Six capabilities, one recurring workflow

Why six capabilities instead of one big tool

Most firms already own a general-purpose file request tool, a shared drive, and a spreadsheet tracking who still owes what. Each piece works fine in isolation, and none of them talk to each other, so a reviewer ends up doing the connective work by hand: checking whether a client uploaded the right statement, matching it against last month’s numbers, and remembering to follow up when a due date slips. Cohesio collapses that connective work into the product itself. Intake defines exactly what’s owed and by when. Collection gives the client one place to answer that ask. Extraction reads what comes back and proposes facts with a citation. Review turns those proposals into decisions a named person stands behind. Close only fires once every decision is in. Automations quietly handle the parts that don’t need judgment — reminders, routing, escalation — so a reviewer’s attention goes to the parts that do.

The reason this matters more than it sounds is that each stage produces the input the next stage needs, automatically. A requirement set at intake becomes the checklist a client sees at collection. A file the client submits becomes the source extraction cites. A citation becomes the evidence a reviewer reads. A reviewer’s decision becomes a line in the manifest. Nothing gets re-typed, re-explained, or re-requested between stages, which is usually where firms lose the most time in a manual process — not in any single step, but in the gaps between them.

A day in the life of a reviewer

Picture a reviewer named Dana at a four-person firm, logging in on the morning of the 12th. Her queue is split into three lanes: work waiting on a client, work waiting on her, and exceptions. She skips the first lane — nothing to do there yet — and opens the second, where six submissions came in overnight from clients on the East Coast.

Four of the six are routine: a bank statement, a merchant processor export, and two sets of answered questions. Each one already shows extracted totals with a page reference, so Dana spot-checks the number against the source image, sees it matches, and accepts. That takes under a minute per item because she’s confirming a citation, not re-deriving a figure from scratch.

The fifth is flagged: a payroll summary where the pay period doesn’t align with the calendar month, which the system can’t resolve on its own because it isn’t a data error — it’s a business fact about how this client runs payroll. Dana opens the exception, reads the evidence panel, sees this is a known pattern for this client’s semi-monthly pay cycle, and accepts with a note so the same shape doesn’t get flagged as a surprise next month. The sixth has no source at all: the client typed "see attached" but attached nothing. Dana requests a correction, which reopens that one requirement for the client without touching anything else already accepted.

By late morning, three more clients have completed their packs, an automation has already sent one client a reminder about a due date two days out, and Dana closes one engagement — the manifest generates, the files push to the firm’s ledger, and next month’s engagement quietly drafts itself in the background. None of that required her to open a second tool.

The architecture of trust

Cohesio is built around a simple constraint: nothing becomes a fact without a citation, and nothing becomes final without a name attached to the decision. Three ideas hold that together.

Every fact points to a source

Extracted values carry the originating file, a page number, and a version identifier. A reviewer isn’t asked to trust a number — they’re shown exactly where it came from and can open the source to confirm it in seconds. If a scan is unreadable, a figure is ambiguous, or a document simply wasn’t provided, the result comes back as unknown rather than a best guess dressed up as a fact.

Decisions belong to people, not the model

Extraction proposes; it never disposes. Only an authorized staff member can accept a submission, send it back for correction, or waive a requirement — and a waiver always carries a reason. That authority can’t be delegated to a rule or a script: automations act on named events like a pack completing or a due date arriving, not on judgment calls about whether a document is good enough.

The record outlives the conversation

Every acceptance, correction and waiver writes to an audit trail that doesn’t get edited after the fact. When a month closes, that trail compresses into a signed decision manifest — who decided what, and why — attached permanently to that period, so a question six months later has an answer that doesn’t depend on anyone’s memory.

Who Cohesio is built for

This platform is scoped deliberately for bookkeeping firms with roughly one to ten staff running recurring monthly engagements for thirty, sixty, or more clients. If your week is dominated by chasing statements, reconciling what a client actually sent against what you asked for, and reviewing the same shape of exception month after month, the workflow described above maps directly onto your practice. Agencies and consultancies running recurring retainer work fit the same shape, just with briefs and approvals instead of statements and waivers.

Who it isn’t built for

Cohesio doesn’t connect to bank feeds, file taxes, run payroll, or post accounting entries — it hands off a reviewed package to the software that does those things. It also isn’t sized for an enterprise finance department managing internal controls across dozens of subsidiaries, and it isn’t a one-off document-request tool for a single project with no recurring cadence. If there’s no next month to draft, half of what makes the platform useful goes unused.