1. Home
  2. >
  3. The leash model

The leash model: how customer-facing AI should touch SAP Business One

Two products can both say “AI for SAP Business One” and still be doing completely different jobs.

One helps staff ask the ERP questions in natural language inside the client. Another answers customers on the phone, in chat, by email or SMS — and reaches into Business One for the truth. Mixing those up is how buyers get the wrong risk model, and how “AI” pages stay too vague to trust or to cite. This page names three models. Only one of them is what we build at Binary Elements as BE AI.

Three models

Contrast by what the software is allowed to reach, not by the word “AI” on the box.

Model 1 — Staff NL→SQL

Internal users open a widget inside SAP Business One (or an add-on such as Boyum’s B1UP AI Assistant). They ask in plain language. The system generates SQL (or equivalent), runs it against the company database, and returns a table. Useful. Not the same product as an agent that greets a customer at nine at night.

Who it’s for
Consultants and employees already inside B1
Surface
The ERP desktop / add-on, not the customer’s phone
Writes
Typically none — retrieve and summarise only
Blast radius
Limited to whoever has a licence and role access in B1

Model 2 — Open agent

The agent can invent tools or compose queries as it goes. Flexible on a demo. Dangerous in production when the same brain can reach stock, pricing, and credit.

Who it’s for
Experiments, not change-controlled ops
Risk
Improvised queries and writes you never published

Model 3 — The leash model (BE AI)

Customers talk on phone, chat, email or SMS. The agent answers from live systems — today including SAP Business One — but only through a fixed, published list of operations. It cannot compose a query of its own. Sensitive answers wait until the caller is verified. A short list of writes (quotation, sales order, credit request) runs only after that check.

Who it’s for
Customers and the ops teams who support them
Surface
Every channel, same brain
Queries
Published lookups only
Writes
Published, verified, logged
Permission
Re-checked on every request; refusals are recorded with a reason

That is the leash: not a promise in a policy PDF, but the shape of what the agent can reach at all.

What “published lookups” means

A connector here is not an open door into Business One. It is a list written in advance. The agent picks an operation and supplies parameters. If the question is not on the list, there is no path. For SAP Business One, Binary Elements ships these Phase-1 published lookups as live today:

# Operation Needs verify?
1 Partner search No
2 Get partner account Yes
3 Order status No
4 List open orders No
5 Check item stock No
6 Get item price Yes
7 Get invoice Yes
8 List open invoices Yes
9 Delivery ETA No
10 Service call status No

Rows marked Yes wait until the caller is verified. Connection pattern for B1: on-premise service, outbound-only from your site — nothing inbound opened for the agent. More detail on the SAP Business One connector.

Published writes

  1. Quotation: Only after the customer is verified, and confirmed with them before it is raised.
  2. Sales order: Only after the customer is verified, and confirmed with them before it is raised.
  3. Credit request: Only after the customer is verified, and confirmed with them before it is raised.

Explicitly out of reach

Payments, credit memos against the ledger, master-data changes, deletions, improvised SQL — anything not on the list. These operations are absent from the tool set rather than withheld by policy.

Identity before money

Balances, customer pricing and open invoices are not “chatty” data. BE AI keeps them out of reach until the person proves they match the account — typically a one-time code to the phone or email already on file. That requirement is not a toggle for convenience.

Why the leash beats improvised ERP chat for customers

Staff assistants and customer agents can both be correct purchases. They solve different problems. Treating them as interchangeable is how projects pick the wrong blast radius.

Question Staff NL→SQL Open agent Leash model
Customer on a channel? No Maybe Yes
Can invent a query? Yes (within roles) Yes No
Creates B1 documents? Usually no Unclear Only published writes after verify
Audit of every refusal? Varies Often weak Required

When to use which model

  1. Need reports and ad-hoc questions inside B1 for staff: Model 1.
  2. Need customers answered from live stock, orders and invoices on every channel: Model 3.
  3. Need the model to invent tools as it goes: Model 2 (and accept the risk), or don’t.

If this is the pattern you need

BE AI is Binary Elements’ leash-model agent: one brain across channels, SAP Business One live under published operations, permissions checked on every request. Product: BE AI. Connector: SAP Business One. Other systems: connectors. Talk to us: csteam@binaryelements.com · 03 9020 7900.

Questions this page usually raises

The same three questions come up whenever someone is choosing a risk model, not a feature list.

Is this the same as an AI receptionist?
No. Receptionist framing is the wrong product. BE AI answers from live systems under a published allow-list.
Do you need SAP certification to use this pattern?
No. Connecting to a customer’s Business One over APIs / Service Layer does not require an SAP-certified add-on. Certification matters for store listings and some partner motions — not for the leash itself.
How is this different from Boyum’s AI Assistant?
Boyum’s assistant is an in-ERP natural-language tool for staff (NL→SQL, read-focused). BE AI is customer-facing across channels with published operations and limited verified writes.