How the work is run
The standard, and the person who keeps it.
One rule produces most of the difference between reporting that gets used and reporting that gets ignored: every figure states where it came from, what period it covers and when it was last checked. One person is responsible for keeping that true: Mikee Shattuck, who sets the systems up and then runs them. Everything below follows from taking that seriously.
If a number cannot name its source and its window, it does not go in front of a decision. That holds for a client dashboard, a weekly brief, a board slide and this website. It is also the reason the figures on this site carry check dates: the standard is easier to trust when it is applied to the person selling it.
Count what fails to join.
Records that do not reconcile are the most useful records in the system. They are reported as a number, not smoothed away with an estimate, because the size of the gap tells you how much to trust everything else.
Publish a range when a point would be a lie.
Some measurement genuinely cannot produce a precise answer. A range with the assumptions attached is more useful than a decimal point nobody can defend.
Make the pipeline fail loudly.
Every feed gets a freshness expectation and an alert when it misses. Silent failure is the default state of every marketing data pipeline, and it is the reason dashboards drift out of use.
Write down every assumption.
Definitions, exclusions, time zones, what a conversion means, which records were dropped and why. Six months later the documentation is the difference between a working system and an archaeology project.
Build to be handed over.
No proprietary layer only we can open, no reporting held inside our tooling. Runbooks, a data dictionary and access transfer are part of the deliverable, not an upsell.
Report the misses.
When a system breaks, it goes on the status line. The Ops agent on this site has been down since 5 September and that is printed above, because a supplier who only reports green teaches you to ignore the report.
Never bill a percentage of spend.
Fees are fixed to scope. A percentage of media spend pays the supplier more when the client spends more, which is the wrong incentive in a practice whose job includes deciding to spend less.
Operate it, don't just install it.
A system handed over to nobody in particular decays inside a quarter, however well it was configured. So the same person keeps watching it: health checks read, alerts answered, budget moved, the brief written. There is no account manager in between and no junior team inheriting the keys.
Say no to the wrong work.
No creative campaigns, no PR, no standalone media buying without measurement. Taking work outside the practice would mean doing it worse than someone who specialises in it.
A rhythm one person keeps.
Weekly brief
Mondays
What changed, what it means, what to do. Drafted from the same data as the dashboard, read by the operator before it goes out, and short enough to be read in full.
Monthly review
First week
Cost per outcome by channel and market, budget allocation for the coming month, and any measurement that moved out of tolerance.
Quarterly review
Every quarter
Whether the measurement itself still fits the business: new offices, new services, new channels, and the assumptions that no longer hold.
Anything broken
Immediately
A feed that goes stale, a flow that errors, a tracking change that breaks a conversion path. Flagged the day it is detected, with the impact stated.
Least access, clearly scoped.
During the diagnostic
- Read-only access to ad platforms, CRM and analytics.
- Nothing is written, changed or published.
- Exports are limited to the figures that appear in the findings memo.
- No client data is used to train anything, and none is retained after the engagement.
During a build
- Access scoped to the properties named in the scope, nothing broader.
- Credentials held in a password manager, never in documents or chat.
- Two-factor authentication on every account, with named-user access rather than shared logins.
- Infrastructure on accounts you own. We are a guest in them, not the owner.
- Every access grant is listed and revoked at handover.
For Collins OS this looked like putting the entire application behind a login: on 4 September 2026 all 69 paths tested were reachable anonymously, and by 23 September 2026 all 69 refused. Server backups went from none to weekly in the same period. Both changes were verified with the same request used before them, which is the only way a security claim means anything.
Your stack, run by a person who can hand it over.
Where your existing tools can carry the load, they are used and configured properly: your CRM, your automation platform, your analytics, your spreadsheet if that is genuinely the right answer. Where something has to be set up, it is set up to be run by a named person and maintained by somebody other than us.
Current working set: Google Ads, Local Services Ads, Meta, YouTube and LinkedIn; HubSpot, Lawmatics and Rocket Matter for CRM and intake; Google Analytics and Google Business Profile; n8n, Zapier and Make for automation; OpenAI and Claude models, with MCP tooling for agent steps; Next.js, React, Postgres, nginx, systemd and Ubuntu servers where a purpose-built application is the honest answer.
- Note no tool is recommended because of a referral fee
- Note no proprietary reporting layer
- Operator the person who sets it up is the person who runs it
Next
Hold us to this standard.
If a figure in your reporting cannot name its source and its window, that is the first thing worth fixing. The standard is free to adopt — and the free training is where the tooling to keep it gets built.