# Context-loss audit: a 10-question score that finds where your agency re-types client context between tools

RevSmith, updated 4 November 2026. The companion to the guide [The marketing agency tech stack in 2026](https://revsmith.in/guides/marketing-agency-tech-stack). Free to copy and use; no sign-up.

## What it measures

Every agency loses client context at the same places: the handoffs where a fact leaves one tool and a person has to carry it into another. The sale reaches delivery without the scope, a call decision never reaches the task list, a budget change never reaches the report. This audit finds those handoffs for one client at a time, scores them from 0 to 20, and turns the minutes your team spends carrying context into a number you can put a cost on.

It does not measure how good your tools are. A stack of excellent tools can score 4 if nothing connects them.

## How to run it

1. **Pick one real client**, ideally a busy retainer. Do not score "the agency in general": averages hide the worst handoffs.
2. **Sit with the account manager and one person who delivers the work** for that client. The owner alone will score too high.
3. **Answer the ten questions** below. For each, pick No, Partly or Yes, using the descriptions, not your intentions.
4. **Fill in the worksheet** for every question that scored 0 or 1: the two tools the context crosses, who carries it, and the minutes a week it takes.
5. **Repeat for a second, different client** (a project client if the first was a retainer, or a newer client if the first was long-standing).
6. **Re-run it after any change to your tools**, and once a quarter.

Allow about 20 minutes per client.

## Scoring

| Answer | Points | What it means |
|---|---|---|
| No | 0 | Someone re-types it from one tool into another, or it lives in one person's head |
| Partly | 1 | It is copied by a person or an automation, and is sometimes late or stale |
| Yes | 2 | It has one home, and it travels with the work without anyone copying it |

## The ten questions

### 1. Sale to delivery
**Does the delivery lead get what was sold (services, allowances, exclusions, prices) without reopening the proposal?**
- Yes: the signed scope is recorded where the team plans the work, line by line.
- Partly: someone summarizes the proposal into the project tool at kickoff, once.
- No: the team asks sales, or opens the PDF, when a question comes up.
- Usual crossing: CRM or proposal tool → project tool.
- Fix: record scope as items (included, allowance, exclusion) in the place the team works, at the handover meeting, before the first task.

### 2. Scope to task
**Can whoever does a task see whether it is inside the retainer?**
- Yes: each task or deliverable shows which part of the scope it uses, and what is left this month.
- Partly: the account manager knows; the person doing the work does not.
- No: nobody checks until the invoice, or never.
- Usual crossing: proposal or spreadsheet → project tool.
- Fix: count deliverables and allowances per month somewhere the doer can see; label work that is extra before it starts.

### 3. Kickoff facts
**Are the client's goals, audience, brand rules, approvers and contacts written once, in one place everyone reads?**
- Yes: one client record, kept current, that the team and any AI tool read.
- Partly: a kickoff doc exists but is out of date, or copies live in several tools.
- No: facts are spread across email, the deck, chat and people's memories.
- Usual crossing: kickoff deck and intake form → docs, chat, the AI assistant.
- Fix: one client record with a named owner; every change goes there first. The [client onboarding checklist](https://revsmith.in/templates/client-onboarding-checklist) lists what to capture.

### 4. Calls to work
**Do decisions and promises from a client call reach the task list and the client record without someone re-typing notes?**
- Yes: decisions, tasks and promises from the call are filed with an owner and a date, and someone checks them.
- Partly: a notetaker writes a summary; someone copies some of it into tasks.
- No: notes stay in the notetaker or a notebook.
- Usual crossing: notetaker → project tool, client record.
- Fix: end every call by reading back decisions, owners and dates; file them within the day.

### 5. Requests
**Does an ask made in Slack, email or WhatsApp land in one list where someone checks it against scope?**
- Yes: every client ask becomes one request in one list, and someone records whether it is included, uses an allowance or is extra.
- Partly: big asks are logged; "quick tweaks" are just done.
- No: asks are answered where they arrive.
- Usual crossing: chat or email → project tool.
- Fix: one request list, one person who checks scope daily. The guide to [scope creep in agencies](https://revsmith.in/guides/agency-scope-creep) has the scripts for saying "that is extra".

### 6. Approvals
**Can you show which exact version the client approved, and when, without searching a thread?**
- Yes: every approval is tied to one version and a date, and a new version reopens it.
- Partly: approvals are in one tool for some work and in email or chat for the rest.
- No: "looks good" in a thread is the record.
- Usual crossing: review tool or email → project tool, file storage.
- Fix: one approval route per client; never accept an approval on a file without a version.

### 7. Changes to the report
**When someone moves a budget, changes a bid strategy or edits tracking, does the reason reach the monthly report?**
- Yes: each change is logged with its reason, and the report draws on that log.
- Partly: the platform's change history shows what changed, not why.
- No: the person writing the report reconstructs it from memory.
- Usual crossing: ad platform or analytics → chat → reporting tool.
- Fix: a one-line change log per client (date, change, reason, who approved), read before the report is written.

### 8. Report to report
**Does next month's commentary start from last month's targets, notes and promises?**
- Yes: the report carries targets and last month's commitments forward.
- Partly: someone opens last month's report by hand.
- No: each month's commentary starts blank.
- Usual crossing: reporting tool → doc or slide.
- Fix: keep targets and commitments in the report template itself. The guide to [automating client reporting](https://revsmith.in/guides/automate-client-reporting) covers the narrative step.

### 9. AI prompts
**When someone uses an AI assistant for this client, does it get the client's context without anyone pasting it in?**
- Yes: the assistant reads the client record itself, and only this client's.
- Partly: there is a shared prompt or project per client, kept by hand.
- No: each person pastes context from memory or old chats.
- Usual crossing: client record, docs, chat → AI assistant.
- Fix: one maintained context per client that the AI reads, with a rule that one client's facts never go into another client's prompt.

### 10. Handover
**Could a colleague run the account from the records alone if its account manager left tomorrow?**
- Yes: scope, contacts, approvers, open promises and history are all in the records.
- Partly: most of it is written down; the relationship history is not.
- No: the account depends on one person.
- Usual crossing: one person's head → everyone else.
- Fix: run this question as a drill once a quarter for one account.

## Worksheet

Copy this table for each client. Fill a row for every question that scored 0 or 1.

| # | Handoff | Score (0, 1, 2) | From tool | To tool | Who carries it | Minutes a week | Fix, owner and date |
|---|---|---|---|---|---|---|---|
| 1 | Sale to delivery | | | | | | |
| 2 | Scope to task | | | | | | |
| 3 | Kickoff facts | | | | | | |
| 4 | Calls to work | | | | | | |
| 5 | Requests | | | | | | |
| 6 | Approvals | | | | | | |
| 7 | Changes to the report | | | | | | |
| 8 | Report to report | | | | | | |
| 9 | AI prompts | | | | | | |
| 10 | Handover | | | | | | |
| | **Total** | **/20** | | | | **minutes** | |

## Bands

| Score | Band | What to do first |
|---|---|---|
| 0 to 7 | **People are your integration layer** | Fix questions 1 to 4. They feed every other handoff, and no new tool helps until they are fixed. |
| 8 to 14 | **Context travels by copy and paste** | Pick one home for client facts (questions 3, 5 and 6) and connect the tools that feed it, so the copying stops at the source. |
| 15 to 20 | **Context travels with the work** | Your handoffs are sound. Consolidate tools for cost and simplicity, not for context. |

If two clients land in different bands, act on the lower one: the agency is only as connected as its messiest account.

## From minutes to money

The minutes column is the hidden cost of your stack. To turn it into the "hours per person per week" input of the stack-cost calculator and the sprawl-cost worksheet:

1. Add the minutes a week for each client you audited, and take the average.
2. Multiply by your number of active clients, then divide by 60: that is agency-wide hours a week.
3. Divide by your number of people: that is hours per person per week.
4. Multiply agency-wide hours a week by 52 and by your loaded hourly cost (pay, taxes, benefits and overhead divided by hours worked) for the yearly cost.

**Example (made-up numbers, not a benchmark).** Two clients audited at 75 and 45 minutes a week average 60 minutes. With 15 active clients that is 15 hours a week across the agency, or 1.5 hours per person for a 10-person team. At a $50 loaded hourly cost: 15 × 52 × $50 = $39,000 a year, or $3,250 a month.

We found no study with a known method that measures this time for agencies, so your own numbers are the only ones that count.

## Use it with

- [The marketing agency tech stack in 2026](https://revsmith.in/guides/marketing-agency-tech-stack): stack maps by agency type, a priced sample stack and the stack-cost calculator.
- The sprawl-cost worksheet (`sprawl-cost-worksheet.csv`, in the same folder as this file): your tools, owners, billing units and renewal dates, plus the copying cost.

RevSmith makes this audit. It is an AI operating system for marketing agencies that keeps a Client Brain for each client and is onboarding its first agencies; the audit works the same whatever tools you use.