Customer cases

Real engagements. Real outcomes.

From AI and Copilot adoption to cybersecurity, automation and dedicated engineering teams, see how we help European enterprises and scale-ups turn technology into measurable results.

AI automation Company Brain Copilot Security SME & startups

70%underwriting email automated at launch

224%three year ROI on recruitment automation

490+people trained on Microsoft Copilot

24x7SOC cover for a 17,000 person group

Insurance underwriting · Automation

AI email automation for an insurance underwriting agency

The challenge

Between 40 and 60% of support effort went to repetitive underwriting email. Manual processing delayed responses, regulation required auditable workflows, and volume kept growing.

What we built

An AI powered email automation platform: classification, prioritisation and routing, a risk based decision engine with human review on critical cases, and compliance logging throughout.

70%automation at launch

Outcomes

✓Reduced manual underwriting workload
✓Faster customer response times
✓Improved compliance and auditability
✓Scalable architecture targeting 90%

Technical architecture

1Microsoft GraphEmail ingestion
2AI processing layerLLMs + RAG on Azure AI Search
3Decision engineRisk based, human review on critical cases
4Policy validation
5Automated response

Intent classification benchmark: 121 identical golden set emails

Mistral Large 3 selected

	Mistral Large 3	GPT-4.1 mini	Grok 4.1 Fast
Cost per 1,000 emails	$0.73	$0.54	$0.26
Classified, not Unknown	95.0%	86.8%	68.3%
Average confidence	0.85	0.79	0.65
Latency P50	0.97 s	1.33 s	10.85 s
Latency P95	1.86 s	1.96 s	26.23 s
API success · schema valid	100% · 100%	100% · 100%	99.2% · 100%

Grok 4.1 Fast is cheapest on paper but leaves 32% unclassified and runs roughly ten times slower at P50. This is what cost per successful task means in practice.

Stack

Azure, Azure SQL, GitHub Actions, Microsoft Graph API. Mistral Large for understanding and generation, Ministral 3B for intent and risk classification, Azure AI Search for policy retrieval.

Team and duration

4 to 5 people: AI/ML engineer, backend engineer, cloud engineer, QA engineer, project manager. Engagement ongoing.

Recruitment · Microsoft 365 · Built in house

Recruitment automation with Company Brain

The challenge

Mid sized firms pay ₹1 to 5 lakh a year or more for HRMS platforms, use half the features, and still screen and schedule by hand. Candidate data leaves the tenant, raising DPDP Act concerns.

What we built

Company Brain, inside Microsoft 365: AI generated JDs with HR approval, automated resume intake and scoring, smart shortlisting with HR keeping the final call, and interview scheduling in Teams or Google Calendar.

224%three year ROI

Outcomes

✓70% less admin work
✓₹1.8 to 3.5 lakh annual savings
✓100% data ownership inside the tenant
✓Payback in under six months

Technical architecture

1Hiring requestInitiated by the hiring manager
2SharePoint formJob request submitted
3Power AutomateWorkflow triggered
4AI pipelineGrok 3 via Azure Foundry, Document Intelligence
5Scoring and rankingAI driven
6Interview schedulingTeams or Google Calendar
7NotificationsOutlook, email, Teams
8SharePoint dashboardCentral storage and reporting

Stack

Grok 3 via Azure AI Foundry, Azure Document Intelligence, Microsoft Power Automate, SharePoint, Teams, Outlook and Calendar.

Team and duration

4 members, 3 to 4 months, 1,920 hours: solution architect, AI engineer, Power Platform developer, SharePoint specialist.

Microsoft Copilot · Adoption

Enterprise Copilot training: 8 hours of setup down to 30 minutes

The challenge

Copilot licences were in place. Usage was not. The gap was not access, it was knowing what to ask, and knowing what not to trust.

What we built

A five phase, five week programme on live Microsoft 365 tenants with real company data, customised per batch. Governance first: limitations taught before capabilities.

490+participants

Outcomes

✓90% retention across two hour sessions (120 people)
✓100+ real use cases submitted, unprompted
✓87% delivery quality improvement
✓32 batch specific questions answered

Programme structure · 5 phases · 5 weeks

1AwarenessWhat Copilot is and isn't. Responsible use.
2Core trainingTeams, Outlook, Word, Excel, PowerPoint
3Advanced agentsResearcher and Analyst. Cross app workflows.
4Hands-on labsAgent building on real workflows
5ReinforcementQ&A, office hours, adoption tracking

Client under NDA · Security operations

24x7 SOC cover for a 17,000 person NASDAQ listed group

The challenge

A German scientific technology group needed continuous security monitoring rather than business hours coverage.

What we built

Microsoft Sentinel implementation, then ongoing incident and alert management. Four dedicated cybersecurity specialists manage the environment end to end inside a 24x7 SOC.

24x7SOC coverage

Outcomes