How It Works

Built from your firm, not from a feature list.

What we believe, how we build, and how grounded drafts, source checks, and human review reduce AI errors. The process below shows what you prepare, what we test, and what you keep.

Request Your Free Systems Audit See the Process

§01The Gap

You bought the software. The work still lands on people.

Your current tools may already save time. The gap is the manual work that remains: intake entered twice, letters assembled by hand, and follow-ups held in someone’s head. AI output can still need grounding, checking, and integration. We map that residual work before recommending a build, without counting gains your tools already deliver twice. The aim is defined back-office work done with review gates, so your people can focus on judgment.

2.9 hrs

Billable time in an 8-hour day

The Clio Legal Trends Report figure is context, not a recovery forecast. Nonbillable time includes work that still needs people; the audit separates that work from the residual manual tasks a system could handle.

60–70%

Estimated task potential, not firm savings

McKinsey’s estimate concerns tasks that absorb working time, not a promise of hours your firm will recover. The audit scopes specific workflows after accounting for current tools, human review, and permissions.

$60K/yr

Illustrative junior-hire cost

An illustrative $60K annual junior-hire cost is a planning comparison, not a verified cost for your firm. A system augments people rather than replacing their judgment. Compare the build, optional care, and separate software or usage costs against your own staffing figure.

§02The Failure Modes

AI, with controls for failure modes.

Invented citations, actions outside scope, and unapproved tools are risks to design for. Grounded drafts, source checks, deterministic rules, visible uncertainty, and human review reduce those risks. Errors remain possible; review is part of the workflow.

Failure mode	Why it happens	How the architecture addresses it
Hallucinated output	Without source grounding, a model can fill gaps with plausible-sounding invention.	Drafts draw from approved templates and sources, with source checks, deterministic rules, and uncertainty made visible. Development, regression, and acceptance tests use synthetic fixtures. Your reviewers approve client-facing output; residual errors remain possible.
Rogue agents	An assistant given broad, standing access can improvise scope, take actions nobody asked for, or act outside its mandate.	Each agent has a defined job, scoped permissions, and logged actions. Deterministic controls and approval gates constrain actions, including spending and external communications, to the written scope.
Subscription sprawl and generic output	A tool may need configuration, source grounding, and integrations before it fits a particular workflow.	We build around useful existing tools and the residual work the audit identifies. You own the delivered workflows and configurations; applicable provider licenses and usage costs remain separate.

§03The Process

How we forge results.

Four phases, in order. You keep something in writing at the end of each one, whether or not you continue to the next.

PH/01

Discover

01 · Systems Audit. A focused 30-minute conversation. Same-day written findings. Your report arrives the same business day once the call and any agreed outstanding inputs are complete. No obligation to build. Have your current tool list, team size, and top two repetitive workflows ready. Estimator results are helpful, not required. We confirm any missing inputs at the end of the call. You receive a written systems recommendation and, if a build fits, a recommended band and price. Timing uses Eastern Time and business days, Monday-Friday excluding U.S. federal holidays. If both requirements are completed on a non-business day, delivery is the next business day. Send contact details and a brief workflow description to arrange the call, not client records or credentials. An NDA is available before confidential discussion. The initial reply is not the completed audit.

inputs: your current stack, a walkthrough of a working week → outputs: a firm systems audit with capacity math shown line by line

You holdThe audit, yours to keep, no commitment attached.

PH/02

Wire

02 · Blueprint. The audit recommends a band; you choose the build. The blueprint fixes scope and price before work begins, connecting approved existing tools and supported open-source workflow components as the agreed blueprint requires. Full self-hosted or on-premises model hosting and hardware management are outside the standard build. It records separate vendor costs, approved data flows, and operating responsibilities.

inputs: the systems audit, your access checklist → outputs: a build blueprint with scope, acceptance criteria, and a fixed price

You holdThe blueprint: scope, price, and acceptance criteria, signed before build starts.

PH/03

Forge

03 · The Build. Development, regression, and acceptance tests use synthetic fixtures against written criteria. The build takes four to six weeks from kickoff, depending on complexity. Kickoff is the next business day after the agreement is signed, the signing payment has cleared, the checklist is complete, and access is verified. Operational activation requires the written agreement and client permissions.

inputs: the signed blueprint, templates, reviewer sign-off per system → outputs: a working system, accepted against written criteria

You holdAn acceptance ledger: each system checked off against the criteria you signed.

PH/04

Grow

04 · Care. Every band hands over scoped configurations, workflows, prompts, operating docs/runbook, and access-control ownership. Optional care covers monitoring, tuning, corrections, and refreshes within agreed hours, month to month and priced by band. Work directly with the founder. Send care requests anytime. Routine requests receive a response within one business day. Business hours: Monday-Friday, 9 a.m.-6 p.m. Eastern Time (America/New_York), excluding U.S. federal holidays. System-blocking incidents take priority, with best-effort assistance outside business hours. A response acknowledges the issue and establishes the next step; it is not a resolution deadline. Resolution depends on the issue and any upstream provider. This is not a 24/7 staffing guarantee. Included care hours and prior written approval for additional work still apply. Changes are reported in writing.

inputs: the accepted system, ongoing corrections → outputs: a monthly care report, approved knowledge-base and workflow corrections, not model training

You holdThe keys, the runbook, and a monthly record of what got sharper.

The result: defined back-office workflows on your stack or infrastructure you control, built in four to six weeks from the qualified kickoff above, with human review and operating docs. The build is a one-time fee. Synapse Forge’s ongoing care plan is optional. Existing software subscriptions and any cloud or usage charges are separate; the blueprint lists the costs that apply to your build.

§04The Paper Trail

What you hold, in writing.

Every engagement runs on a paper trail. These fictional samples illustrate the formats, not customer results. All sample figures and acceptance cases are synthetic; the actual scope, tests, and responsibilities are written for your build.

01Firm Systems Audit

Where your hours actually go, mapped system by system against your own numbers, with the capacity math shown line by line.

precedes any commitment · fictional sample: Whitfield & Marsh LLP Open the sample →

02Build Blueprint

Scope, systems, acceptance criteria, and the fixed price, in writing, before the build starts. No number, no invoice.

the contract-grade document · fictional sample Open the sample →

03Monthly Care Report

What the machine did, what was corrected, what got sharper. Accountability for the system, monthly, in writing.

under the optional care plan · fictional sample Open the sample →

04Pre-Build Readiness Checklist

Access grants, templates, reviewers, and an owner for each preparation item, with hands-on person-hour estimates and secure handover instructions. Elapsed lead time for access approvals and template agreement is separate. Preparation depends on permissions and readiness; there is no assumed one-hour total.

sent with the signed agreement · included with every engagement

05Data Boundary Sheet

A factual summary of the approved processing boundary and sign-offs for your build. Provider, region, retention, and training terms must be documented and verified against agreed data requirements before activation; client-owned accounts alone do not establish those terms.

boundary approved before activation · final as-built sheet at acceptance · included with every engagement

Illustrative approval sequence: storyboard only

Synthetic storyboard. No system is executing these steps here, no real message is sent, and no performance result is claimed.

01 / Bounded inputAn invented intake record has incomplete fields. No client records are used.
02 / Visible uncertaintyA draft follows an invented template. An unsupported statement is flagged, not presented as fact.
03 / Human correctionA reviewer corrects the draft. It remains queued.
04 / Silence is not consentAn expired review window triggers an internal reminder. The client-facing send stays blocked.
05 / Explicit approvalA reviewer approves current content and recipient. The storyboard illustrates the audit record; it does not send anything.

FAQ

The build, explained.

An operating system for judgment work: software workers, built inside accounts you already control, that do defined back-office jobs. They open and route intake, draft documents from your templates, file meeting notes to the right matter or project, and chase follow-ups. Each system is tested with synthetic fixtures through written acceptance. Operational activation requires the written agreement and client permissions. Every band includes scoped configurations, workflows, prompts, operating docs/runbook, and access-control ownership.

Any business that trades on judgment: law firms, CPA firms, consultancies, and increasingly the founder working alone. If your hours leak into intake, documents, meeting notes, and follow-up, the math works the same, only the scale changes. One honest caveat: builds carry a $12,000 minimum, so for very small operations the numbers sometimes say not yet. The free audit tells you either way, and the findings are yours regardless.

Your existing tools stay where they help. We scope approved integrations and supported open-source workflow components on infrastructure you control. The blueprint confirms the components and prerequisites; not every configuration is offered. Full self-hosted or on-premises model hosting and hardware management are outside the standard build. Applicable license, infrastructure, provider, and usage dependencies remain. The blueprint lists data flows, costs, and responsibilities, and providers must satisfy the agreed data requirements before activation.

Four to six weeks from kickoff, depending on complexity, with the schedule set in the blueprint. Kickoff is the next business day after the agreement is signed, the signing payment has cleared, the checklist is complete, and access is verified. The checklist itemizes access, templates, reviewers, owners, and hands-on person-hour estimates. Elapsed approval time is separate and depends on permissions and readiness.

Authorized people, integrations, and providers within the approved processing boundary. Provider, region, retention, and training terms are documented in the blueprint and verified against agreed data requirements before activation. Client-owned accounts alone do not remove processing or retention. Synapse Forge does not pool, resell, or use client data for model training. The build is designed to support your firm’s confidentiality obligations, not guarantee privilege or compliance. An NDA is available before confidential discussion.

Seen Enough

See where your hours actually go.

Request Your Free Systems Audit →