How Dexi.io Cut Cloud Execution Costs by 60% Without Sacrificing Scale
Rebuilt the robot builder UX, elastic execution layer, and BI connector suite for a cloud-native data extraction platform used by enterprises worldwide.
Dexi.io is a cloud-based web data automation platform, part of the Engage3 group, that enables businesses to build extraction robots without writing code. Their three robot types (Extractor, Crawler, and Pipe) power data pipelines for analytics teams across retail, finance, and media. Despite strong market traction, rising cloud execution costs and a technically complex robot builder were limiting growth and customer satisfaction.
60%
reduction in cloud execution costs
40%
faster robot build time
15+
BI connector integrations live
99.7%
job completion rate (from 94.2%)
The challenge
Dexi.io's core robot builder worked but was intimidating for non-developer users, leading to high support ticket volumes and slow enterprise onboarding. Concurrently, running large numbers of extraction robots simultaneously on dedicated cloud instances was creating runaway infrastructure costs. And the BI integration layer was falling behind as Tableau and Power BI released new API versions that broke existing connectors.
Robot builder too complex for business users
The node-based builder required understanding of DOM structures, XPath, and CSS selectors to use effectively. Business analysts were opening support tickets for configurations that should be self-service.
Cloud execution costs scaling faster than revenue
Each robot ran on dedicated instances provisioned at job creation. Concurrent jobs during peak hours were spinning up hundreds of instances simultaneously, making infrastructure costs unpredictable.
BI connectors failing on API changes
Tableau, Power BI, and Looker connectors were built as one-off integrations with no shared abstraction layer. Every platform API update required individual fixes, creating a maintenance backlog.
No job-level monitoring or recovery
When long-running extraction jobs failed, there was no automatic retry logic. Failed jobs silently dropped data, which clients discovered only when their downstream dashboards showed gaps.
What we built
We redesigned the robot builder around non-developer workflows, replaced dedicated instance provisioning with an elastic Kubernetes-based execution pool, rebuilt the connector layer with a shared abstraction framework, and added comprehensive job monitoring with automatic recovery. The result was a platform that costs less to run and is faster for users to adopt.
Simplified robot builder
New drag-and-drop action palette with guided configuration for common patterns: table extraction, pagination handling, login flows, and form submission. Inline live preview shows extracted data as the robot is built, eliminating the guess-and-run cycle.
Elastic execution layer
Kubernetes-based robot execution pool with autoscaling tied to queue depth rather than individual job creation. Spot instance integration on AWS reduced average execution cost per robot run by 60% compared to on-demand dedicated instances.
Connector abstraction framework
Unified integration layer with standardised authentication, data mapping, and version management. New BI connectors for Tableau, Power BI, Looker, Redshift, Snowflake, and BigQuery built on this framework. API version changes now require a single update rather than individual fixes per connector.
Job monitoring and auto-recovery
Per-job execution logs surfaced in the dashboard, real-time progress tracking, configurable retry on transient network failures, and escalation to human review for persistent failures. Clients receive alerts before data gaps appear in their dashboards.
Cost and performance analytics
Per-robot cost tracking, execution time trends, data volume metrics, and quality scoring. Operations teams can identify which robots are consuming disproportionate resources and optimise or schedule them accordingly.
Results
Measurable outcomes delivered, not projected.
60%
execution cost reduction
Switching from dedicated instance provisioning to the elastic Kubernetes pool reduced cloud execution costs by 60% in the first billing cycle after migration. The savings funded further product development.
40%
faster robot builds
Business analyst users now build robots 40% faster on average following the UX redesign. Support ticket volume for robot configuration fell by 55% within 60 days of the new builder launching.
99.7%
job completion rate
Automatic retry with smart backoff lifted the job completion rate from 94.2% to 99.7%. Enterprise clients report virtually no unexplained data gaps in their downstream dashboards.
15+
BI connectors live
The connector marketplace now offers 15+ production-ready integrations maintained by the shared abstraction layer. Adding a new connector now takes days rather than weeks.
Technologies used
Frontend
Execution
Backend
Integrations
“Working with Kivva Tech was an absolute pleasure. They took the time to understand our business needs and delivered a platform that our customers and our finance team both love.”
Sofia Evlo
Graphic Designer, Dexi.io
Want results like these?
Tell us your problem. A senior engineer will respond with a practical approach within 24 hours.