Embedded AI vs bolt-on AI in Supply Chain Planning | Ep. 27

"Why don't we just pour all our data into Claude and use that as our planning tool?"
If you work in supply chain planning, you have probably heard that question in your own company. It sounds reasonable.

The general AI assistants are impressive, everyone is already using them, and they will happily answer any question you throw at them. This episode takes that question seriously and answers it properly.

Søren Hammer Pedersen hosts this session of the S&OP MasterClass and brings in Benjamin Obling, CPO of Perito IBP at Roima Intelligence.

Benjamin spends his working life embedding AI agents into integrated business planning software, which makes him exactly the right person to explain where the general assistants end and dedicated planning AI begins.

The surprising starting point is that the two are built on the same raw material. The LLMs behind an embedded planning agent and behind Claude or Copilot can be identical.

What separates them is instruction, context, and governance. An embedded agent knows which page you are on, which step of the process you are in, which of your six different forecast versions you actually mean, and which data your role allows you to see. A general assistant knows none of that, and it will still give you a confident answer.

By the end of the conversation you will understand where each kind of AI belongs in your planning setup, why verification is the difference between a useful alert and a dangerous one, and why the real test of any AI analysis is a simple question. Did we make a better decision?

In this episode

• Why "just pour the data into Claude" keeps coming up, and what that question misses
• What actually separates embedded AI from bolt-on AI when the underlying LLMs are the same
• How role-based governance keeps planners inside the data they are allowed to see
• When a general AI tool is still the right choice for a planner
• What changes for planners and executives in the monthly process, and what stays the same

Chapters
01:10 Welcome and the question every planner hears
02:04 Bolt-on and embedded AI defined
05:05 Instructions, context and the six forecasts problem
09:17 Security, roles and data governance
11:10 The cost of iteration and pre-prepared analysis
14:18 Sharing best practice through one data model
16:38 When to use which tool
19:26 Same monthly process, different division of labour
21:47 The shadow AI pitfall and proving the answer
25:01 How planners actually receive embedded AI
27:32 Less is more and the decision test

Contact and follow
Questions, topic ideas, or guest suggestions: podcast@roimaint.com
Find more episodes at: https://www.roimaint.com/en/catalog/node/insights-webinars-events-and-podcasts

If you want to know more about what we do in Perito IBP, we are here to help.

Production
This podcast is brought to you by Roima.
This podcast is produced by Montanus.