EPAM's energy churn case study describes a model that reached production after about a year and now feeds reports used by marketers. The work combined an AWS lakehouse, a governed data catalog and SageMaker models. EPAM reports more than 90% accuracy and 4.6x model lift against contractual thresholds, plus a test-and-control retention campaign.
Four questions we ask every AI vendor
What was deployed, what did the provider implement, and what evidence covers evaluation and operation?
- What went live?A working AI system in production — not a demo or a slide.
- What did they build?Models, data pipelines, integrations or the app around them.
- How was quality measured?Evaluation before release and monitoring after it.
- Who runs it next year?Hosting, retraining and the bill for both.
Side by side
Scroll sideways on small screens →| No. | Company | AI use case | Shipped project | Published min. | Delivery | Profile |
|---|---|---|---|---|---|---|
| 01 | | Enterprises taking a predictive model from pilot into production | European energy customer churn model | Not publicly disclosed | Not documented | Profile of EPAM → |
| 02 | | Adding a conversational AI layer to a data product | Fusedash AI | $10,000+ | Not documented | Profile of CorLab Tech → |
| 03 | | Enterprises building agent-based research workflows on a vendor platform | PharmaMar multi-agent drug candidate assessment system | Not publicly disclosed | Not documented | Profile of Globant → |
| 04 | | Regulated firms deploying voice AI agents into existing operations | Atlas Credit lending AI agents | $5,000+ | End-to-end | Profile of ScienceSoft → |
| 05 | | Organisations wanting a scoped, evaluated Gen AI proof of concept | ServiceTrade Gen AI proof of concept | $50,000+ | End-to-end | Profile of SoftServe → |
| 06 | | Field-operations teams automating documentation with AI agents | AI field documentation agents for an HVAC group | $25,000+ | Not documented | Profile of Simform → |
| 07 | | Adding engineers to an AI product company's existing team | EliseAI leasing assistant platform | $50,000+ | Team extension | Profile of Vention → |
| 08 | | Adding engineers to an in-house team building LLM features | Freeplay LLM application MVP | $50,000+ | Team extension | Profile of BairesDev → |
The AI dossiers
Fusedash AI includes a chat interface for questions about data and AI-generated recommendations, with OpenAI in the listed stack. The case study does not describe model choice, prompting or how answers are checked. CorLab Tech's AI evidence rests on this one product page.
In a February 2026 announcement, Globant describes a multi-agent system for PharmaMar that is in use and runs on Globant Enterprise AI. More than 20 agents process documents and rank options, narrowing over 8,000 treatment-indication combinations to 10, according to the company. A further phase covering hypothesis generation and compliance checks is described as planned.
ScienceSoft's Atlas Credit work covered a feasibility study, architecture, development, QA and UAT support through to production. The verification agent ran as a single-branch pilot before rollout to all branches. The bioAffinity project shows integration of a client-owned model rather than model development.
SoftServe staffed ServiceTrade's Gen AI PoC with its own delivery director, project manager, analyst and data scientists. The work covered use-case workshops, a RAG component, documentation and a roadmap to production. The output was a validated approach deployed to EC2, not a production system.
Simform built five agent workflows on Azure AI Foundry that route voice, image, receipt and LiDAR inputs to specialist agents. These feed a mobile app and a supervisor portal with Procore integration. Simform reports 80% faster report creation and 60% less manual expense logging.
Vention engineers worked on EliseAI's leasing assistant, building AI-powered automatic replies and property-system integrations. Much of the work was platform engineering, such as database tuning, security and QA. Vention reports a 125% rise in conversions.
Freeplay's CTO describes BairesDev engineers joining the in-house team's sprints to build a React app, Python backend and SDKs. They also integrated OpenAI, Anthropic and Azure models. The client calls the arrangement staff augmentation, so the product belongs to Freeplay's team.
Buyers ask
How do I check that an AI vendor has shipped real systems?
Ask for a named deployment, who uses it and what the vendor itself built: the model work, the data pipeline, the integration or the application around it. Ask how output quality was evaluated before release and monitored after it. A demo is not a deployment.
Who should run an AI system after launch?
Agree in the contract who hosts the system, monitors output quality, updates prompts or models and pays the inference bill. If your team will take over, ask for documentation and a handover period.
Sources & corrections
- EPAM — European energy customer churn model — checked 17 Nov 2026
- EPAM — Published commercial information — checked 17 Nov 2026
- CorLab Tech — Fusedash AI — checked 17 Nov 2026
- CorLab Tech — Published commercial information — checked 17 Nov 2026
- CorLab Tech — Clutch profile — checked 17 Nov 2026
- Globant — PharmaMar multi-agent drug candidate assessment system — checked 17 Nov 2026
- Globant — Published commercial information — checked 17 Nov 2026
- ScienceSoft — Atlas Credit lending AI agents — checked 17 Nov 2026
- ScienceSoft — Published commercial information — checked 17 Nov 2026
- ScienceSoft — Clutch profile — checked 17 Nov 2026
- SoftServe — ServiceTrade Gen AI proof of concept — checked 17 Nov 2026
- SoftServe — Published commercial information — checked 17 Nov 2026
- SoftServe — Clutch profile — checked 17 Nov 2026
- Simform — AI field documentation agents for an HVAC group — checked 17 Nov 2026
- Simform — Published commercial information — checked 17 Nov 2026
- Simform — Clutch profile — checked 17 Nov 2026
- Vention — EliseAI leasing assistant platform — checked 17 Nov 2026
- Vention — Published commercial information — checked 17 Nov 2026
- Vention — Clutch profile — checked 17 Nov 2026
- BairesDev — Freeplay LLM application MVP — checked 17 Nov 2026
- BairesDev — Published commercial information — checked 17 Nov 2026
- BairesDev — Clutch profile — checked 17 Nov 2026
Company profiles list every source used. Spotted an error? Request a correction.