AI matter briefings

AI matter briefings turn a matter's full state — tasks, recent messages, doc-request progress, invoices, calendar events, time entries, pinned notes — into a focused 1-page brief. Available on the Firm plan (aiBriefing flag).

The problem this solves

Before a client call, you want to know:

• Where does this matter actually stand?
• What's open, what's stuck, what's at risk?
• What should I do next?

Without AI: you click through 6 sections of the matter, scan 30 timeline entries, mentally synthesize. Takes 5-10 minutes per matter, longer if the matter is sprawling. Across a 6-call morning, that's 45 minutes of context loading.

With AI: 1 second to generate, 30 seconds to read.

Generating a briefing

On any matter detail page, scroll to the AI matter briefing card (right column, between Signatures and Activity). Click Generate.

Within 5-10 seconds, the card displays:

• Summary — 2-3 sentences bringing you up to speed. Lead is the case posture (where it stands), then pressure points (deadlines, missing docs, unanswered messages).
• Key points — 3-5 short bullets capturing the most important facts.
• Next actions — 1-3 short bullets of what you should likely do next, time-aware ("Call client re: missing W-2 before April 1" not "Follow up").

Click Regenerate to produce a fresh briefing. The previous one is replaced.

What goes into the briefing

The AI is given a structured snapshot of:

• Matter name, vertical, status, practice area, description, opened date, responsible attorney/CPA
• Client name + email
• Open + recent tasks (status + due date)
• Upcoming calendar events (next 30 days, with type)
• Recent doc-request submissions (status + date)
• Recent messages (sender, visibility, brief preview, date)
• Invoices (number, status, total, due date)
• Time entries summary (hours logged in trailing 30 days)
• Pinned notes (your firm's internal notes)

Everything else on the matter — full message bodies, document content, every line of every invoice — is intentionally NOT sent. The briefing summarizes posture, not content. Privacy + token economy both win.

Using it well

Best uses:

• 30-second prep before a client call
• Re-orienting on a matter you haven't touched in 2 weeks
• Onboarding a new associate to an in-progress matter
• Drafting an email update to the client (use it as a starting point, edit for tone)

Don't use it for:

• Legal/tax advice — the AI doesn't opine on strategy, it summarizes posture
• Privileged work product — it's a status snapshot, not a brief in the legal sense
• Anything you'd quote in a client deliverable — paraphrase, don't paste

Cost + caching

Each generation costs roughly $0.001-0.005 in Anthropic API spend depending on matter complexity. BopSuite tracks input + output tokens on each briefing (matter_briefings.input_tokens, matter_briefings.output_tokens) for visibility into AI cost per org — visible in the admin panel.

Briefings are cached at the matter level. The card shows the timestamp of the last generation; regenerating overwrites. There's no auto-staleness yet — you decide when to regenerate. A reasonable cadence: regenerate before any major meeting, or once a week for active matters.

Model + reliability

Briefings use Claude Haiku 4.5 (claude-haiku-4-5-20251001) — Anthropic's fast small model. Output is structured JSON (summary, key_points, next_actions) parsed strictly; if the model returns malformed output, the API throws a clear error rather than guessing.

Generation typically completes in 3-8 seconds. If it takes longer, it's usually waiting on the database for context gathering, not on the model.

What about more AI?

Other AI features in BopSuite (varying tiers):

• AI doc validation — runs on every portal submission (Practice+, aiValidation flag). Validates SSNs, EINs, ZIP, etc. with the right regex; uses Claude for fuzzy fields like "filing status."
• AI metadata extraction — runs on file uploads with semantic_type set on the field. Pulls out period, institution, tax year, etc. so files auto-rename to readable names.
• AI personalized nudges — Firm-tier aiNudges flag. Drafts personalized reminder emails. See Sending the first reminder.
• AI box completeness check — When a tax form is uploaded, AI checks that key boxes have values (W-2 boxes 1-12, etc.) and flags incomplete uploads.

All four use the same Anthropic Claude infrastructure. All cache via Redis where appropriate.

Privacy + data handling

Briefing prompts are sent to Anthropic's API. Per Anthropic's data-handling policy:

• They don't train on API customer data by default
• Prompts + responses aren't retained beyond 30 days
• API access uses TLS in transit

If your firm has a hard "no AI on client data" policy (some BigLaw firms do), keep your plan on Practice tier and AI features stay off — aiBriefing: false, no calls made. The rest of BopSuite functions identically.

What's next

• What is a matter? — what gets into the briefing
• Plan limits + upgrading — see which AI features come at which tier