The Enterprise Discovery Vault
Used in complex business transformation initiatives.
The Enterprise Discovery Vault is where thousands of details about how a business runs are stored for use in improving enterprise business processes and the systems that support them, like ERP.
These details are kept in their natural form, in one place the AI agent can search in full, so the implementation team can find and easily leverage the information at the moment they design each process and its metrics. That moment is where most such projects fall short: the influencers on a business process are spread across hundreds of spreadsheets, forms, applications, emails and people's memories, and the design team rarely sees them in any coherent way that enables absorption. The vault stays confidential and under the client's control throughout.
The problem: the design team cannot easily see what shapes each process
In the business process design phase of a major business systems project, such as an ERP implementation, the company team decides how each process and its metrics will run in the new system. To decide well, they need to see legitimate influencers about business process intent, while systematically avoiding legacy processes that conflict with modern systems. That is rarely done well.
The influencers are everywhere and in no particular order:
- Business process support tools. Spreadsheets, Word and paper forms, checklists, email templates, small databases, whiteboards and scheduling boards, and commercial and custom point applications. A reasonably thorough listing for a small company totals roughly 500 separate tools; a larger company has far more.
- Data sets that will have to be cleaned up and moved to the new systems, or archived or deleted.
- Documents and correspondence: policies, procedures, reports, emails, meeting notes and interview transcripts, all of which can hold important influencers.
- Tribal knowledge: how things are actually done, not formally documented, known only to the people who do them.
These hold hundreds, often thousands, of details about how the company actually operates that interviews and other investigation may pick up but do not organize in a useful way.
From scattered sources to coherent information
The Enterprise Discovery Vault is EAI's answer to the problem just described. It gathers the scattered influencers into one place, keeps each in its natural form where AI can search it, and puts them in front of the design team as each process is designed.
Everything on the left side of the chart at the top of this page exists in every company, supporting enterprise business processes and describing objectives or ideas for improvements. However, when the time comes to pursue such improvements, this vast background information is rarely in an organized and usable place. The vault gives each item one record in plain text, searchable by what it says and tagged to business processes and other topics it bears on, so what comes out on the right can be found by any question and is also organized by process, ready to use.
The vault's foundation: Obsidian
The vault is a set of folders of plain text files holding all client discovery, with links to the client's actual artifacts. The files are managed in an elegant application called Obsidian (obsidian.md), a mature, modern platform built for privacy and autonomy.
Obsidian is developed by its own dedicated team and is widely used, and EAI chose it as the vault's backbone because it fits this purpose closely:
- Plain text the client keeps. The files are ordinary text (markdown) on the client's own storage, readable in any text editor, even without Obsidian.
- Private by design. Obsidian works offline and stores nothing on anyone else's server unless the client chooses to.
- Extended by plugins. A large library of add-ons supplies the views, task boards and pickers the vault uses.
- EAI adds the method. EAI's contribution is what goes into the vault and how: the vault kit, the agent's instructions, and the method for finding, preparing and using the information.
A client vault is built by an EAI script from a standard vault kit, so every client's vault has the same general structure, the same fields and the same views. The structure can then be tuned to a client's particular needs.
Who controls the vault and owns its content
The vault holds the client's own material, and that content belongs to the client, always. EAI provides the vault's structure — its templates, views, guides and AI agent instructions — under a license for the client company's use, for as long as it likes, to improve its business processes in the areas set up in the project, at no cost beyond the Intellectual Property fees in EAI's proposal. EAI's team runs the vault with the client during the work and keeps it in order, and the client can take direct access or full control whenever it chooses. The Obsidian vault can be hosted anywhere the client chooses, including locally. There is no remote third party demanding involvement, no ongoing subscription of consequence, and no new intrusive commercial terms to wade through. The vault is ready to hand to the client at any point along the way. It is never handed to an implementer or any other outside party, and whenever EAI departs, at any point and for any reason, it is turned over with its trained AI agent and with training in its use.
The content stays confidential. The AI agent reads the files where they sit and sends them only to the AI model it runs on (Anthropic's Claude, under a business account where the content is not used to train AI models); after handover it runs under the client's own account.
This is an example of the autonomy and control that will increasingly be the standard for a company's valuable information and critical processes.
EAI's part: finding the information, preparing it, and using it
The EAI-designed Enterprise Discovery Vault holds the information. Finding it is the hard part. The influencers are scattered across drives, desks, inboxes and people's memories; many are undocumented, and the people who rely on them often do not think to mention them. A vault filled only with what is easy to collect may look complete, and then misleads the design team at exactly the moment it is trusted. EAI's part is on either side of the vault:
- Locating the influencers: scanning the company's drives for tools and data, interviewing the people in each process area, and drawing on EAI's deep experience of where influencers are usually found.
- Preparing them for the vault: gathering them in one pass, naming and dating them, and adding details that make each piece of information easier to find and understand, with AI doing much of this work. Filing is light: an item goes in largely as it is, and from then on the AI agent can read and search it along with everything else.
- Using them on the other side: in a project, for process and metrics design sessions, tool decisions, data cleanup and training.
The vault is used most heavily in our Implementation Control service.
Business process design going forward
The following summarizes the three major results of having comprehensive, organized and searchable company details in the Enterprise Discovery Vault. These results are expected to greatly improve business process and metrics design decisions, help control project scope, and largely remove the factors that cause drift and dilution in complex business systems projects.
Result 1: Thousands of influencers in view when each process is designed
The largest result is cumulative. Each item in the vault, and each detail within an item, contains insight about how the company works: a column in a spreadsheet, a step on a checklist, a report someone runs every Friday, a rule a supervisor applies from memory. On its own, each one is easy to miss in the design of future processes. Brought forward systematically — every item filed, tagged to the processes it bears on, and put in front of the team when each of them is designed — thousands of these details add up to a complete picture of discrete process intent, placed in front of the people who are redesigning it.
When the project team works to design a process, the AI agent, set up specifically for the vault's structure and the client's content, scans the entire vault for salient influencers to review. The items are gathered into an information pack on the subject, grouped by what they are about, and the pack is kept in the vault as a record of what the team had in front of it.
The groundwork is done as each item is filed, not saved for the end. AI proposes how the item is filed: the business processes it bears on, the questions, decisions, risks and build items in it, and the actions it leads to. A person on EAI's team reviews the proposal before anything is applied, and for meeting results the client lead reviews it too. The requirements list carries the status of each requirement, so what is not yet covered shows there. AI's own implementation considerations are kept apart, as labeled AI suggestions in the information packs.
Anyone on the team can also ask in plain language. For example: “Today we are discussing excess inventory of items with a shelf life: what is causing it, and how should the go-forward process work in the new system?” The agent searches the vault for related items, including ones that never use those words, and returns a report that cites the record behind every point, keeps what the vault shows apart from its own suggestions, and lists what the vault does not yet hold.
Vault-wide dashboards show the same information across all processes: tools still waiting for a decision, evidence still being collected, data sets by cleanup status, and the applications in use.
For business process and metrics design, this is a transformational change. The design of each process starts from comprehensive evidence of what the company actually does and intends to do, not from memory and a blank page.
Result 2: Every existing tool managed into the future
Existing tools are a large share of the influencers in Result 1. After the tools are reviewed for influence, each tool is managed systematically into the new business process support structure. Every tool record carries the decision for that tool:
- Retire — a new system takes over what the tool did, and the tool is removed after go-live.
- Modify — a new system takes over part of what the tool did; the tool is kept, with the replaced parts taken out after go-live.
- Remain as is — the tool still has a job the new systems will not do, and it stays.
The dashboard lists every tool still waiting for a decision and every retired or modified tool not yet removed or adjusted after go-live, so retiring and modifying tools becomes a managed list rather than a hope. Every tool retired means that process or metric now runs inside the new systems or another go-forward method, instead of in an isolated spreadsheet or form that conflicts with the new systems and is often known only to its one or two users.
Result 3: Operational data sets tracked from their current state to their final home
The company's operational data sets are listed in the vault the same way, one record each, tagged to the processes they serve. Each record carries three decisions that are managed until the data has reached its home:
- Status — whether the data set is in the scope of the project, partly in scope, out of scope, or still uncertain.
- Improvement — the cleanup the data needs — how much and how far back, tracked as not started, in progress or done. Cleanup can start as soon as a process has been reviewed, and run for months up to the move.
- Destination — a new system, an archive, or deletion. Data that is not moved forward is archived or deleted deliberately, with the decision recorded.
The dashboard shows every data set by cleanup status and destination, so the state of the data is known at every point up to the move.
Managing the vault and its AI agent
The vault is not a project management system. The project keeps its own plan, schedule and project management tools. The vault manages what no project management system can: the thousands of details about how the business works, and the decision on each, kept in front of the people who need them at the right moment. The actions it raises come only from that content (a tool to decide on, a sample to collect, a call to make), not from the project plan's tasks, milestones or resources.
Managing the vault's details day to day
Working the vault's information spawns hundreds of open items: a tool to decide on, a sample still to collect, a data set to clean, a question for the client, a follow-up from an interview. Most are tied to one record; some are broader. The vault keeps them in front of the team in three ways.
These are content actions, not project management tasks. Each one comes from an item in the vault and closes when that item is settled. The project plan, schedule and resources stay in the project's own management tools.
1. Work the records track themselves. Much of the open work needs no separate action at all, because the records' own fields already say what is incomplete. The views read those fields and list the work:
- tools still waiting for a keep, modify or retire decision, and retired tools not yet removed;
- tool samples and other evidence still being collected;
- data sets by cleanup status and destination.
No one has to create an action to keep these visible. When a record is updated, it leaves the list of open work and stays in the vault as part of the permanent record.
2. Actions. Work the fields cannot express — a call to make, a document to request — becomes an action item. Each action identifies a responsible person, when it is due, its priority and status, and links to the business process and the record it came from, so the reason for the action is one click away.
3. Questions. A question for the client is not made an action. It is flagged where it arises, in the record it belongs to, and the agent gathers the open questions into one list by business process, so the team can see where a meeting with the subject matter expert for that area would help, and how long it needs. The answer is written into the record, and the question is marked answered.
Who runs it. During the project, especially the business process design phase, someone has to make the work happen. EAI runs the vault's management of its content while it is engaged: the project manager works the vault's views daily, alongside the project plan in the project's own tools, and a review of them is held at least weekly to assign new actions and chase what has stalled. The vault agent, described in the next section, prepares both.
Why AI changes the effort of managing this information
EAI has built this kind of vault repository before, as a database application. Every detail had to be entered as structured data before anything could be tracked or reported, and that entry effort was ultimately too much for most clients to sustain.
The vault reverses that. Each record carries only a few attributes, and the rest is the plain text people write anyway: meeting notes, interview summaries, emails, observations. AI reads both, and finds an item by what it says, not only by how it is tagged. Set up properly, it is expected to keep the salient work and the open issues in front of the team with little added entry effort. The vault agent is Claude, an AI model from Anthropic, working on the vault's files under standing instructions that travel with the vault. These are its standing outputs:
- A daily note and a weekly status brief: each involved person's work for the day, and for each business process and each person what moved, what is open and what is overdue. This includes team members from the client, the implementer, EAI and others involved.
- Open items found in the text. AI reads new records for commitments and follow-ups, and drafts actions for a person to accept, change or discard.
- Gathered lists: with each weekly brief, the open questions for each subject matter expert, a log of the decisions made, and a list of the build items.
- Drift flagged early: actions overdue or waiting too long, a process with no movement, work added faster than it is closed, deferrals piling up, decisions past due, and open questions piling up in one process area. The client sets the limit for each at a briefing.
- Gaps and inconsistencies flagged: rule errors found by a daily vault check, and suspected errors found by reading — the same item filed twice, a record about a process it is not tagged to, a question answered but still open.
- Depth of evidence: for each business process, a rating of how deep its evidence runs — light, adequate or deep — so a thin area is worked harder rather than trusted. Before the implementer immersion, a snapshot of each process: what the vault holds, what is still open, where it is light, and who on the client side collects there.
- Answers on request: questions about the client's content, and about how to run the vault, answered from the vault itself. The client runs the vault by opening it in Claude and asking, without reading EAI procedures.
AI proposes; people decide. Every action it drafts is accepted or rejected by a person, and every change to a decision field is made by a person. EAI estimates that this reduces the administrative effort of running an enterprise knowledge repository to a small fraction of what a structured database required, and makes it practical for such work to be done well. In most complex business transformation projects it has been done poorly, causing cost overruns and calendar delays while reaching results well short of what was possible.
Bounded and confidential. The agent writes only its own notes and draft actions. It never changes a record, a decision or a status. It has no web access. Where outside guidance would help, such as industry practice for a process, it drafts search phrases that do not name the client, and a person runs the search outside the vault and files what is useful. These limits are written into its instructions and set again in the application's permission settings. It runs under a Claude business account, where the client's content is not used to train AI models, and after handover under the client's own account. Each version is tested against a vault seeded with known problems before it reaches a client.
After go-live: the vault carries on
The vault covers the whole enterprise, while each go-live has a controlled scope. When a system goes live, many items in the vault are still pending: processes and requirements outside that scope, items meant for other applications, instructions for improving tools that are being kept, data sets still being cleaned, and more. The vault carries on after go-live in much the same capacity as before, holding those items and keeping them in front of the team until each one is settled.
The vault and where business systems are heading
AI options, and the large benefits they offer buyers, are estimated to move business logic and data out of today's large commercial applications over the coming years, in steps, with ERP as the core example. How a company handles that move decides whether it reaches the destination in one piece.
The vault is built for that. It holds what the company does and intends in its own words, in plain text it owns, outside any seller's system. The same record serves an upgrade, a new ERP, or a later move of business logic to new AI business process frameworks. If a seller fails or shows signs of failing, it is the basis for an accelerated plan. See where business systems are heading.
Through all of this, the vault remains confidential and under the client's control. It works beside the project's management tools, not in place of them: they run the project, and the vault manages the vast details the project depends on.
Contact us to discuss the Enterprise Discovery Vault for your company.