What the firm knows, and how it decides.

Most information a firm needs is already somewhere inside the business.

One ordinary week at Whitlow Ferris Partners, and a proposal written during it. Watch it both ways.

Once it exists, everything that answers a question or drafts a document reads from the same place.

The drawing below is the core from stage three of the walkthrough, with everything that reads from it drawn in. Six of those readers are software and the seventh is a person with her phone. We draw the person the same size as the software, because the core is worth building whichever of the two ends up doing the job.

It remembers three things, and the facts are the easiest of them.

Most attempts at this collect documents and stop there. Documents carry the first of the three well enough. They carry very little of the second, and the second is the part that makes the rest usable by somebody who was not there.

None of this decides anything. A partner still scopes the engagement, sets the fee, and approves the proposal. The memory is what they and their software are working from when they do.

An entry records how a decision gets made, with the answer as only one part of it.

This is the format every approved entry uses. A bare answer goes stale the first time an engagement is slightly different, and a partner will put her name to a rule that has its reasons and its exceptions written down when she would not put it to a bare number.

Exceptions get a field of their own, because most of the real judgment at a firm is in the exceptions and not in the rule. Nobody has to approve a universal rule they would not follow themselves.

This is the entry the proposal drafter cites when it scopes P-2026-041, the entry Ellen reads when she adds the data-readiness week, and the entry Aisha reads in her first month. The record does not change depending on who is reading it.

A small set of approved positions, and a much larger set kept for reference.

Approving something spends the scarcest attention in the firm, so the approved set stays small. Everything else is still worth having, and it is kept exactly as it arrived with a note on where it came from.

Material in the second column never gets quietly rewritten into the first. That would be approving something through the back door, and the value of the approved set is that a named person looked at it.

Where it starts, it is a folder and nothing more.

Nothing is installed and nothing is integrated, so it works with whatever the firm already pays for and it survives a change of mind about vendors. It sits in the SharePoint library the firm already has, and the firm points its own AI tools at it. If Whitlow Ferris stopped working with us the week after it arrived, it would keep working.

An entry that has gone stale still carries a name and a date, and people trust it.

The scoping tab said six weeks at three FTE. It had a name on it and a date, 2021, and proposals were still being scoped from it in 2026, two years after the firm had moved to five weeks. That is the reason the engagement does not end at delivery. What follows is ordinary maintenance, and it is what keeps the memory worth reading a year later.

• New material comes in through one pathA rate card reissue, a debrief, a sponsor preference somebody learned on a call, an engagement that ran over. All of it arrives at one place and not in three partners’ inboxes, so nothing depends on who happened to be copied.
• Conflicts go on a list before anybody settles themWhen the Beltline debrief disagrees with module 4 of PB-100 on how often inventory gets counted, that pair goes on a short list. Nobody edits the playbook underneath Priya.
• The list goes to the person who owns that areaThe debrief line proposes and the owner decides. Priya decided the four Beltline proposals on 23 September. This is the only recurring meeting the whole thing needs, and it stays short because the entries were settled before it started.
• Entries expire on a scheduleEvery approved entry carries a review date. SC-4 comes back each January with the rate card. When a date passes the entry is flagged for a fresh look, and any answer drawn from it says so.
• What it does not cover stays written downThe list of open questions is maintained the same way the entries are. Carve-out pricing sat on it until 18 September. A tool that cannot answer from the memory says so, and people come to rely on that more than on the answers.

Two of the other samples show this happening. At the playbook desk, the carve-out question from the walkthrough is routed to Marcus and comes back as FR-3. In the debrief line, the Beltline lessons are matched to the module, the rule, and the case example they bear on, and Priya decides each one.

The entries are connected, so one question can reach across all of them.

Each entry names the things it concerns: the sponsor, the portfolio company, the engagement, the proposal, the rule, the person who approved it. Those names are connections, and they are what let a question reach material that was never filed together. Here is the chain the memory holds around one sponsor.

Each of these records stands on its own. Because they are connected, they can also answer questions that no single one of them holds. The one below is the kind a managing partner asks, and the memory answers it by walking the chain.

How the walking happens depends on how far the system has been built. While the memory is a folder, the connections are written into the entries by name and a person follows them one step at a time. Once software runs on it, an answer like this one is assembled the same way, only faster, and every step stays cited.

This chain is the same one the other samples walk one step at a time: the Monday brief reads UT-1 when it reports realization by engagement, and the meeting prep reads the same three engagement records three days later. They agree because they read from the same connected records.

Every other sample in this suite is reading from the memory, including the one with no software in it.

These do visibly different jobs and they all work for the same reason. Each one is ordinary software pointed at knowledge that has been written down properly, with a person making the decision at the end.

The seventh reader in the drawing is a managing partner with her phone, preparing for a sponsor call the way she always has and getting the right answer the first time because the answer exists in one place. That workflow has a sample of its own, and nothing in it is drafted by software. If the preparation is worth doing with no software drafting anything, the memory was worth building.