Turning Senior Engineers' Expertise into a Knowledge Base: First-Contact Resolution Up About 15 Points

A company providing technology products and long-term after-sales support to business customers had built up a large body of customer issues, product documentation and technical solutions.

Yet customer service still depended heavily on a handful of senior employees.

When new staff ran into a problem, they had to search back and forth through old emails, chat groups, the internal wiki and product documentation.

Even if a customer had raised the same question before, a different employee might investigate it all over again.

The company had a wealth of experience, but had not turned it into a stable, maintainable body of organizational knowledge.

The challenge

Scattered knowledge sources

Product manuals, technical solutions, operating procedures and past customer cases were stored in different tools.

Without unified classification and version control, staff found it hard to tell which document was the latest.

Customer issues were disconnected from knowledge assets

The ticketing system recorded problems and resolutions, but these were never consistently distilled into reusable standard solutions.

Direct AI search could create permission risks

Information about different customers could be commercially sensitive, and some internal documents should not be open to all employees.

Without access controls, a unified knowledge search could risk information leaks.

SHEPHARD's digital solution

Step 1: Take inventory of knowledge assets

We identified product documentation, standard operating procedures, historical tickets, customer service records and internal technical documents.

These were classified by knowledge type, product version, applicable customer and business scenario.

Step 2: Establish knowledge governance

Each piece of knowledge content has an owner, a scope, a version, a review status and an expiry date.

Valuable resolution experience from historical tickets becomes a formal knowledge article only after review.

Outdated or superseded documents are clearly marked so they are no longer used as a basis for current work.

Step 3: Build permission-aware knowledge search

Combining structured search with RAG (retrieval-augmented generation), employees can query internal knowledge in natural language.

Search inherits each user's access rights, restricting access to information across customers and departments.

AI answers come with the corresponding document sources and version information, making human verification easy.

Step 4: Connect CRM and service tickets

The system retrieves relevant solutions based on the current customer, product model, issue type and service history.

Support agents can see similar cases, standard troubleshooting procedures and points to watch directly.

Step 5: Continuously improve knowledge

The system records which issues can be resolved with existing knowledge and which still need escalation to experts.

Frequently unresolved issues are fed back to knowledge owners, who keep adding to and revising the content.

For actions with clear authorization and verifiable rules, knowledge search can be further connected to business tools, while high-risk actions still require human approval.

Implementation roadmap

We started with a pilot to organize and search knowledge for one category of high-frequency product issues.

A manually labelled question set was used to test retrieval hit rate, answer accuracy, citation quality and permission isolation.

Once acceptance criteria were met, the solution was gradually extended to more business scenarios.

Business value

The company has turned individual experience into reviewed, shareable organizational knowledge.

New staff can find reliable information faster, and senior employees spend less time answering basic questions repeatedly, freeing them for more complex customer needs.

Project results (approximate): first-contact resolution about +15 points; average handling time about −35%; answer citation accuracy about 95%.

Metrics tracked on an ongoing basis: first-contact resolution (FCR), average handling time (AHT), knowledge search effectiveness, knowledge reuse rate, answer citation accuracy, permission violation incidents.

What this case teaches us

An enterprise knowledge base is not about uploading every file to a large language model.

Effective knowledge intelligence requires ongoing content governance, permission management, version control and quality evaluation.

Only when knowledge sources are reliable can AI become a trustworthy assistant for employees.

*To protect client confidentiality, company details in this case study have been anonymized, interface screenshots are illustrative, and results are rounded approximations.*