Enterprise knowledge search
A permission-aware search layer that finds evidence across company systems and returns cited answers without exposing material the user cannot access.
Retrieves permitted sources and returns cited answers
Enterprise knowledge search connects approved company repositories. For each question, it retrieves relevant passages, applies existing access rights and returns results or a cited answer.
Success means a permitted answer supported by current material. Fit is strongest where recurring questions have documented answers.
Less search time and fewer repeated requests
Employees spend less time trying keywords across tools or reconstructing old decisions. Specialists receive fewer routine requests when colleagues can find a cited answer first. Teams also repeat less analysis.
Measure these effects separately. Search time and an avoided expert interruption may describe the same episode, so adding both could double-count value. Faster onboarding needs its own causal evidence.
The primary KPI starts when the need arises and stops at an answer that is relevant, current enough, cited and accessible under policy.
| KPI | What it shows | Measurement approach |
|---|---|---|
| Time to a supported answer | End-to-end retrieval performance | Timed task set and sampled production sessions |
| Successful answer rate | Share of questions resolved with adequate evidence | User confirmation plus reviewer audit |
| Citation support rate | Whether answer claims follow from cited passages | Blind review of sampled answers |
| Expert interruptions | Whether routine questions leave specialist queues | Message sampling or specialist diary |
| Duplicate work rate | Whether teams rediscover and reuse prior work | Project retrospectives and artifact matching |
| New-hire time to proficiency | Whether onboarding becomes faster | Comparable role cohorts |
| Permission violations | Whether restricted content crosses access boundaries | Automated tests and incident logs |
| Stale-answer rate | Whether outdated sources affect responses | Sample audit against document status |
Adoption diagnoses use, not value.
Discounts broad time-saved estimates
Estimate the gross time pool:
Gross annual productivity value =
employees in scope
× hours saved per employee per week
× 52
× loaded hourly cost
× active adoption
Broad time savings are not automatically EBIT. Small fragments may improve responsiveness without changing payroll, hiring or output. Financial value needs a credible route to avoided cost, contribution profit or lower rework.
Realised annual benefit =
gross productivity value
× realisation factor
+ separately measured duplicate work avoided
+ separately measured onboarding cost avoided
Conservative fictional ROI case
These assumptions are illustrative only. They are not a benchmark, forecast, guarantee or quote.
| Input | Fictional assumption | Evidence needed internally |
|---|---|---|
| Employees in scope | 1,400 | Identity and role data |
| Time saved | 1.2 hours per person per week | Baseline study and controlled pilot |
| Active adoption | 75% | Usage telemetry |
| Loaded hourly cost | 520 SEK | Finance-approved blended cost |
| Realisation factor | 15% | Capacity, hiring or output plan |
| Year-one platform cost | 2.80 MSEK | Contracted commercial terms |
| Implementation cost | 1.50 MSEK | Scoped delivery estimate |
Gross productivity value =
1,400 × 1.2 × 52 × 520 × 75% = 34.07 MSEK
Realised annual benefit =
34.07 × 15% = 5.11 MSEK
Year-one cost =
2.80 + 1.50 = 4.30 MSEK
Fictional net first-year value =
5.11 - 4.30 = 0.81 MSEK
Fictional payback =
4.30 ÷ (5.11 / 12) = about 10.1 months
The 15% realisation factor is deliberately severe and still needs proof through avoided hiring, overtime, rework or a genuine operating constraint.
Technical design and controls
Connectors read approved content and metadata. Incremental indexing handles changes and deletions. Access-control lists, group memberships and document restrictions travel with each record. Authorisation is enforced during retrieval and before return across embeddings, caches, citations, conversation history and answer caches.
Source documents are untrusted data, never instructions. This limits indirect prompt injection. Authorisation and tool permissions remain outside the model. The model receives only permitted passages, cites material claims and declines without enough evidence. Apply output DLP and provider retention controls where required.
Chunks retain titles, timestamps, owners and source links. Keyword and semantic retrieval can be combined and reranked. Freshness rules expose dates and warn on expired material.
Build an evaluation set of real questions, including ambiguity, conflicts, missing answers and required refusals. Score retrieval, relevance, citation support and freshness. Use a permission matrix across roles, regions, teams and exceptional grants. Test direct queries, paraphrases and follow-ups against restricted material, plus revocation, freshness and deletion lag against agreed service windows.
Compare historical questions under both processes using one rubric. Monitor connector failures, indexing and ACL lag, unsupported claims and stale citations. Privacy-respecting logs should reconstruct the user, query, passages and permissions behind an answer.
Where to start
Choose one domain with repeated questions, credible documents, named owners and baseline tasks. Connect necessary sources, normalise permissions and release to a small group.
Set thresholds before the pilot for supported-answer time, citation support, freshness, permission tests, revocation and deletion lag. Expand only after meeting them for an agreed period and confirming reliable ownership and access rules in the next domain.
Do not build when:
- Most valuable answers depend on undocumented judgment or sensitive political context
- Source permissions are inconsistent or cannot be synchronised reliably
- Documents are mostly obsolete, duplicated or ownerless
- Search demand is too low to justify connectors and governance
- A configured feature in an existing platform meets the tested need at lower total cost
- Leadership expects distributed minutes saved to appear automatically as EBIT
In those cases, first clean sources, redesign access, capture expert knowledge or test a narrower assistant.
Sources and methodology
The value model and implementation recommendations are Epicube analysis. The worked economics are illustrative. These independent references inform the retrieval, evaluation and permission controls:
- NIST TREC 2025 RAG Track documents evaluation work for retrieval-augmented generation
- OWASP: Vector and Embedding Weaknesses covers access control, data leakage and retrieval risks
- NIST AI Risk Management Framework provides a general framework for governing and measuring AI risks