Field notes

What it looks like when the systems get put in order.

Short accounts of real engagements: what we found, what we built, what changed. Client names are withheld where the client prefers; the systems and the numbers are as they were.

Note 01 — Reporting on the spreadsheets you keep

Thirteen spreadsheets stayed. The reading and the chasing didn’t.

Client · Health careProgram · SAP Concur travel & expense implementationPhase · Requirements through testing

A large implementation was being run, faithfully, out of thirteen spreadsheet tabs. Requirements, fit/gap analysis, test scenarios and results, status, and the defect log — all of it, updated live by the team every day. Knowing where the project stood meant reading all thirteen.

What we found

The spreadsheets were the system of record and they worked — people knew them and kept them current. The cost was everything downstream: someone had to read them all, write the status, and remind each tester and defect owner what was due. Every day.

What we built

An end-of-day run. The latest tabs go to an AI-driven process that produces:

  • A live dashboard of scenarios, defects, and status, rebuilt daily.
  • A status summary emailed to every member of the project team.
  • A reminder to each tester listing their scenarios due or past due, with everything they need to run them.
  • An update to anyone holding an open defect, with the current detail.

Every one of them carried the client’s and the project’s branding, so it read as project material at a glance.

What changed

Nobody had to change how they worked — the spreadsheets stayed exactly where they were. The daily reading, writing, and chasing stopped being a person’s job. Status went out every evening without anyone composing it, and testers heard about past-due work from the system, not from a project manager.

Spreadsheets (13 tabs) · AI pipeline · Email · Live HTML dashboard · Client & project branding
Note 01Illustrative
Note 02 — Reporting, after go-live

Stability you could see every morning.

Client · Health careProgram · Same implementation, post-go-live hypercare

Once the system was live, three questions mattered to leadership: is it stable, are tickets being worked quickly, and how do the people using it feel? The evidence lived in three unconnected places.

What we found

The vendor emailed usage and system reports daily. Incidents were tracked in a separate log. Stakeholder sentiment was collected by polling. Each was informative alone; none answered the stability question on its own.

What we built

A daily and weekly stability dashboard that combines all three: the vendor’s emailed reports, the incident log, and the polling results with stakeholder feedback. Interactive, branded to the client and the program, and sent to stakeholders every day.

What changed

One view of how the system was being used, how quickly tickets were closing, and how project leads were perceived — refreshed daily, in everyone’s inbox, for the whole hypercare period.

Vendor email reports · Incident log · Stakeholder polls · Interactive HTML dashboard · Client & program branding
Note 02Illustrative
Note 03 — Software build

Meeting summaries without the enterprise add-on.

Client · Health careNeed · Near-daily meeting summaries, professional grade

The client’s Microsoft Teams plan didn’t include automatic meeting summaries. Summaries were needed almost every day, and they had to read like a professional wrote them — consistent, polished, and actionable.

What we found

Transcripts existed; turning them into something a stakeholder would read was manual work, and the result varied by who did it that day. Procuring a new add-on would take longer than the need could wait.

What we built

A custom tool that lives in a single portable HTML file. Paste the raw transcript, choose the AI you want to use, pick the formatting and what to include, and generate. Then edit in-line, drag sections into the order you want, and send.

  • Works with the AI of the client’s choice
  • Formatting and content options, so every summary matches — in the client’s and project’s branding
  • In-line editing and drag-and-drop reordering
  • No install, no server — one file that runs in the browser
What changed

Daily summaries that look the same every day, produced in minutes, with nothing new to procure or administer.

Microsoft Teams transcripts · Single-file HTML app · AI of choice
Note 03Illustrative
Note 04 — A knowledge base with its own agent

The project’s second brain.

Client · Health careScope · Whole project, every meeting

Over a long implementation, decisions get made in meetings and then get made again, because nobody can find the first one. Action items drift. The second brain was built so that neither happens.

What we found

Meeting notes were being taken, but they were notes: no consistent record of what was decided, by whom, or who owed what by when.

What we built

Every meeting’s notes are turned into a polished summary that carries the key decisions and the action items with assignees. Consistently formatted and branded to the project, all of it lands in a project knowledge base that can be asked questions: what did we decide about X, what’s still open, who owns it.

What changed

Past decisions get revisited from the record rather than from memory. Action items get followed up because the system knows they exist. It is the same pattern we set up for clients of every size — and this is where it was proven.

Meeting notes · Summaries with decisions & actions · Project knowledge base · Client & project branding
Note 04Illustrative
Note 05 — Software build for a consumer group

A fantasy league, run like a production pipeline.

Client · Fantasy pick’em league · a group of individualsType · Website + automations

Not every client is a company. This one was a fantasy pick’em league — a group of individuals who wanted recaps, standings, and advice without anyone doing the work. Picks live on the league’s fantasy site; the system imports them at game lock, ingests results, and publishes weekly recaps and personal coaching for every member, automatically.

What it does

Picks and results pull in after the fantasy site reveals them at game lock. Each week publishes a newspaper-style league recap, private performance trends and a strategy report card for each picker, and league math that explains why an outcome mattered — who took the lead, which confidence miss hurt most.

The discipline underneath
  • State-driven scheduler. Pregame → live → settling → weekly edition, driven by the actual slate, not “Sunday mode.”
  • Per-game lock watcher. A pick is ingested the moment its game locks on the fantasy site and not a second before. Hidden means hidden.
  • Two clocks that never mix. Pre-kickoff advice can only use what was publicly known before lock; recaps may use post-game news. No hindsight leaks into recommendations.
  • One sealed fact package. Once a week is verified, one immutable record feeds every output — recap, coaching, audio — so six products never tell six stories.
  • Fail-safe jobs. A retry produces one clean state, not duplicates. If a source stalls, the site freezes on the last verified fact and says so.
Why it’s here

Consumer software gets the same standard as enterprise software. Logged, reviewable, reversible — the rules we hold every agent to — are easiest to see in a system with no excuse for cutting corners. This one has none.

Fantasy site (picks & results) · Scheduled pipelines · Sealed fact package · Generated recaps & audio
Note 05Illustrative
Next

Which of these looks like your problem?

Tell us what’s not working. We’ll tell you what we’d look at first.

Back to the map