Scattered information
Start with
what you already know.
Select the sources useful for a business question, in your tools and with the right access.
Company Brain · AI consulting and engineering
What your business knows. Finally connected.
A knowledge base that AI queries for your team, with the source and date of each answer.
Company Brain: the operational memory of your company, finally shared.
To find an answer, understand its context
and return to what justifies it.
An email
“Validated, for this file. ”
A decision to find again A report
The reason for the decision.
A context to keep A folder
Its history. Its exceptions.
The context found
A memory.
Verifiable links.
01 / From fragment to action
More than a tidy Drive.
A memory that responds.
A fact, a hypothesis, a decision: they do not have the same meaning. Company Brain connects information to what can be used.
Memory architecture
DocumentsThe written rule
ConversationsThe nuance exchanged
FilesThe work already done
MemoryFacts · Decisions · Exceptions
Who decided?
In what context?
What has changed?
A sourced answer.
Better informed action.
Better informed action.
Their links become usable.
Start with
Structured memory
Preserve the meaning.
Not just the text.
Distinguish between facts, hypotheses and decisions. Maintain dates, versions and exceptions.
Shared context
Find the why,
with the answer.
Open the sources of an answer, understand its scope and see what is missing or contradicts itself.
Better informed decisions & actions
Move forward without
starting from scratch.
Prepare a file, apply a rule or consult the right person in charge. The context informs the work, without replacing business judgment.
The value is not in accumulating more information.
It’s finding the one that changes the decision.
02 / The raw material
Your tools already
have a history.
We start from the sources necessary for first use, not from your entire information system. Access, compatibility and update rate are verified during scoping.
The source indexChoose a source to see what it provides
Source sheet / 01Illustration
REFERENCEVersion → scope
Reference procedure
The rule is not enough.
Its version matters.
“This procedure applies to new requests. Files already initiated must be examined separately. ”
What becomes usefulFind the applicable rule, without confusing an old version with the procedure in force.
Source sheet / 02Illustration
PURPOSEValidating an exception
judgment in an email
One sentence.
A precise range.
“Exception validated for this folder only. To be confirmed if the request arises again. ”
What becomes usefulFind judgment without transforming a one-off exception into a general rule.
Source sheet / 03Illustration
FILEExchange → decision → continuation
History of a file
A relationship.
Not an isolated plug.
An identified need, an adjusted proposal, a pending item. These steps take on meaning when they are linked to exchanges.
What becomes usefulResume a file with its commitments and open points, rather than just its last status.
Source sheet / 04Illustration
REPORTDocumented decision
Meeting minutes
The choice is written.
The why too.
“The hypothesis is adjusted in view of these elements. The person responsible for the file will check their relevance before reuse. ”
What becomes usefulUnderstand a previous decision and know who to contact when it needs to be reconsidered.
Source sheet / 05Illustration
TRANSMISSIONTo be clarified → to be validated
Knowledge to formalize
What's on people's minds
do not invent yourself.
A rule known to only one person must first be explained, documented and then validated by the right people.
What becomes usefulTransform a recurring response from the manager into validated knowledge accessible to the team.
03 / Memory over time
The present is evolving.
The why remains readable.
A decision may be correct on one date and then no longer apply. Updating memory should not erase its history.
Customer exception logIllustrative example Explore the three dates
Documented decision
What memory must restore on this date
“This exception has been validated for this file. It is not a general rule. ”
- Source
- Report from June 12
- Context
- A specific customer request
- Limit
- Confirm before any reuse
Contradiction reported
What memory must restore on this date
“The general procedure has changed. The old exception needs to be revisited. ”
- Sources
- Report of June 12 + procedure of June 18
- Conflict
- The recent rule and the exception are not enough to decide
- Expected sequel
- A decision of the manager, not a silent deduction
Clarification validated
What memory must restore on this date
“The exception remains limited to the initial file. The new rule applies to other requests. ”
- Source
- Clarification from the manager of June 24
- History
- Initial decision and contradiction retained
- Learning
- The exception must no longer be taken outside its scope
04 / Work, concretely
Start from a question
your team actually asks.
Two examples of initial uses, each built around your needs.
01 / M&A teams
A useful precedent.
With the reasoning behind it.
Why was an assumption adjusted in a comparable file? The response is divided between a data book, emails and reports.
The first corpus
Closed files Versions Decisions
An authorized set of records, with their versions and associated decisions.
Our experience in operational monitoring of M&A files feeds into this scoping process.
Scope an M&A knowledge baseReading note / M&AExample answer
The expected response
Find the fit.
And what explains it.
Data book · file version
The relevant passage, not an isolated extract.
Return to the hypothesis in its document and its version. Situate the passage before interpreting it.
Email or report · associated decision
The reason for the adjustment.
Find out what was arbitrated, on what date and in what context. A hypothesis does not become a fact.
Reference file · application limits
Which is not automatically transposed.
The conditions of the original file and the limits of comparison must accompany the response.
To be validated by the team
Relevance of the comparable, confidentiality and application of the hypothesis to the current file.
How to measure success
On known questions: the source found, the accuracy of the explanation and the research time.
02 / Managers & teams of VSEs / SMEs
We've already decided.
Let's find out where.
What rule should be applied to this customer request? The answer exists, but the team still has to interrupt the person who remembers it.
The first corpus
Scope a recurring question
Procedures Validated answers
A repetitive customer flow, in the tools that the team already uses.
Team marker / VSE · SMEExample answer
The expected response
The rule. Its exception.
The right handoff.
Procedure · applicable reference
A landmark that the team can find.
The rule in force and its scope of application, with the source which allows them to be verified.
Answer validated · special case
An exception does not become the rule.
Find out what was accepted and for what case, without silently extending this decision to all requests.
Responsible · non-framework case
Know when to pass the baton.
Identify the person to consult when the request goes beyond the documented scope, rather than deducing a new rule.
The handoff remains human
A new exception or commitment to the customer falls to the manager.
How to measure success
Recognize recurring questions, test the answers, then observe their use by the team.
05 / Conditions of trust
Memory informs.
You keep control.
A useful memory is not a stand-alone authority. Rights, traceability and limits are defined from the scoping stage.
A generated response may be incorrect. Human sources, testing and validation reduce risk; they do not replace professional judgment.
06 / The BM method
Start with the right scope.
Then build the rest.
04
weeksA primary objective,
over an agreed perimeter.
over an agreed perimeter.
The starting point
A question. A corpus. A team.
Frame what matters.
Choose the recurring question and reference sources. Define access and collection rates.
Selected question Bounded corpusTest the answer.
Compare responses, sources, refusals and contradictions to a human reference. Measure initial research.
Known cases Trust criteriaGetting down to real work.
Observe if the team finds a reliable answer and actually uses it. Correct before enlarging.
Observed use Next to decide
Our MVP aims for a first value in four weeks, on an agreed scope. cost estimate depends on the stream, corpus and access requirements.
Build what is needed. No more.
A well-maintained wiki can sometimes be enough.
We check what your tools already allow. Wiki, documentary search and assistant can complement each other; none of these terms are guarantees of quality.
Wiki, documentary research, assistant: compare roles
| Approach | What it brings | What to organize |
|---|---|---|
| Wiki or knowledge base | Procedures and reference pages that the team can read. | Those responsible, updating and research. |
| Documentary research / RAG | Retrieving passages to support an answer. | Corpus quality, rights, versions and verification. |
| Conversational assistant | An interface to ask a question or prepare work. | Accessible sources and the limits of its actions. |
| Custom Company Brain | A memory of facts, decisions and exceptions related to the company's work. | Precise use, responsible sources and tested trust criteria. |
Before you start
Company Brain:
frequently asked questions.
A shared memory project starts
with a clear scope.
Should we change our tools?
The project starts from your tools and their available access. A reading connection may be sufficient for first use. Compatibility and data quality are checked during scoping.
Can we trust all the answers?
A generated response may be incorrect. Citations, known case testing, missing detection, and human validation reduce risk; they do not dispense with professional judgment.
What if our procedures are not written?
You must first identify the reference people and decisions. The system cannot invent rules that the company has not defined. The collection and validation of this knowledge can constitute the first task.
How to measure usefulness?
Choose representative questions, measure initial search time and verify the accuracy of answers. In use, track repeated responses, errors, corrections and necessary escalations. The criteria are defined with the team before the project.
Who can build a Company Brain for an SME or M&A team?
BM Automation designs it to measure, based on your tools and their access, for a limited initial use. We first check what your tools already allow: if a well-maintained wiki or documentary research is enough, there is nothing to build.
Let's open your first file
Your first
question.
Our starting point.
What information does your team have?
does it still have to reconstitute?
The next file
starts with you.
starts with you.
An exchange on your reality.
Come with a recurring question and the relevant tools. We will define the first use, its sources and the way of measuring the result.
Custom made. With your tools. On your scale.