Do I need a paid CoCounsel or Lexis+ AI account, or can a partner test with free tools?

You need the paid legal-grade product if you want the output to be defensible. The free consumer tier of ChatGPT or Claude can summarize a deposition, but it does not retrieve from a verified case law corpus and it cannot pin a citation to a Bluebook-correct source. CoCounsel, Lexis+ AI, Westlaw Precision, and Harvey all use retrieval-augmented generation against vetted legal databases, which is the part that matters when a partner signs the brief. Pricing for mid-size firms typically runs four to twelve thousand per attorney per year depending on the vendor and seat count. The honest answer: a partner can test the workflow on a free tool to feel the rhythm, but the production review work has to happen on a vendor with an audit trail, an enterprise data agreement, and a known-good citation corpus.

Is CoCounsel or Lexis+ AI privilege-safe for client documents?

Yes when configured correctly, no when used out of the box without an enterprise agreement. Thomson Reuters and LexisNexis both offer enterprise tiers that contractually exclude your inputs from training, run on tenant-isolated infrastructure, and produce an audit log your general counsel can show in a malpractice deposition. The consumer or trial tier of any AI tool is not privilege-safe. Privileged client documents pasted into ChatGPT consumer, Gemini personal, or any free chat interface have arguably been disclosed to a third party, which a clever opposing counsel will run with. The practical rule for mid-size firms: get the enterprise agreement signed before a single privileged document touches the tool, and write a one-page firm policy stating which tools are approved and for what.

How do I share AI-drafted output with attorneys who do not have access to the tool?

Export and route through your document management system. CoCounsel exports to Word and PDF. Lexis+ AI integrates directly with NetDocuments and iManage so the output lands in the matter folder with a version stamp. Harvey produces Word output with track changes ready for partner review. The pattern most firms run: associate uses the AI tool, exports the draft into iManage or NetDocuments under the matter number, partner reviews and edits inside Word with track changes on, and the AI tool itself is just the drafting surface. Avoid emailing AI output as attachments outside the DMS. That breaks the audit trail and makes a privilege defense harder if the matter ever gets reviewed by outside counsel.

What if my firm has restrictions on AI tools or the managing partner is skeptical?

The skepticism is reasonable. Three moves that work: one, propose a 60-day pilot scoped to a single practice group with a written policy and a verification protocol. Two, point the managing partner at the state bar AI opinion in your jurisdiction. As of 2026, New York, California, Florida, Illinois, and Texas have all issued formal opinions, and most are supportive of AI use with appropriate verification and supervision. Three, run the pilot on non-billable internal work first, like CLE summaries or internal know-how, before it touches client matters. Most managing partner skepticism collapses when they see the verification workflow actually working on a low-stakes matter. Do not try to install AI by stealth. That is how firms end up in the news for the wrong reason.

Can paralegals and litigation support staff use these tools too?

Yes, and they should. Paralegals running document review on a moderately large case are the highest-value users of AI document review tools in most mid-size firms. They handle volume that would otherwise consume weeks of associate time. The workflow that works: paralegals run first-pass review and tagging, associates run the second pass and pull privileged material, partners do the substantive review and verify any citation that would appear in court papers. Each tier reviews the layer below. Most enterprise vendors include paralegal seats at a reduced rate. The pushback to expect: some partners worry that paralegal AI use somehow erodes attorney supervision. The opposite is true when the audit trail is in place. Every action is logged and reviewable.

What about contract drafting with AI? Different rules?

Different stakes, similar rules. Contract drafting AI like Harvey, Spellbook, or LexisNexis MarkUp works on the structural and boilerplate layers, not on the negotiated terms. The pattern: AI drafts the first version against a playbook, the associate edits to fit the deal, the partner approves the negotiated provisions. The hallucination risk is lower in transactional work than in litigation because there are no fabricated case citations to worry about, but the malpractice risk is real on missed indemnity caps, wrong jurisdiction clauses, or stale carve-outs. The verification step is still mandatory. Spellbook and Harvey both produce comparison reports against your firm playbook, which is the move that makes contract AI defensible. Read every clause that touches risk allocation. Trust nothing the AI marks as a standard provision without checking it.

Will AI document review get me in trouble with my malpractice insurance carrier?

Most carriers now require disclosure of AI use in your annual application. As of 2026, ALPS, ProAssurance, and the major specialty malpractice carriers have AI riders that either require attestation that AI output is verified by a human attorney, or impose a small premium adjustment for firms using generative AI in production work. The riders are not punitive. They are calibrating to the new failure mode. The practical advice: call your broker before your next renewal. Ask what disclosure the carrier requires, what verification protocols they expect to see, and whether your written firm AI policy meets the bar. Firms that get caught off-guard by an AI-related claim almost always failed at the disclosure step, not the technology step. Underwriters reward firms that show their work.

How Do Mid-Size Law Firms Run AI Document Review Without Hallucinated Cites?

Q: Will the output sound generic or like every other AI brief?

Only if the prompts are generic. The associates who get the best output give the AI a specific frame: the jurisdiction, the procedural posture, the standard of review, and the firm's house style. Vague prompts produce vague summaries. A prompt like 'summarize this deposition' returns mush. A prompt like 'extract every admission relevant to the duty-of-care element under California negligence law, organized by witness, with page and line citations' returns something a partner can actually use. The other quality move is feeding the AI a sample of the firm's prior briefs as a style reference. CoCounsel and Harvey both let you do this. The voice problem solves itself when the model has a real example of how your firm writes.

The mid-size firm partner I talked to last quarter spent six weeks on a single document review for a regulatory matter. Forty thousand documents, four associates, a litigation support team. The bill on that review was just under three hundred thousand dollars. He told me he could probably have done it in two weeks with AI. He also told me he was scared to try, because of Mata v. Avianca and the way that case ended a lawyer's career.

This is the conversation in mid-size firms right now. The technology works. The verification protocol is the question.

This guide walks through how mid-size firms (50 to 200 attorneys) actually deploy AI for document review without inventing case law, breaking privilege, or ending up in a sanctions hearing. It covers the real tools (CoCounsel, Lexis+ AI, Westlaw Precision, Harvey), the verification workflow, the privilege architecture, and the malpractice insurance reality. Read this before your firm's next big matter.

Why this matters for mid-size law firms specifically

BigLaw and mid-size firms are in different boats. The Am Law 100 has dedicated knowledge management departments, a CIO with a budget for vendor pilots, and the political capital to absorb a failed experiment. A 120-attorney firm with three offices does not. The managing partner is also a producer, the IT director also handles e-discovery vendors, and a single bad story about an AI brief would carry through the local bar in 48 hours.

What changes when a mid-size firm uses AI document review well: associates stop drowning in first-pass review, partners get to the strategic questions faster, and realization rates climb because clients stop pushing back on review-time line items. Firms doing this well report 30 to 50 percent reductions in document review hours. Firms doing it badly are the ones in the news.

What CoCounsel, Lexis+ AI, Westlaw Precision, and Harvey actually do

These are not chatbots wrapped around legal databases. They are retrieval-augmented systems that combine a large language model with a vetted legal corpus, plus structured workflows for specific lawyer tasks (deposition summary, contract review, brief drafting, discovery analysis).

Three things make them different from a generic AI tool:

They retrieve from a known-good source. CoCounsel pulls from Westlaw. Lexis+ AI pulls from the LexisNexis corpus. The model is grounded in verified case law, not making it up. Hallucinated citations are not impossible but they are dramatically less likely than with a consumer chat tool.
They produce audit trails. Every query, every document fed in, every output, logged with a timestamp and a user ID. This matters when a partner has to defend the work product to a client, an opposing counsel, or a court.
They run under enterprise contracts that exclude your inputs from model training. The standard CoCounsel and Lexis+ AI enterprise agreements include this clause. The consumer tier does not. The contract is the privilege firewall.

Think of it as a senior associate who reads ten thousand pages an hour, never gets tired, and will confidently make things up if the prompt is bad. Your job is to design a workflow where the things it makes up get caught before they hit the brief.

Before you start

You need:

An enterprise agreement with a legal-grade AI vendor. CoCounsel, Lexis+ AI, Westlaw Precision, or Harvey are the four serious options for mid-size firms. The contract has to include training data exclusion, tenant isolation, and an audit log.
A document management system with AI integration. NetDocuments and iManage both have direct integrations with the major legal AI tools. PracticePanther, MyCase, and Clio are catching up but trail on enterprise AI integrations.
About 40 minutes for a partner or senior associate to run the first real session.
A real matter to test on. A finished matter you have already billed is the right starting point because you know the right answer and can grade the AI's output against it.

One thing to settle before you paste anything: privilege, work product, and the Mata v. Avianca lesson. We have a dedicated section below. It is non-negotiable. Two attorneys lost their careers in 2023 because they trusted a chat tool to retrieve case law and did not verify the citations. The technology is better in 2026 but the rule has not changed. Every citation that goes in a brief gets human-verified before the brief goes out.

Task 1: First-pass deposition summary at scale

The failure pattern: a third-year associate spends three full days summarizing a 400-page deposition, working from highlighted PDFs and a Word doc, missing 20 percent of the testimony that matters to your case theory because nobody reads with full attention for 24 straight hours.

What to ask CoCounsel for instead:

Summarize the attached deposition of John Smith, taken March 14, 2026, in Smith v. Acme Logistics (Northern District of California, Case No. 25-cv-01234). Extract every admission relevant to the duty-of-care element under California negligence law, every statement about Acme's training procedures for forklift operators between 2020 and 2024, and every reference to the warehouse safety audit conducted by Sentinel Compliance in November 2023. Organize by topic, then by witness statement, with page and line citations for each. Flag any apparent inconsistency with the witness's prior interrogatory responses (Exhibits A through D).

The prompt is doing several things. It scopes the legal frame (duty of care under California negligence law), it scopes the factual targets (training procedures, the Sentinel audit), it specifies the output structure (topic, witness, citation), and it asks for one analytical layer (inconsistency flagging). Generic prompts produce generic summaries. This kind of prompt produces a working document an associate can hand a partner.

The associate then verifies. Spot-check 15 of the citations. If any are wrong, the verification expands. If they all check out, the summary is ready for partner review. The 24 hours of associate time becomes 90 minutes of CoCounsel time plus 90 minutes of verification.

For the appellate practice variant: feed the AI the full record on appeal and ask for a chronological factual summary keyed to the issues on appeal. The retrieval and citation structure are the same. The output is the skeleton of the statement of facts.

Task 2: Privilege screening on large document productions

The failure pattern: a paralegal flags 8,000 documents for attorney review based on a basic search-term hit list, an associate scrolls through them at 90 seconds per document for two weeks, and three privileged emails slip through into the production.

What to ask the AI tool for instead:

Review the attached batch of 8,000 documents from the custodial collection of Sarah Chen, in-house counsel at Acme Logistics. For each document, classify as: (1) clearly privileged attorney-client communication, (2) clearly privileged work product, (3) potentially privileged requiring attorney review, (4) not privileged but responsive to RFP No. 14, (5) not privileged and not responsive. For categories 1, 2, and 3, identify the specific basis for the privilege claim (parties on the email, subject matter, document type). Output as a structured table with document ID, category, basis, and a 2-sentence justification.

What the AI does well: triage at speed and consistency. What it gets wrong: edge cases involving non-attorney consultants, dual-purpose communications, or post-acquisition privilege questions. The associate reviews every document the AI flagged as category 3, and randomly audits a sample of categories 1, 2, and 4. Anything in category 4 (not privileged but responsive) gets a second pass before production.

The efficiency gain: the associate's two-week review collapses to three to four days, and the privilege accuracy goes up because the AI catches the pattern-based flags the human eye misses by hour twelve.

For multi-custodian productions, run each custodian as a separate batch. The AI is better at maintaining privilege context within a single relationship than across multiple relationships in one prompt.

Task 3: Contract review against a firm playbook

The failure pattern: a senior associate marks up a 90-page commercial agreement against the firm's negotiating playbook in eight to ten hours, missing two of the partner's hot-button clauses because the playbook is in a Word doc nobody updated since 2022.

What to ask Harvey or Spellbook for instead:

Review the attached Master Services Agreement between client Mid-State Manufacturing and vendor Cloud Logistics, dated April 2026. Compare against the firm's commercial agreements playbook (uploaded). Flag every clause that deviates from our playbook positions on: indemnification scope and caps, limitation of liability, IP ownership of work product, data security and breach notification, and termination for convenience. For each deviation, provide the playbook position, the agreement language, the risk to the client, and a suggested redline. Output as a marked-up Word document with track changes and a summary memo for the partner.

The AI produces the comparison report and the redline. The associate reviews every flagged clause and verifies the suggested redlines reflect current playbook positions, not stale ones. The partner reviews the negotiated terms (price, scope, term, key business points) and the high-risk redlines.

The key prompt move: name the specific risk categories you care about. Generic 'review for risk' produces generic output. 'Indemnification caps, limitation of liability, IP ownership, data security, termination for convenience' produces a focused review that mirrors how the firm actually negotiates.

For M&A due diligence, the same pattern works at scale across hundreds of contracts. Define your three to five risk categories, run the AI batch, route the flags to associates, escalate the deal-killers to partners.

Task 4: Legal research with citation verification built in

The failure pattern: a junior associate runs a Westlaw search, picks the first 10 cases, writes a research memo, and a partner discovers in deposition prep that two of the cited cases were overruled in 2024.

What to ask Westlaw Precision or Lexis+ AI for instead:

Research the current state of California law on the apex deposition doctrine for senior corporate executives, focused on the Northern District of California and the California Court of Appeals. I need: (1) the leading authority, (2) the current standard for whether an apex deposition is appropriate, (3) the three most recent published opinions applying the doctrine (post-2024), (4) any pending appellate review or recent rule changes, (5) flag any cited authority that has been overruled, distinguished, or superseded. Provide pinpoint citations and a confidence score for each citation.

Westlaw Precision and Lexis+ AI both verify citations against KeyCite or Shepard's automatically. The AI tool returns research with a treatment indicator next to every citation. Overruled cases get flagged in red. Distinguished cases get flagged in yellow. The associate verifies the top three to five cases manually before relying on them in any brief.

This is the workflow that prevents a Mata v. Avianca outcome. The verification is built into the tool. The associate's job is to confirm the verification, not to do it from scratch. Skipping the verification step is what gets attorneys sanctioned.

For unfamiliar jurisdictions or specialized areas (admiralty, ITC, FERC), the verification is even more important because the associate has less internal sense of what looks wrong. The AI catches the obvious issues. The partner catches the subtle ones.

Task 5: First-draft motion or brief skeletons

The failure pattern: a partner spends 12 hours drafting an opposition brief from scratch under deadline pressure, produces a serviceable but uneven draft, and the associate then rewrites the structure because the partner ran out of time on the legal section.

What to ask Harvey or CoCounsel for instead:

Draft an opposition to defendant's motion for summary judgment in Smith v. Acme Logistics. The motion (attached) argues no triable issue of fact on the duty-of-care element. Use the firm's prior opposition briefs (uploaded as samples) for voice and structure. Build: (1) introduction framing the disputed facts, (2) standard of review under Rule 56 in the Ninth Circuit, (3) statement of facts citing the deposition summary (uploaded) and the Sentinel audit report (uploaded), (4) legal argument addressing each of defendant's three arguments in turn, (5) conclusion. Use the firm's preferred citation format (Bluebook, with parallel citations to West Reporter). Flag every citation with a Shepard's or KeyCite indicator.

The AI produces a structured first draft. Every citation flagged. Every section organized. The partner edits substance, the associate verifies citations, the paralegal does the cite-check. Twelve hours becomes four hours, and the structural quality is more consistent because the AI does not get tired in the middle of the legal section.

The critical move: feed the AI prior firm briefs as voice samples. Without them, the output reads like every other AI brief on the internet. With them, the output reads like your firm.

For demand letters and pre-litigation correspondence, the same pattern works at lower stakes. Feed the AI three of the firm's prior demand letters, a fact pattern, and the legal theory. The output is 70 percent of a finalized letter in 10 minutes.

Task 6: Discovery responses and objection drafting

The failure pattern: an associate spends a full day drafting responses and objections to 75 interrogatories and requests for production, copying boilerplate objections from a 2019 Word doc, missing an attorney-client privilege objection on three requests because they read the request too quickly.

What to ask CoCounsel or Lexis+ AI for instead:

Draft responses and objections to the attached Defendant's First Set of Interrogatories and Requests for Production in Smith v. Acme Logistics. For each request: (1) identify all applicable objections (privilege, work product, overbreadth, vagueness, undue burden, scope of permissible discovery, third-party privacy), (2) draft the objection language using the firm's standard objection bank (uploaded), (3) draft a substantive response based on the available facts (uploaded fact summary), (4) flag any request that requires an attorney judgment call (e.g., scope of privilege log, confidentiality designations under the protective order). Output as a Word document formatted for filing in the Northern District of California.

The AI produces the structural draft. The associate makes every judgment call. The partner reviews privilege calls and any flag the AI raised. Eight hours of associate time becomes two hours of AI work plus two hours of attorney review and judgment.

For responses to requests for admission, the same pattern works with one change: the AI is generally not allowed to draft admissions or denials without attorney sign-off, only to flag the request, identify the operative fact, and recommend the response. The judgment stays with the lawyer.

The law firm-specific prompts that actually work

After watching mid-size firms deploy these tools for the better part of a year, the difference between an AI workflow that compresses billable hours and one that creates malpractice risk comes down to four prompt moves.

Specify the legal frame. Jurisdiction, procedural posture, standard of review, and applicable causes of action. 'Negligence claim under California law on summary judgment' produces a different output than 'review this motion.' The AI is better at retrieval when the frame is explicit.

Specify the verification requirement. Tell the AI to flag confidence levels, to include treatment indicators on citations, and to mark anything it is uncertain about. 'Flag any citation that has been overruled, distinguished, or superseded' is a more useful instruction than the implicit assumption that the AI will catch it. The tools can do it. They have to be asked.

Specify the firm voice. Upload three to five examples of how your firm writes briefs, demand letters, or memos. Without samples, the AI defaults to a generic legal-writing register that does not match any specific firm. With samples, the output sounds like your firm at the first pass.

Specify what the human must verify. End every prompt with the verification list: which citations the partner must confirm, which factual claims need cross-checking against source documents, which legal conclusions need attorney sign-off. This embeds the verification protocol into the prompt rather than relying on a separate checklist that gets forgotten under deadline pressure.

The privilege and malpractice non-negotiables

This section is short because the rule is simple, but it is the most important section in this guide.

Do not put any of the following into the consumer tier of any AI tool (free ChatGPT, free Claude, Gemini personal, Copilot personal):

Privileged attorney-client communications
Attorney work product (mental impressions, case strategy, witness assessments)
Client identities tied to matter substance
Settlement positions or negotiation strategy
Witness names and substantive testimony content
Sealed or protected-order materials
Trade secrets or proprietary information disclosed under NDA

Use the enterprise-tier legal AI tools (CoCounsel, Lexis+ AI, Westlaw Precision, Harvey) under signed enterprise agreements that include training data exclusion, tenant isolation, and audit logging. These contracts are the privilege firewall. Without them, an opposing counsel can argue privilege was waived by disclosure to a third-party AI vendor whose terms permit training on your inputs.

The Mata v. Avianca case is the canonical lesson. Two attorneys filed a brief citing six fictional cases that ChatGPT hallucinated. They were sanctioned, fined, and reported to the bar. The current state-bar AI opinions, including New York's 2024 opinion, California's 2024 guidance, Florida's 2024 ethics advisory, Illinois' 2024 opinion, and Texas' 2025 opinion, all converge on the same rule: AI is permitted, but the lawyer remains responsible for verification, supervision, and competence under the rules of professional conduct. Hallucinated citations are sanctionable regardless of whether the lawyer used AI or copied from a bad memo.

The practical workflow that respects these rules: build prompts and templates in the AI tool, run all client-matter work through the enterprise-licensed product with the matter associated to the client folder, verify every citation with a Shepard's or KeyCite check before any document leaves the firm, and document the verification in the matter file. If your firm has signed an enterprise agreement with a Data Processing Addendum, the rules can be different on permitted use. Ask your general counsel or the firm's risk partner what is covered. Do not assume.

One more layer: malpractice insurance carriers as of 2026 require disclosure of AI use in production work. ALPS, ProAssurance, and the major specialty malpractice carriers have AI riders. Some require attestation that human attorneys verify all AI output. Some impose a small premium adjustment. Call your broker before your next renewal and confirm what your written firm AI policy needs to say to satisfy the underwriter.

When NOT to use AI for legal work

AI legal tools are powerful but not universal. They are the wrong answer for:

Anything client-facing without attorney verification. A client-facing letter, an email to opposing counsel, a filing with the court. AI drafts; lawyers send.
Novel legal questions in unsettled areas. AI is excellent on settled doctrine and weak on emerging case law where treatment is contested. For first-impression questions, the AI gives you a starting point, not an answer.
Anything with sealed or protective-order materials outside the licensed enterprise tool. Sealed pleadings, protected discovery, confidential settlement terms. The audit trail and tenant isolation matter most here.
High-stakes fact-finding from witness statements. AI is good at extracting and summarizing but not at evaluating credibility or noticing what a witness did not say. The judgment call stays with the attorney.

A simple rule: AI is an unfair advantage on the 80 percent of legal work where speed and structure matter. Trust the official channels, the verification protocols, and human judgment for the 20 percent where the document or decision has career or client-defining weight.

The quick-start template

Here is the prompt scaffold that works across most mid-size firm AI use cases. Copy it, fill in the brackets, paste into your enterprise legal AI tool.

[Task: summarize, review, draft, research] the attached [document type] in [matter name and case number, jurisdiction].

Legal frame: [applicable law, procedural posture, standard of review, claims at issue].

Specific outputs needed: [list 3 to 5 specific things you need extracted, drafted, or analyzed].

Verification: flag every citation with a Shepard's or KeyCite indicator. Flag any factual claim that requires cross-checking against the source. Flag any conclusion that requires an attorney judgment call.

Voice and format: use the firm's [brief / memo / letter] style as reflected in the uploaded samples. Output in [Word with track changes / structured table / formatted memo].

Confidentiality: this matter is privileged. Process under the firm's enterprise agreement only.

That is the whole pattern. For 80 percent of mid-size firm document review work, this is enough. For complex matters, extend the scaffold with matter-specific risk categories and prior firm work product as voice samples.

Bigger wins beyond document review

Once a firm has the document review workflow running cleanly, the next layer of value shows up in places that are not single matters.

Firm-wide knowledge management. Spend a quarter loading the firm's prior briefs, motions, and memos into a tagged corpus inside the AI tool. Now every associate can ask 'show me how this firm has argued the apex deposition issue in the past three years' and get an answer in 30 seconds. The institutional knowledge that used to live in three partners' heads becomes searchable and usable by the entire firm. Total time investment: one IT-led project quarter. Time saved: every research session for the next decade.

Practice group playbooks. Build structured AI prompts for each practice group's most common matters. Commercial litigation has its set, employment has its set, real estate has its set. Each playbook codifies the firm's preferred legal frame, citation patterns, and risk categories. Associates onboard faster. Senior associates supervise consistently. Partners get consistent first-pass quality regardless of which associate is on the matter.

Client intake and conflict checks. AI can run conflict-check analysis across the firm's matter database faster and more comprehensively than the existing conflict software in most mid-size firms. The output is a flagged list for the conflicts partner to review. Five-minute clearance becomes 30-second clearance with better recall on indirect conflicts.

Matter budgeting and post-mortems. Feed the AI the time entries, the matter outcome, and the client communications from a recently-completed matter. Ask it to identify which tasks took longer than expected, which billing categories overran the budget, and which workflow changes might have saved time. The firm gets better matter pricing data. The next pitch on a similar matter becomes more confident.

The law firm AI consulting connection

This is one tool category in one practice area. The bigger AI question for mid-size firms is structural. The firms that figure out where AI fits, where it does not, and how to deploy it with the right privilege architecture and malpractice protection end up with better realization rates, faster matter turns, and a competitive position against the BigLaw firms that used to win every cross-jurisdiction pitch on resources alone. The firms that wait usually end up either banning AI awkwardly, using it badly under the table, or both.

If your firm is wrestling with the bigger AI question, the AI Consulting for Law Firms page covers the full scope: where AI actually fits in mid-size firm operations, what the common failure modes look like, how the privilege architecture works, and what an engagement looks like when it works.

Closing

The goal is not for partners to become AI prompt engineers. It is for the firm to ship better work in less time without trading away privilege protections or malpractice posture. The verification workflow is what makes the difference. Without it, AI is a Mata v. Avianca waiting to happen. With it, AI is the most consequential operational lever a mid-size firm has had in 30 years.

Pick one matter. Pick the practice group most willing to test. Run a 60-day pilot with a written verification protocol and a malpractice-disclosure memo on file. Then extend it.

If you want to talk about how AI fits into your firm at the program level, the AI Consulting for Law Firms page lays out the full picture and how an engagement works.