What we scope
Exposed scope, rights by profile, query traceability, data excluded from access.
Home — AI
The agent builds fast. You still have to know what to build.
Pigment's AI agents write models, analyse variances and simulate scenarios. They move the effort of a planning project, they do not remove it. Here is what that changes in practice, and what it does not.
Collection plan — assistant
Ask the plan a question
The Pigment agents
Three roles, inside your model.
The agents work inside the platform, on your governed data and your business logic. Every action stays traceable, and the decision stays yours.
Analyst
Monitors indicators, spots anomalies, explains variances and writes the summary. Your teams stop hunting for figures and focus on the decision.
Modeller
Builds and evolves models from a plain-language description, respecting your structure and documenting what it produces.
Planner
Simulates scenarios in real time and proposes trade-offs grounded in your data and your business context.
What this changes for your project
The effort moves upstream.
When building a model takes minutes instead of days, the limiting factor is no longer build capacity. It is the quality of the question asked and the cleanliness of the underlying data.
That is precisely the work we do: turning a vague business need into explicit rules, and making the data trustworthy. An agent amplifies that work when it has been done, and amplifies the error when it has not.
- Framing matters more than before: a badly framed model is generated as fast as a good one
- Data governance becomes the real prerequisite, not an end-of-project step
- Testing changes in nature: you check reasoning, not only formulas
- Iterations with the business happen during the session, not between two workshops
- Adoption accelerates, because a user can query the plan in plain language
- Documentation stops being the poor relation: the agent produces its own
Connecting to assistants
Query the plan from Claude, ChatGPT or Copilot.
Pigment's MCP server opens governed access to the models from external assistants. The assistant reads the data with the rights of the user asking the question, reasons and proposes. Pigment remains the calculation reference: the rules do not move.
The business user
asks, arbitrates, approves
- Asks a question in plain language
- Triggers a recurring task
- Receives a scheduled alert
The assistant
understands, reasons, explains, proposes
- Your business rules and your vocabulary
- Analytical models built on history
- Reasoned answers, notes and proposals
reads the indicators
MCP
submits the proposals
Pigment
calculates, stores, secures
- Calculation rules fixed and unchanged
- Reference data: budget, forecasts, plans
- Rights by profile, approval and change log
Out of scope
Data warehouse, external sources and open data remain accessible separately, for the few uses that need them.
In practice, in four steps
1
Ask
“Why is margin falling on this family?”, or a review scheduled every Monday.
2
Read
The assistant queries the indicators the user is entitled to, and nothing else.
3
Explain or propose
A reasoned answer, or a proposed value with the factors behind it.
4
Approve
The user approves in Pigment. The calculation runs again under the established rules.
What we test
Answer reliability on your real cases, before opening access to the teams.
Where we stand
What AI will not do for you.
Decide a business rule
An agent proposes. Whether replenishment favours service level or inventory value remains a leadership call.
Reconcile definitions
When finance and merchandising do not count margin the same way, no model settles it for them.
Clean your master data
An agent on wrong data produces a wrong answer, faster and with more confidence.
Carry the change
Adopting a new process is won in workshops, with the teams, not in an interface.
Stand behind a forecast
Responsibility for a figure committed in front of a committee stays human, whatever machine produced it.
Replace the product owner
Turning a need into an executable request is precisely the work agents make more valuable.