’s existing stack
Customer data and product details
Complex, fragmented retail data, as the brand already holds it.
BigQuery
Gather, normalise, enrich
Turns customer data and product details into unified records.
Gemma on Cloud TPUs
Cost-effective, high-volume processing
Open-source Gemma models that Replenit hosts and fine-tunes on Tensor Processing Units generate customised insights in minutes.
Gemini platform
A multi-agent workflow
Specialised agents each run a pipeline stage. Gemini Pro handles deep contextual reasoning; Gemini Flash handles high-speed, volume-heavy execution.
Brand’s channels
Ready-to-use instructions
Which shopper, when, and what context, sent directly to the brand’s communication channels.
Memory for the customer, the product and the brand is built in the pipeline; the skill matched to the KPI being optimised is then applied for every retailer on the platform.
What the shopper sees
A shopper might see a compatible skincare item suggested precisely when their routine requires a restock.
Built-in compliance from Google Cloud, which carries certifications such as SOC 2
- SOC 2
- Data separation
- Tenant isolation
- Auditable decisions
This is how Replenit meets what enterprise retailers demand.
“We built an engine that runs hundreds of specialized agents simultaneously to read and enrich retail data. Using BigQuery and Gemini platform, we can cleanly orchestrate these complex agentic workflows, turning raw behavioral signals into precise, actionable intent.”
The results
Optimising data operations on Google Cloud cut pipeline time and cost, and freed resources for product innovation.
With Gemini platform pipelines the team reduced pipeline execution times by more than 10x. Combining self-hosted open models with services from Gemini also cut overall pipeline costs by 10x, from over $200 to just $20 per run. That 90% reduction lets resources be reinvested in the product, and new pipelines ship in hours, so the team can test new models and run experiments continuously instead of waiting days for results.
Exhibit 4
Results on Google Cloud
360%
business growth in one quarter
with Google Cloud
95%
faster pipeline processing
with Gemini platform
90%
lower pipeline costs
using Gemma on Cloud TPUs
Exhibit 5
The same work, at a fraction of the cost and time
Cost per pipeline run
Before
$200+
On Google Cloud
$20
Time to ship a new production pipeline
Before
3 days
On Google Cloud
4 hours
Bars are drawn to scale, with three days read as 72 hours.
What comes next
Clearing enterprise security checks and running a flexible, low-cost engine let Replenit scale quickly, and the same architecture can travel.
Exhibit 6
Why the engine scales
Enterprise-ready
360% growth in a single quarter
Replenit could confidently onboard enterprise retailers at scale, because it could clear stringent enterprise security checks.
Every retail type
No vertical-by-vertical path
The platform enriches data flexibly, without industry limits, so Replenit expanded its total addressable market immediately across all retail types.
Replenit Labs
The same architecture, new sectors
Replenit Labs is replicating this automated architecture, scaling and fine-tuning identical autonomous decision-making systems for other fast-moving sectors.
- Banking
- Airlines
- Insurance
- Automotive
“Our focus is on building AI systems that remain reliable and cost-effective no matter how much a business grows. The architectural flexibility we now have means we can confidently expand beyond retail and take our decision engine into new global industries.”
At a glance
Maestro, its AI CRM Manager, runs on that engine, owns lifecycle workflows, and decides 1:1 for every customer, powered by memory, skills and data enrichment.
Industries
Retail, Startup
Location
Poland
Google Cloud products
- BigQuery
- Gemini
- Gemini platform
- Gemini platform pipelines
- Gemma
- Google Cloud
- Tensor Processing Units (TPUs)
The team
Ilyas Kurklu
CEO
Egemen Akdan
CTO
Omer Ozden
CPO
Caner Demir
CMO
Cenk Karacaev
CRO
Alp Karacaev
CCO
See the engine at work on your own retail data.
Maestro runs on this engine: it owns the lifecycle workflows and decides 1:1 for every customer.