Cost savings in invoice reconciliation using Agentforce.

Inspiration

I have worked in Retail and Semi-Conductor industries, which have 1000+ sites of operational facilities with 1+ million vendors and 10+ million orders. In-correct invoice costs $8-$10 million / year per facility. This is daunting task for, Head of Operation, to detect, track and take corrective actions to reduce loss.

What it does

Each site places orders for particular materials/services based on pre-approved order budget. Each site has pre-selected set of vendors, who fulfill the orders each month. By end of month vendor submit fulfilment status to site manager for approval and release of funds. Site managers, check order status, vendor delivery status, advance payment status (for pre-approval of budget) before authorizing vendor fulfilment status.

In the afore use-case, am solution does following;

a) Flow pulls order details from Salesforce CRM b) Flow also pulls vendor status details from Salesforce CRM c) Actions check if vendors have fulfilled the orders per site and here we created two flows. d) If Vendors haven't fulfilled the vendor status , then we have trained LLM model on False Positive use-cases, e) If Vendors have fulfilled the vendor status, then we have trained LLM model on True Positive use-cases.

How we built it

a) We have all our data created in Salesforce CRM i.e 5 tables. Table : ORDER
Order ID
Order Date
Order Status
Location ID Order Value

Table : VENDOR Vendor ID Vendor Name Materials Location ID Order ID

Table : LOCATION Location ID Order ID Vendor ID

Table: VENDOR_FullFilment
Vendor ID
Location ID Order ID
Order_Delivered

Table : A/P APID Order ID AP_Approved_Amount AP_Approval_Status

Our solution had following business rules :

1 Order ID will have 1 : 1 Mapping with 1 Vendor ID
1 Order ID will have 1 : 1 Mapping with Location ID
1 Vendor ID could have 1: n Mapping with Order ID
1 Location ID could have 1 : n Mapping with Order ID
1 Location ID could have 1 : n Mapping with Vendor ID
1 APID will have 1 : 1 Mapping with Order ID

Challenges we ran into

We ran into multiple challenges *1) How to bring data for LLM into Salesforce cloud: * We have multiple choices a) Connect to external Order Management tool, Vendor Management tool, Advance Payments tool b) Create external db and link with Saleforce c) Create data using AI , and upload via Salesforce Cloud However after exploring all the options, based on $$ and infrastructure , feasible option was to create tables inside Salesforce CRM. Challenges also made use understand power & flexibility of Salesforce CRM tool.

2) Developing Action: We had difficulty in identifying right actions i.e, what actions even to create , for whom are the actions are going to be created, what benefits each of the actions going to provide etc.

It was learning experience, in the end we learnt building Actions for LLM is primarily driven by Persona i.e in our case Head of Operation (per site).

Accomplishments that we're proud of

From start to finish , inspite of lack of knowledge on Salesforce and its capabilities , we built working prototype in 2 weeks.

What we learned

Salesforce agentforce is super powerful , if only you know how to use it Extended modules, API ,features of larger Salesforce cloud is all but available to address use-case of any industry.

What's next for Cost savings in invoice reconciliation using Agentforce.

We are a startup company, started during worst possible economic condition 2025. Thus our focus is COST SAVINGS to our customers. We want to use our prize money to buy cloud infrastructure, build additional use-cases, and help customer (both non-salesforce and salesforce) to appreciate agentforce and help save them millions of $$.