From a supply chain AI demo to a question legal operations will eventually have to answer
Law firm AI planning will not begin with a better chatbot.
It will begin with something less visible: the business logic layer that explains how decisions actually get made.
That thought stayed with me after attending an AI Native Planning workshop during NYC Tech Week, presented by Rodrigo Durán of Pyplan, a Buenos Aires-based planning platform used in industries such as supply chain, operations, and finance. The product was not built for law firms. The demo was not about legal operations. But the planning problem underneath it felt very familiar.
How do you make better decisions when the data is messy, the future is uncertain, and the cost of getting it wrong compounds over time?
Pyplan connects supply chain, operations, and finance into a single decision model. The architecture Rodrigo walked through was simple to understand: an LLM at the base, a business logic layer above that, and a conversational interface on top. Users can ask questions in plain language, run simulations, and test scenarios without writing code.
But the important part was not the conversational interface.
The important part was what sat underneath it.
Rodrigo kept returning to one point: the quality of the data and the business rules underneath determine everything. The AI is only as valuable as what it is sitting on.
This is not yet a law firm story — but it could be
Very few law firm COOs are anywhere near this kind of capability today. Most firms are still working on getting reliable data out of their practice management systems, let alone building AI-driven simulation models on top of them.
And the honest expectation in many firms is that Clio, Aderant, Elite, or whichever platform they run will eventually deliver something like this: packaged, pre-built, and ready to turn on.
That may happen.
But waiting for a vendor to solve a strategy problem is rarely how competitive advantage gets built.
What is worth paying attention to now is the underlying logic. Replace supply chain forecasting with matter resourcing, client revenue planning, pricing assumptions, or capacity management, and the architecture starts to look much closer to what law firms will eventually need.
The question is not whether every firm should build this now.
The better question is what a firm can prepare now, before these capabilities arrive inside familiar platforms.
The part ERPs or Practice Management tools will never give you
Rodrigo made one point that stopped me: you cannot extract the business logic layer from your existing systems.
SAP does not encode how an organization makes decisions. It records transactions.
The same is true of Aderant, Elite, Clio, and other systems law firms rely on. They capture what happened. They do not capture why it happened, what should happen next, or how experienced people weighed the tradeoffs along the way.
The logic that drives good decisions lives somewhere else.
It lives in the rules, exceptions, relationships, constraints, and judgment that experienced people carry in their heads. It shows up in how a matter is staffed, how a client relationship is forecast, how a budget is trusted or discounted, and how a partner knows when an estimate is quietly too optimistic.
That knowledge is real.
But in many firms, it is not modeled anywhere.
And that is the business logic layer.
A digital twin by another name
Rodrigo never used the term digital twin, but that is effectively what Pyplan demonstrated: a working simulation of the business where you can test what happens when a key input changes, a timeline shifts, or a constraint is removed.
His core insight was about compounding.
At the planning horizon, small errors grow. A 5% demand forecast error does not stay a 5% problem. It cascades through inventory, production, margin, and timing.
Law firms face the same dynamic.
An inaccurate matter estimate at intake does not stay contained in the intake meeting. An optimistic staffing assumption does not remain a staffing issue. A poorly modeled pipeline does not simply produce a forecasting miss.
These assumptions surface months later as write-downs, margin pressure, overextended teams, awkward client conversations, and partner frustration — often long after the original cause has disappeared from view.
This is where law firm AI planning becomes interesting.
Not because AI can magically predict the future.
But because better planning requires firms to make their operating assumptions visible enough to test.
The data question will matter
One audience question during the workshop is worth noting for law firm readers: does client data feed back into the LLM and become part of its training?
The answer was no.
The LLM was described as the reasoning interface, not the repository. The data stays in the client environment.
That distinction will matter enormously when this conversation reaches legal. Before any partner, general counsel, or client signs off, firms will need clear answers about where the data sits, how it is used, and what the model can and cannot retain.
But even if those questions are answered well, the harder work remains.
A secure interface does not create a useful planning model.
The firm still has to know what it wants the model to reason over.
What law firms can do today
Most law firms are not ready to buy a platform like Pyplan. And that is not the point.
The useful work starts earlier.
-
Get honest about your data quality
Matter data, billing narratives, staffing records, client intake, budgets, write-downs, and realization data all become more important when a firm wants to reason across them.
Are they clean enough?
Are they consistent enough?
Are they structured in a way that reflects how the firm actually works?
This is not glamorous work. But it is the foundation for anything more advanced.
The firms that move fastest when AI-native planning becomes more available in legal will likely be the ones that have already done the housekeeping.
-
Write down how decisions actually get made
Not the policy.
The practice.
How does your firm actually staff a matter? How does it forecast a client relationship? How does it decide whether to pursue a new practice area? How does it know when a budget is credible? How does it adjust when the stated plan and lived reality start to diverge?
That tacit knowledge is the business logic layer.
No vendor will be able to hand it to you pre-built, because no vendor knows how your firm makes judgment calls under pressure.
Capturing that logic now puts a firm ahead of the conversation.
-
Pick one forecasting problem and name it
Not AI in general.
One specific place where better foresight would change a decision the firm makes regularly. That could be:
- Matter budget accuracy.
- Capacity utilization by practice group.
- Client churn signals.
- Pipeline reliability.
- Pricing assumptions.
- Lateral integration.
Naming the problem matters. It narrows the conversation from possibility to usefulness. It also helps the firm know what to ask for when planning tools become more sophisticated.
The real preparation is not technical
Pyplan counts companies such as Puma, Pirelli, Nestlé, and Coca-Cola among its clients. Law firms are not yet the obvious market.
But the underlying problem is not industry-specific.
Decisions are often made with incomplete models of the business, at the moment when the cost of error is highest.
That is true in supply chain planning.
It is also true in law firm management.
The firms paying attention to where planning is going, even in industries that look nothing like legal, will be less surprised when these capabilities arrive closer to home.
Not because they chased the newest tool.
Because they understood what the tool would eventually need from them.
The business logic layer has to come first.