Salesforce and AI
We run Salesforce. Should we use Agentforce, or get AI built around Salesforce and our other systems?
Last updated
Use Agentforce when the work starts and ends in Salesforce: service cases, leads, opportunities and field service work orders, with data Salesforce already holds. Salesforce supports it, so talk to your account executive and Salesforce partner first. Get AI built around Salesforce when the work also touches your ERP, shared inboxes, PDF documents or field systems, and has to land in each as a checked draft. Many organisations use both.
Who this is for
This page is for sales operations, customer service, finance and IT leads in organisations that run Salesforce alongside an ERP such as SAP, NetSuite, Dynamics 365, Business Central or Pronto Xi. It sets out what Salesforce publishes about Agentforce, where an AI layer across several systems is the better tool, and who is responsible for each part.
Solve8 builds AI and integration around Salesforce for multi-site organisations in Sydney, Melbourne, Brisbane, Perth and across Australia and New Zealand. Solve8 is not a Salesforce partner. Salesforce and your Salesforce partner own your org, its configuration and Agentforce, and Solve8 works alongside them.
What Salesforce says Agentforce does
Agents by function
Salesforce lists agents for customer service, contact centres, field service, sales, employee service and IT service, with examples such as qualifying inbound leads, scheduling technician visits and updating work order status.
Where Einstein fits
Salesforce describes Agentforce as formerly Einstein Copilot and says Einstein is embedded in the Salesforce Platform. Older Einstein features and newer Agentforce agents sit on the same platform.
Data from other systems
Salesforce says agents can use external data from Data 360, and builders can use MuleSoft API connectors and Flows to reach other systems. That connection work still has to be designed, built and supported by someone.
How it is licensed
Salesforce publishes several models: usage-based Flex Credits drawn per action, per-conversation pricing aimed at customer-facing agents, and per-user add-ons for Sales, Service and Field Service. Flex Credits and Conversations cannot be used in the same org, so ask your account executive which fits your edition and contract.
Inside Salesforce or around it
| The job | Where it fits | Who to talk to |
|---|---|---|
| Summarise a case or opportunity and suggest next steps | Agentforce | Salesforce or your Salesforce partner |
| Qualify inbound leads that arrive through Salesforce web forms | Agentforce | Salesforce or your Salesforce partner |
| Sort a shared Outlook inbox, match each email to an account and log the case | AI built around Salesforce and Microsoft 365 | An AI consultancy, with your Salesforce partner for the data model |
| Turn an emailed RFQ into a quote with ERP pricing and stock | AI built across Salesforce and the ERP | An AI consultancy, with both platform partners |
| Pass a won opportunity into the ERP as a draft sales order | Integration around Salesforce | An integration or AI consultancy |
When Agentforce is enough
The data is already in Salesforce
Cases, accounts, contacts and opportunities hold what the agent needs, and staff already work in Salesforce screens.
The channel is Salesforce
Customers reach you through Salesforce web forms, chat or Experience Cloud sites, so the conversation starts inside the platform.
You want one vendor accountable
Agentforce is a Salesforce product, so support and releases stay with Salesforce and your partner.
When the AI has to span several systems
Customers still email
Orders, quote requests and complaints land in shared Outlook inboxes with PDF attachments. The AI has to read them, find the account in Salesforce and the customer number in the ERP, and draft the next step.
The ERP holds the answer
Pricing, stock, credit status, order status and invoices live in the ERP. A reply or quote built from Salesforce data alone will be wrong.
Field work closes the loop
Jobs completed in simPRO, Uptick or AroFlo have to update Salesforce and raise an invoice in the ERP.
Data must stay in your environment
The AI layer can run on your own infrastructure in Australia and reach Salesforce through an integration user that only reads and writes what the job needs.
Salesforce APIs, data quality and CRM hygiene
Use a dedicated integration user
Give the AI its own Salesforce user with a permission set limited to the objects and fields it needs, so every read and write is logged against it.
Match accounts to ERP customers
Most errors come from accounts with no ERP customer number, duplicate accounts across branches, or contacts attached to the wrong site. Fix the match key before the AI depends on it.
AI will not fix bad CRM data by itself
It can flag likely duplicates, missing fields and stale opportunities for a person to merge or close. Reps who do not update Salesforce are a process problem, and logging activity from email and calendar helps more than another reminder.
Watch API limits and field changes
Agree with your Salesforce partner how the AI layer uses the Salesforce APIs and who gives notice when a field or validation rule changes.
How a quote moves from inbox to Salesforce to the ERP
Read the request
The AI reads the email and attachments, identifies the customer and matches their part numbers to your items.
Look up both systems
It finds the account and open opportunity in Salesforce, and contract pricing, stock and credit status in the ERP.
Draft, then a person approves
A draft quote is created where your team already approves quotes, with each line traced to its source and exceptions flagged.
Hand over on acceptance
When the customer accepts, the opportunity is updated in Salesforce and a draft sales order is created in the ERP for order entry staff to release.
How Solve8 runs the work with your Salesforce partner
Discovery on one process
Fixed-fee discovery on one queue at one site, such as quote requests or the service inbox, with a measured baseline of effort and errors.
Shadow pilot
Staff keep doing the job while the AI's output is compared with theirs on live work. You set the accuracy bar before anything writes to Salesforce or the ERP.
Build into your systems
The AI layer is built on your infrastructure in Australia, using integration users your Salesforce and ERP partners agree. Workshops run on site and builds run remotely under your access controls.
Support or handover
Solve8 supports the AI layer, or hands it to your team with the code and runbooks. Your Salesforce partner keeps the org.
Related answers
Connect AI to your CRM and ERPService accounts, read-only first, approvals and audit logs.
AI for a shared inbox and CRMSorting emails, logging cases and drafting replies for approval.
AI quotes from emailed RFQsMatching SKUs, applying customer pricing and creating draft quotes in the ERP.
ERP partner or AI consultancyWho should build the AI layer, and what to ask each provider.
Sources
Salesforce, Agentforce, checked November 2026: https://www.salesforce.com/agentforce/
Salesforce, Agentforce pricing and licensing models, checked November 2026: https://www.salesforce.com/agentforce/pricing/
Questions
Who can help us add AI to Salesforce in Sydney, Melbourne or Brisbane?
For Agentforce inside Salesforce, start with Salesforce or your Salesforce partner. For AI that reads email and documents and works across Salesforce and your ERP, use an AI consultancy that works alongside that partner. Solve8 does this second kind of work across Australia and New Zealand. It is not a Salesforce partner and does not configure your org.
Can Agentforce read our shared Outlook inbox and create quotes in our ERP?
Salesforce describes connecting agents to other systems through MuleSoft connectors, Flows and Data 360. Whether that covers your inbox and ERP depends on what you license and what someone builds. Ask your Salesforce partner for a written design, and compare it with an AI layer that sits across Microsoft 365, Salesforce and the ERP.
Do we need clean Salesforce data before we start?
You need a reliable match between Salesforce accounts and ERP customers, and the fields the AI depends on filled in. You do not need the whole org cleaned first. The discovery phase measures the gaps on the one process being automated.
Where does our data go?
Agentforce runs on Salesforce's platform under your Salesforce contract. An AI layer built around Salesforce can run on your own infrastructure in Australia and read or write only what its integration user allows.
Running Salesforce alongside an ERP across several sites?
A 30-minute review with a senior consultant. You leave with a written one-page summary you can share internally.
Fixed-fee phases, agreed before we start. Stop after any phase.