Spark Labsby Sam

Work I have done

Four projects from more than a year of building and running operations for a US service business with several programs and a remote team. The details are kept general because client work is confidential. The work itself is all real.

Switching the platform a business bills through

7platforms scored against a weighted checklist before choosing one

The situation

The company had outgrown the software it used for scheduling, billing clients, and paying staff. Changing it meant touching every invoice and every pay run, so a bad choice or a messy move would have been felt by everyone.

What I did

I turned the owner’s requirements into a weighted checklist and scored seven platforms against it. The two finalists got hands-on trials, including a stress test of the hardest billing scenario the business deals with. I also modelled payment processing costs across the options using real collection data, which showed that how clients pay mattered far more than which platform was chosen.

Once the decision was made, I planned what to export from the old system, coordinated the data move with the vendor’s migration partner, and wrote guides for staff and for clients. The old system’s shutdown was planned around what records had to be kept.

RequirementsScorecardTrialsMigrationLive

What changed

Scheduling, billing, and staff pay now run on the new platform, with written guides for everyone who uses it. The decision is documented too, so the reasoning is on record if the question ever comes up again.

An operating system built from nothing

80+automations running across more than 20 connected tools

The situation

Leads were in one place, clients in another, tasks in people’s heads, and documents wherever they had last been saved. What held it together was people copying information between tools and remembering the steps.

What I did

I built one workspace around a single source of truth: a contacts database where every person exists once, and everything else links to it instead of duplicating it. Work is organised in a simple hierarchy, and each task links to the documents, people, and meetings it touches.

AreasProjectsTasks

On top of that sits the automation layer. Enquiries flow in from forms, bookings and payments post to team chat, weekly and monthly numbers log themselves into a history nobody can overwrite, and a failed automation raises an alert instead of failing quietly. Because the team includes seasonal staff, access is set by role, so each person sees what they need and nothing else.

What changed

The day-to-day runs through one documented system. New people get up to speed from written guides, and leadership sees leads, meetings, and deals on live dashboards without asking anyone to pull a report.

AI doing real back-office work

19 to 1automation runs per meeting, after rebuilding the meeting notes pipeline

The situation

A lot of recurring admin had many steps and many small decisions: processing orders, handling applications, importing registrations. Done by hand it was slow and easy to get slightly wrong.

What I did

I wrote each process as a precise procedure with its decision rules spelled out: when to update a record and when to create one, which details must match, and when to stop and ask a person instead of guessing. AI then carries out the procedure directly in the live tools, and anything it flags comes to a human.

Meetings got the same treatment. Each recorded meeting becomes a filed note with a summary, the transcript, and the action items. After a person reviews them, the actions become assigned tasks.

RecordedSummarisedReviewedTasks assigned

What changed

The procedures run the same way every time, whoever starts them. Rebuilding the meeting pipeline also cut what it costs to run: a long meeting that used to take up to 19 automation runs now takes one.

Weekly reporting without the scramble

1screen for the whole weekly review, in the order the meeting runs

The situation

The weekly status meeting covered tasks, sessions, client activity, money, and the sales pipeline. Preparing for it meant clicking through several systems and assembling the picture by hand.

What I did

I worked out with leadership which numbers the meeting needed, then built a dashboard laid out in the order the meeting runs. It opens with a written briefing and a list of what needs action, then moves through each area, with week-on-week history so trends are visible.

The data collection is a documented weekly procedure. When the first approach to pulling the data turned out slow and unreliable, I replaced it with a faster, exact method.

What changed

The meeting starts with the numbers on screen and the problems already flagged, so the time goes on decisions instead of on finding out what happened.

Tell me what is slowing you down.

A 15-minute call is enough to work out whether I can help and what it would involve. If I am not the right person, I will say so.