What is KAM?

The firm's CE knowledge,
retrieved in seconds.

Knowledge Asset Manager connects sellers to every proposal, case study, playbook, and battle card the firm has ever produced — through a natural language interface powered by Claude. Instead of navigating SharePoint folders manually, a seller types a question and gets the right asset back with a link. KAM lives on SharePoint; Claude is the retrieval layer on top of it.

3Folders
9Asset Types
6Verticals
5Platforms
6Columns
Who uses it

Three roles, three different relationships with KAM.

S
Sellers

Ask Claude a question before a call, an RFP, or a proposal. Get assets back without opening SharePoint. No upload access.

Read only
SA
SAs & Directors

Upload assets after every engagement. Write a plain-text Summary before requesting Approved status. The quality of KAM is the quality of these summaries.

Contribute
PL
Practice Leads

Approve assets and own the taxonomy. Decide what reaches sellers. Enforce the Summary gate and remove stale or duplicate entries.

Govern
What's inside

Three folders for navigation. Nine asset types across them. All tagged and searchable by Claude.

External

Customer-facing assets and partner materials. Anonymised by default. The most-retrieved folder during pre-sales.

Case Study Proposal SOW
Internal

Sales enablement, vertical playbooks, and delivery assets. Delivery-only assets are hidden from seller queries.

Playbook Capability Deck Battle Card Discovery Questions
Knowledge

Regulatory reference, platform notes, and competitive intelligence. Carries a ValidUntil date so stale intel is flagged automatically.

Architecture Pattern Knowledge Doc
Folder structure

Folders are for human navigation only. Claude queries across all folders using column filters — subfolder depth does not affect retrieval.

kam/ ← one SharePoint document library │ ├── external/ ← three subcategories: customer work · events · partner │ │ │ ├── customer_artifacts/ ← account-specific, one folder per engagement │ │ ├── [2026]-[insurance]-[wtw]-[marketing-data-lake-on-databricks]-[data-engineering]/ │ │ │ ├── estimates/ │ │ │ ├── proposals/ │ │ │ ├── sow/ │ │ │ ├── case_study/ ← added post-delivery, anonymised before tagging approved │ │ │ └── delivered_decks/ │ │ ├── [2026]-[banking]-[barclays]-[bcbs239-compliance]-[data-engineering]/ │ │ │ ├── estimates/ │ │ │ ├── proposals/ │ │ │ ├── sow/ │ │ │ └── delivered_decks/ │ │ └── [year]-[domain]-[customer]-[requirements]-[practice]/ ← one per engagement │ │ │ ├── events_artifacts/ ← market-facing content, brand and lead generation │ │ ├── webinars/ │ │ │ ├── [2026]-[ce-on-databricks]-[kpi-databricks-joint]/ │ │ │ │ ├── slide_deck/ │ │ │ │ ├── recording_link.md │ │ │ │ └── follow_up_assets/ │ │ │ └── [year]-[topic]-[partner-or-solo]/ │ │ ├── white_papers/ │ │ │ ├── [2026]-[context-engineering-in-fsi]/ │ │ │ └── [year]-[topic]/ │ │ ├── conference_talks/ │ │ │ ├── [2026]-[databricks-summit]-[ce-for-fsi-session]/ │ │ │ │ ├── speaking_proposal/ │ │ │ │ ├── session_deck/ │ │ │ │ └── recording_link.md │ │ │ └── [year]-[event]-[session-title]/ │ │ ├── blog_posts/ │ │ │ └── [year]-[topic]-[platform-or-solo]/ │ │ └── awards_submissions/ │ │ └── [year]-[award-body]-[category]/ │ │ │ └── partner_artifacts/ ← co-sell and alliance materials │ ├── databricks/ │ ├── snowflake/ │ ├── microsoft_fabric/ │ ├── anthropic/ │ └── aws_bedrock/ │ ├── internal/ ← sales enablement and delivery assets │ ├── capability_decks/ │ │ ├── ce_on_databricks/ │ │ ├── ce_on_snowflake/ │ │ ├── ce_on_fabric/ │ │ ├── ce_on_bedrock/ │ │ └── ce_on_anthropic/ │ ├── vertical_playbooks/ │ │ ├── banking/ │ │ ├── insurance/ │ │ ├── healthcare/ │ │ ├── wealth_management/ │ │ └── capital_markets/ │ ├── sales_enablement/ │ │ ├── battle_cards/ │ │ ├── discovery_questions/ │ │ ├── objection_handling/ │ │ ├── icp_profiles/ │ │ └── pricing_models/ │ └── delivery_assets/ │ ├── architecture_patterns/ │ ├── sprint_templates/ │ ├── assessment_framework/ │ └── project_governance/ │ └── knowledge/ ← regulatory reference and competitive intel ├── regulatory/ │ ├── bcbs_239/ │ ├── dora/ │ ├── eu_ai_act/ │ └── sr_26-2/ ├── platform_knowledge/ │ ├── databricks/ │ ├── snowflake/ │ ├── microsoft_fabric/ │ └── aws_bedrock/ └── competitive_intelligence/
Customer_Artifacts naming:  [year]-[domain]-[customer]-[requirements]-[kpi-practice-at-work]
Mirrors the workbench naming in the my-agent-foundry repo. Each KAM engagement folder maps one-to-one to a workbench — create both at the same time. Subfolders: Estimates/ · Proposals/ · SOW/ · Case_Study/ · Delivered_Decks/. Case_Study/ is added post-delivery and must be anonymised before Status is set to Approved.

Events_Artifacts naming:  [year]-[topic]-[partner-or-event]/ for webinars and conference talks  ·  [year]-[topic]/ for white papers and blogs. Each event folder holds slide decks, recording links, and follow-up assets together so the full context of a market moment stays in one place.

Claude ignores folder path. A webinar deck inside events_artifacts/webinars/[2026]-[ce-on-databricks]-[…]/ is retrieved by a query for AssetType: Capability Deck · Technology: Databricks regardless of where it sits. Columns drive retrieval; folders drive human navigation.
How assets are tagged

Every asset carries six columns. Claude filters by them; people browse by them. Nothing else is required.

ColumnTypePurpose
Vertical Choice (multi) The FS vertical this asset covers. Insurance · Banking · Capital Markets · Healthcare · Wealth · RegTech
Technology Choice (multi) The platform the asset is built on. Databricks · Snowflake · Fabric · Bedrock · Anthropic · Multi-Model
AssetType Choice (single) What kind of asset this is. One of the nine types across the three folders above.
Status Choice Draft → Review → Approved → Archived. Claude retrieves only Approved assets.
Audience Choice Sellers · Delivery · Both. Delivery-only assets are excluded from all seller-facing queries.
Summary Text (300 chars) The plain-text description Claude reads when deciding relevance. The most important field in KAM — see below.
The Summary field

This is the only field Claude reads when deciding which asset is most relevant to a seller's question. Plain text, two to three sentences, under 300 characters.

Good
"CE build on Databricks Unity Catalog for a UK retail bank. Reduced report generation time 60%. BCBS 239 lineage captured as a byproduct. Anonymised, reference-ready."
What it does · Outcome · Regulatory angle · Reference status — Claude can match this to five different seller questions.
Avoid
"Big bank project on Databricks — great outcome, check it out."
No vertical. No outcome metric. No regulatory or use-case context. Claude cannot match this to any structured seller query.
The gate: Status cannot be set to Approved unless the Summary field is filled. An asset without a Summary never reaches sellers, regardless of how well the other columns are tagged.
How Claude retrieves

Three steps on every query. The same pattern regardless of how the seller phrases the question.

1
Parse the query

Claude reads the seller's question and maps it to column values: vertical, technology, asset type. If the query is open-ended ("anything on BCBS 239?"), it skips the column filter and does a full-text keyword scan across all Summaries instead.

2
Filter by columns via Graph API

Claude queries the KAM SharePoint library through Microsoft Graph API, filtering to Approved assets that match the inferred column values. Delivery-only assets are always excluded.

3
Read Summaries and return results

Claude reads the Summary of each matching asset, ranks by relevance to the original question, and returns the top results with a SharePoint link. If nothing matches exactly, it tells the seller and suggests the closest alternatives — it never returns a blank response.

Can Copilot do the same?

Yes — Microsoft offers three routes to the same KAM access patterns. Each has different effort and precision trade-offs.

Route 1
SharePoint Agent
Zero dev · ~30 min

Scope a native SharePoint agent to the KAM library in one click. It appears as a chat interface in SharePoint and Teams. Sellers ask questions and it searches column values, document titles, the Summary field, and file content.

Less precise — reads all content, not just Summary. Results are ranked semantically rather than by structured column filter.
Route 2
M365 Copilot Chat
Zero dev · already there

If the SI has M365 Copilot licences, sellers already have this. Copilot Chat searches all M365 content the user can access — SharePoint, Teams, Outlook — and surfaces KAM assets alongside everything else.

Too broad. Searches the entire tenant, not just KAM. Loses the precision of a library-scoped query.
Route 3
Copilot Studio Agent
1–2 days

A custom agent in Copilot Studio that calls Microsoft Graph API directly — the same API the Claude MCP design uses. You define exact column filters, read only the Summary field, and exclude Delivery-only assets.

Requires a Copilot Studio licence. Functionally identical to the Claude MCP approach; the only difference is the model (GPT-4o vs Claude).
SharePoint Agent M365 Copilot Copilot Studio Claude MCP
Dev effort None None 1–2 days 1–2 days
Retrieval precision Medium Low High High
Column filtering Partial No Yes Yes
Summary-only read No No Yes Yes
Where it appears SP + Teams All M365 Teams + SP Anywhere
Licence needed SharePoint M365 Copilot Copilot Studio Claude API
Practical recommendation: Start with a SharePoint Agent on the KAM library — 30 minutes, zero cost, gives sellers something usable immediately. If retrieval precision is a problem, build the Copilot Studio agent with structured column logic. The Claude MCP approach is right when KAM is part of a broader Claude workflow — drafting proposals, summarising case studies, generating discovery packs — not just retrieval. Either way: build the schema once, choose the front end later.
Live sample
Ask KAM

Select an example below or type your own question. Claude queries the asset library and returns matching results with the filter it applied.