Insights / AI in your business
What we built for our own business
We mapped our own business before we offered to map yours. HandBuiltBrands is a small, distributed team that uses AI every day. We ran the same phases on ourselves that we run for clients: map the workflow and assign value, make the fix, tune it. Here’s what that produced. The phases are described in how we build your AI system, step by step.
What the map and the scoring covered
- 101 steps mapped, from a first call to a monthly report.
- 12 candidate fixes, scored on four factors.
- 4 factors: how often, how many hours, how risky, how ready.
- Every hour, our data pipelines get checked, and safe failures get fixed.
The map, the scoring, and what got built
The map
The job: See the whole operation at once, from the first call with a prospect to the monthly report.
What we built: A map of 101 steps. For each one: who or what does it, which software, how much a machine already does, and where a person has to approve.
Where a person stays in: Any step we could only infer was marked for a person to confirm.
The scoring
The job: Decide what to fix first.
What we built: A score for each of twelve candidate fixes, from 1 to 3 on four factors: how often the step happens, the hours it takes, what it risks, and whether a fix already works somewhere else in our business.
Where a person stays in: The top score went to the monthly client reports, and we chose to fix them first.
Monthly reports
The job: Give every client a monthly report and deck from the data we collect for them.
What we built: A scheduled routine that drafts the report and the deck.
Where a person stays in: A person checks every draft before it goes to the client.
Data pipelines that repair themselves
The job: Keep the nightly data loads behind every report running.
What we built: A job that checks every hour for failed pipeline runs. It reads each error and applies only the fixes it’s allowed to make: run it again, reload a limited window of data, raise one resource limit, or wait. Then it checks the tables and writes what it did in the alert thread.
Where a person stays in: Code changes wait for a person.
The operating picture
The job: Know what changed, what’s due, and what we promised each client.
What we built: An assistant that reads our email and our project tool and writes one picture of the business: what changed since yesterday, what’s due, what each person has open, what we owe each client, and what we’re waiting on from them.
Where a person stays in: When it proposes an action, a person signs off first.
Meeting notes
The job: Keep a usable record of every call.
What we built: Calls are recorded, transcribed with each speaker labeled, cleaned up, and turned into notes, decisions and action items with one owner each.
Where a person stays in: A person runs it and checks the result.
One written record
The job: Make every piece of work start from what we already know.
What we built: One written record of how each account, and HandBuiltBrands itself, works. Every agent reads it before it starts. When anyone learns something that isn’t obvious, it goes into the record during the work.
Where a person stays in: People write and correct the record. Agents read it.
What we won’t claim
A number we can’t back up
We don’t publish an hours-saved figure for our own fixes yet. When we do, it goes here with how we measured it.
That the agents run themselves
Every fix here drafts, flags or repairs inside written limits. A person approves anything that reaches a client.
That this is a client result
Everything in this article is our own business. Client results are published separately, blinded.
Want the same map of your business?
We’ll talk through how your work gets done and whether an Assessment is the right first step. See how the AI Assessment works.
Questions about your own situation? Request a call with Matt.