Key-Person Risk Brief

Someone has to own it. Key-Person Risk Brief is not a quick AI prompt. It's extracting, verifying, and organizing years of one person's work into something trustworthy. Your team is already stretched, and a distracting side project is the last thing you need when a key person is walking out the door and you need answers now.

The information you need is scattered and buried. A key employee's knowledge could be spread across years of email, docs, and code. The decisions that matter may be hidden in the noise. Extracting and verifying the results could take days or weeks of work, with no assurance of finding the facts that matter now.

Consumer AI tools can expose sensitive records. Pasting a departing employee's records into an unapproved consumer tool may expose them to retention, review or training under that service's terms. KPRB uses the governed Amazon Bedrock no-retention path, with separate controlled engagement storage and review.

The data is a legal and privacy minefield. It contains regulated, privileged, and personal material: health information, legal matters, private communications. Handling it without proper scope, consent, and governance isn't just risky. It's a liability your counsel will flag.

An answer you can't verify is worse than no answer at all. Raw AI can produce confident answers with no source and no way to check results against the facts. Act on knowledge you can't trace or verify, and a confident guess becomes an expensive mistake.

That's why this is a managed service, not a tool. We own the process, the security, and the provenance, so you get knowledge you can trust, without exposing your data or your team to the risk of doing it yourself.

AI can't hold all the information at once. An LLM can only reason over what fits in its working memory. A key employee's knowledge may span years of email, documents, and code. As the content grows, the model drops earlier detail, loses the thread, and compresses away the nuance that mattered. KPRB extracts and stores the entire body of work intact, so no answer depends on what fit into a single prompt.

AI makes things up. When an LLM doesn't know, the model doesn't say so. AI generates the most plausible-sounding answer, which reads exactly like the truth. For a departed employee's reasoning, a confident fabrication is worse than a blank, because a false answer corrupts every decision built upon it. KPRB ties every answer to actual source material, so there's nothing to invent.

AI has a poor sense of time. An LLM can't reliably tell what came first, what was later reversed, or what's still true. A decision and its reversal look equally valid. Knowing the difference is often the whole point. KPRB preserves the sequence, so you see not just what was decided, but what still stands.

AI doesn't show you where the answer came from. Raw AI gives you a fluent answer with no way to check the source. You can't audit the result, defend it to a board or a court, or confirm the model even used the right inputs. KPRB binds every answer to its exact source, so you can trace it, prove it, and stand behind it.

AI doesn't know what matters. An LLM reads all text as roughly equal. The model misses that one load-bearing decision buried in a routine thread, the offhand constraint that the whole system secretly depends on. KPRB surfaces and preserves exactly those high-significance facts.

Controlled processing on AWS. Your engagement uses a dedicated AWS workspace, with AI processing through Amazon Bedrock rather than a consumer chatbot. The upload and client portal use encrypted connections.

Model-service retention set to "none". The required setting is checked at each governed provider launch before processing is allowed. This is distinct from the temporary engagement storage needed to prepare and verify your brief. Client records are not used to train AI models.

Secure upload without extra steps. We provide your engagement-specific AWS upload page. Send the approved ZIP archive without adding your own encryption or password. The connection is encrypted in transit and the material is encrypted at rest. Never send source records or credentials by ordinary email.

Least-privilege access. Access is restricted to the authorized people and processes needed for the engagement. AWS supplies the processing and storage infrastructure; Cloudflare protects named-user portal access. These infrastructure services are not permission to use your records for unrelated purposes.

Every engagement is fully isolated. Each client runs in a dedicated per-engagement workspace on AWS, created for that engagement and destroyed when it ends. Nothing is shared between clients, so there's no possibility of cross-contamination.

Verified closeout. The written delivery notice states the agreed destruction date. Final confirmation follows verified deletion of Provider-controlled engagement material, including our brief copy, and required access and infrastructure removal. If anything remains pending, we tell you what remains rather than claim completion. Retained administrative agreements, invoices and contentless verification evidence are separate from the source corpus and analytical results; any legally required retention exception is handled under the agreements.

Protected identifiers are masked before AI processing. Screening masks credentials, personal identifiers, and financial and tax identifiers before the text reaches an AI model. The surrounding business record can remain useful with those values masked. Screening is automated and is not a guarantee of perfect detection.

Sensitive topics are classified separately. After masking, the remaining material is analyzed and classified for topics such as medical information, privileged legal matters, complaints and investigations. Your written ALLOW/BLOCK category election determines what may be used in the brief and query answers. Written approval may include an eligible sensitive category; it never restores always-masked identifiers.

Evidence policy and source context are different. BLOCK prevents analytical use. Full click-to-source context remains Screen-safe but is not category-redacted. Classification is automated; human review means pre-delivery report review, not a required per-email adjudication process.

You define the scope. Information stays inside company boundaries. We work only with data the company already owns and has the right to access: its own systems, accounts, and records.

Trust works best in the open, not in secret. The strongest engagements are ones where the departing employee knows their knowledge is being preserved for the team. Handled transparently, this is a professional handoff, not employee surveillance.

Your rules govern the engagement. You remain in charge of what's collected and how the results are used. We build the process to fit your privacy, HR, and legal requirements, not to work around them.