Internal tickets from 850 to 270 a month in 6 months.
Every department had its own documents, wikis, and tribal knowledge, and employees Slacked a colleague every time they needed an answer. KeyDelta built an AI assistant that gave everyone instant access to HR policies, product specs, sales playbooks, and live business intelligence. The assistant kept earning trust because we fixed the operating layer first: clear ownership of each knowledge source, a governance cadence that keeps answers current, and accountability for accuracy, the discipline that separates a living system from an abandoned wiki.
Situation
A mid-market cloud services provider with 500+ employees had a knowledge problem disguised as a people problem. Every department (HR, engineering, product, sales, marketing, finance) had its own documents, wikis, and tribal knowledge. When employees needed answers, they Slacked a colleague, filed an internal ticket, or just guessed. The result: constant interruptions, inconsistent answers, and a growing internal support burden that pulled senior people away from high-value work.
- HR fielding 200+ repetitive policy questions per month: PTO, benefits, expense policies, onboarding steps
- Sales reps couldn't find current product specs or competitive positioning without asking product managers
- New hires took weeks to become self-sufficient, tribal knowledge lived in people's heads, not systems
- Marketing and sales using outdated collateral because the latest versions were buried in shared drives
- Finance and ops leaders navigating Power BI dashboards to answer questions that should take 10 seconds
Approach
KeyDelta built a conversational AI assistant that connected to every content source in the company and gave every employee a single place to ask anything.
- 01
Content ingestion and knowledge graph
Indexed content repositories across HR, product, sales, marketing, engineering, and finance. Built a unified knowledge index that understood relationships between documents, policies, products, and processes.
- 02
Copilot Studio and Power BI integration
Deployed on Microsoft Copilot Studio for natural language chat with deep Microsoft 365 integration. Connected Power BI for real-time business intelligence queries, revenue, service metrics, and operational KPIs accessible through plain English questions.
- 03
AWS Lambda backend intelligence
Complex queries requiring Python-based ML inference and multi-source synthesis routed to AWS Lambda, using the company's existing AWS infrastructure and ML libraries not available in Azure Functions. Serverless architecture meant costs scaled with usage, not headcount.
- 04
Department-specific tuning and rollout
Tuned response accuracy per department. HR got policy-specific citation and compliance guardrails. Sales got competitive intel with freshness dates. Finance got BI answers with drill-down links. Each department validated before launch.
Stack
Outcome
Over six months, internal tickets fell from 850 a month to 270 and answer time dropped from hours or days to under 8 seconds, with 84% of questions resolved on the first ask. Employee NPS rose from 32 to 61, customer CSAT from 4.0 to 4.4, and the company avoided $420K a year in headcount. Builds like this now ship under the KeyDelta Method and stay current under AAMP.
| Measure | Before | After | Change |
|---|---|---|---|
| Internal tickets. 68% reduction in internal asks | 850/mo | 270/mo | −68% |
| Answer time. From waiting to answers | Hours/Days | <8 sec | Instant |
| First-ask accuracy. Resolved without follow-up | N/A | 84% | 84% |
| Employee NPS. Staff love instant answers | 32 | 61 | +29 pts |
| Customer CSAT. Faster staff = faster customer | 4.0 | 4.4 / 5 | +10% |
| Cost avoidance. 3.5 FTEs redirected to high-value | $0 | $420K/yr | $420K |
“We had 500 people constantly interrupting each other for answers that already existed somewhere in our systems. The AI assistant didn't replace anyone, it freed everyone. Employee NPS jumped 29 points. New hires onboarded in 8 days instead of three weeks. We avoided $420K in headcount by redirecting people from answering repetitive questions to actual high-value work. Our customers noticed too, response times dropped because our people had answers at their fingertips.”