AI Recommendations
This is a complete, from-scratch guide for Salesforce administrators. It assumes no prior knowledge of REDA AI. It explains what the feature is, the vocabulary you’ll encounter, why it’s valuable, and every step to set it up, configure it, verify it, roll it out, and troubleshoot it. If you have never used REDA AI at all, start with About REDA AI and Installing REDA AI, then come back here.
By the end of this guide you will have a recommendation agent and a record page that shows automatic, advisory recommendations to your users. Plan roughly 15–30 minutes for a first setup. No coding is required — everything is done with clicks in Salesforce Setup and the Lightning App Builder.
Quick start (for experienced admins)
- Create an Internal REDA AI agent, make it read-only (remove action tools), and give it short “recommend next actions and risks” instructions. Enable Allow Degraded Record Context so it can read the record. Note its Agent Id.
- Open the record page in Lightning App Builder and drag on the REDA AI Recommendations component.
- Set Agent Id, leave the other options at their defaults, Save, and Activate.
The rest of this page explains each step in full detail.
1. What is REDA AI, and what is AI Recommendations?
REDA AI is an AI assistant that runs inside your Salesforce org. As an administrator, you create agents — configurable AI helpers — and place them where your users work. Normally an agent is reactive: a user types a question and the agent answers.
AI Recommendations makes the assistant proactive. You place a small panel on a record page (for example an Opportunity, a Case, an Account, or one of your custom objects). When a user opens that record, the panel automatically asks an agent to review the record and its recent activity, and it presents:
- A bottom-line takeaway — two or three lines summarising the single most important thing to know or do right now.
- Next Actions — the most useful things to do next, most important first.
- Risks / Red Flags — anything that needs attention, such as a stalled follow-up or an approaching deadline.
The recommendations are advisory and read-only — they are suggestions for the person to act on. The panel never changes any data on its own.
2. Key concepts (a quick glossary)
A few REDA AI and Salesforce terms are used throughout this guide:
| Term | What it means |
|---|---|
| Agent | A configurable AI helper you set up in REDA AI. It has a name, a purpose (its instructions), and a set of capabilities. See Understanding Agents. |
| Agent Id | The unique identifier of an agent. You enter this on the component so it knows which agent to use. |
| Channel | Where an agent is allowed to run. AI Recommendations uses the Internal channel (logged-in Salesforce users). See Internal Channel. |
| Tool | A capability you grant an agent. “Read” tools look things up; “action” tools change data. For recommendations you keep it read-only. See Understanding Tools. |
| Context Fields | The record fields you let an agent read for context. See Understanding Context Fields. |
| Lightning App Builder | The Salesforce point-and-click tool for designing record, app, and home pages. This is where you add the component. |
| Permission set | A bundle of permissions you assign to users. Users need the REDA AI permission set to use REDA AI features. |
| Record page | The Lightning page shown when a user opens a record (for example, an Opportunity detail page). |
3. Why it’s useful
- Nothing slips through the cracks. Overdue follow-ups, unanswered messages, and approaching deadlines are surfaced automatically, right on the record.
- Faster decisions. Users get a clear “what to do next” without digging through related lists, activity history, and files.
- Consistent best practice. Everyone — including brand-new team members — sees the same high-quality guidance for how to handle a record.
- Works anywhere. The same component sits on any standard or custom object, with no custom development.
- Speaks your users’ language. Recommendations are returned in each user’s own Salesforce language automatically.
- Sales — Opportunity: “Chase the buyer for a reply — three follow-up emails have gone unanswered,” and a flag: “A support issue logged in April is still open.”
- Service — Case: “Send the customer an update — no activity in 5 days,” and a flag: “Resolution deadline is tomorrow.”
- Any custom object: next steps and risks specific to your process, defined by the agent you configure.
4. Before you begin (prerequisites)
Confirm each of the following before setting up:
| Requirement | How to confirm / where |
|---|---|
| REDA AI is installed and activated | See Installing REDA AI. |
| Your users have REDA AI access | Assign the REDA AI permission set to the users who should see recommendations. See App Access & Permissions. |
| You can edit Lightning pages | You need the “Customize Application” permission (standard for System Administrators). |
| A recommendation agent | Reuse an internal agent, or create one in Step 1 below. |
5. Step 1 — Create or choose the recommendation agent
The panel doesn’t generate recommendations by itself — it uses a REDA AI agent that you configure. The agent decides what to recommend and how to phrase it. You can reuse an existing internal agent, but we recommend a dedicated one so you can tune it independently.
-
Open REDA AI and start a new agent
From the App Launcher (the grid icon, top-left) open REDA AI, go to Agent Definitions, and create a new agent. Full details: Creating an Agent.
-
Name it clearly and note the Agent Id
Give it a recognisable Display Name (for example, “Opportunity Recommendations”). Note the Agent Id — you’ll type this into the component in Step 3.
-
Set the channel to Internal
Recommendations run for logged-in Salesforce users on record pages, so set the agent’s Channel to Internal. See Configuring Agent Settings.
-
Make it read-only
Because recommendations are advisory, remove any “action” tools (create, update, delete, send, schedule, and similar) so the agent can never change data while generating suggestions. You may keep read-only tools if you want it to look up related records. See Configuring Tools.
ImportantKeeping the recommendation agent read-only is the key safety setting. It guarantees that simply opening a record and generating recommendations can never modify your data.
-
Write short, action-focused instructions
Tell the agent to recommend what to do rather than describe the record (users can already see the record). Keep points to one line. You can adapt this sample:
Sample agent instruction (adapt to your process)“You review this record and its recent activity and recommend the next best actions, and flag any risks. Start with a two- to three-line bottom-line takeaway, then a short Next Actions list and a short Risks / Red Flags list. Keep each item to one line. Do not summarise the record — recommend what to do.”
-
Let the agent see the record
So the agent can read the record’s fields, either add Context Fields for the object, or turn on Allow Degraded Record Context on the agent. The latter is the quickest option — it lets the agent read the record’s standard fields with no per-object setup.
-
Activate the agent
Make sure the agent is Active. See Activating & Deactivating Agents.
6. Step 2 — Add the component to a record page
-
Open a record of the object you want
For example, open any Opportunity from the Opportunities tab.
-
Enter the page editor
Click the Setup gear (top right) › Edit Page. The Lightning App Builder opens for that record page.
-
Find the component
In the Components panel on the left, use the search box and type “REDA”, or scroll to the Custom section, and locate REDA AI Recommendations.
-
Drag it onto the page
Drop it where you want it. A wide region near the top gives it prominence so users notice it. You can place it in the main column or a sidebar — it adapts to the width.
7. Step 3 — Configure the component
With the component selected on the canvas, set its properties in the right-hand panel. Each property is explained below.
| Property | What it does | Recommended |
|---|---|---|
| Agent Id | The recommendation agent to use (the Agent Id from Step 1). This is the only required setting. | Your recommendation agent |
| Generate on Load | On: recommendations appear automatically when the record opens. Off: users see a Generate recommendations button and choose when to run it. | On for high-value pages; off to make it opt-in |
| Panel Title | An optional heading override for the panel. | Leave blank for the default title |
| Prompt (optional) | An optional one-off instruction override for this placement. Leave blank to use the agent’s own instructions. | Blank (keeps behaviour consistent) |
| Scan Emails | Include recent emails on the record in the analysis. | On |
| Scan Tasks | Include recent tasks on the record. | On |
| Scan Events | Include recent and upcoming events on the record. | On |
| Scan Files & Attachments | Include recent files/attachments on the record. | On |
| Scan Notes | Include recent notes on the record. | On |
| Scan Field History | Include recent field-history changes (requires field history tracking on the object). | On where useful |
Turn a scan option off when a source isn’t relevant on that object, or to keep the analysis focused. For example, on a Case you might scan emails and tasks but not files.
8. Step 4 — Save and activate the page
- Click Save.
-
Activate the page if prompted. You can:
- Set it as the Org Default (everyone sees it), or
- Assign it to a specific app, profile, and/or record type using the App, Record Type, and Profile options. This lets you, for example, show recommendations only to the Sales app.
- Click Back to return to the record. The panel appears and (if Generate on Load is on) begins producing recommendations.
9. Verify it’s working
Run through this quick checklist on a real record that has some recent activity:
- ✓ The REDA AI Recommendations panel appears on the record page.
- ✓ With Generate on Load on, a short takeaway appears within a few seconds; with it off, a Generate recommendations button appears and works when clicked.
- ✓ View details opens a window with the full Next Actions and Risks lists.
- ✓ The thumbs up / down in the header respond when clicked.
- ✓ The refresh icon regenerates the recommendations and updates the “as of” time.
- ✓ The content reflects the record’s real activity (not generic filler).
If any step fails, see Troubleshooting below.
10. What your users will see
- A titled panel with a short bottom-line takeaway (two to three lines), so it never crowds the page.
- A View details link that opens a window with the full Next Actions and Risks / Red Flags lists.
- A thumbs up / thumbs down in the panel header so users can rate how helpful the recommendations were.
- A refresh icon to regenerate on demand, and an “as of” time showing when the recommendations were produced.
- If Generate on Load is off, a Generate recommendations button instead of automatic results.
11. Worked example — Opportunity
Here is a complete example you can mirror.
- Agent: Create an Internal agent named “Opportunity Recommendations”, read-only, with the sample instruction from Step 1, and Allow Degraded Record Context enabled.
- Placement: On the Opportunity record page, drag the REDA AI Recommendations component to the top of the main column.
- Settings: Set Agent Id to the new agent; leave Generate on Load on and all scan toggles on.
- Save & activate the page for the Sales app.
Takeaway: “This deal has stalled — three follow-ups are overdue and the buyer hasn’t replied in two weeks.”
Next Actions:
- Call the buyer today — email follow-ups have gone unanswered.
- Confirm the delivery date — the close date is in 8 days.
Risks / Red Flags:
- Single-threaded contact — all activity is with one person.
(Your actual results depend on the record and your agent’s instructions.)
12. Choosing what the AI scans
The scan toggles control which recent activity the agent considers. Richer input generally produces sharper recommendations, but you can turn off sources that aren’t relevant to keep the focus tight (and the response quick). Guidance:
- Emails / Tasks / Events — usually the most valuable signals of momentum or stalling. Keep on.
- Files & Attachments — useful where documents matter (contracts, proposals). Turn off if the object rarely has files.
- Notes — useful where your team logs notes on records.
- Field History — useful for spotting changes (for example a slipping close date). Requires field history tracking to be enabled on the object.
13. Automatic vs. on-demand
The Generate on Load option decides whether recommendations run automatically. Turn it on for pages where recommendations are always useful (users get value with no clicks). Turn it off when you’d rather users request recommendations explicitly — they’ll see a Generate recommendations button instead.
14. Feedback — and improving recommendations over time
The thumbs up / thumbs down in the panel header lets users tell you whether a recommendation was helpful. This feedback is stored with the conversation. Use it as a signal: if certain recommendations are consistently marked unhelpful, refine the agent’s instructions (Step 1) — for example, ask it to be more specific, or to prioritise differently. Small instruction tweaks usually produce noticeably better results.
15. Languages
Recommendations are returned in each user’s own Salesforce language automatically — there is no per-language setup. A user whose Salesforce Language (Settings › Language & Time Zone) is French will see recommendations in French. See Multi-Language Agents for how language handling works across REDA AI.
16. Best practices
- Use a dedicated, read-only agent — keeps recommendations advisory and independently tunable.
- Recommend, don’t summarise — focus the agent on next actions and risks.
- Keep it lean — scan only the sources that matter for the object.
- Pick the right load behaviour — automatic for always-useful pages, on-demand elsewhere.
- Give the agent record access — Context Fields or Allow Degraded Record Context.
- Review feedback regularly and refine the agent’s instructions.
- Use a separate agent per object when different objects need different guidance.
17. Rolling it out safely
- Build and test in a sandbox first. Set up the agent and page in a sandbox and confirm the recommendations are accurate and useful on real-looking data.
- Pilot with a small group. Activate the page for one app or profile (for example, a single sales team) and gather feedback via the thumbs up/down and directly from users.
- Refine the agent. Adjust the instructions and scan toggles based on what you learn.
- Roll out widely. Once you’re happy, activate the page as the Org Default or for the relevant apps/profiles.
See Best Practices for general REDA AI rollout guidance.
18. Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| The component isn’t in the App Builder list | Not editing a record page, or the editor cached | Make sure you are editing a record page; refresh the App Builder. |
| Panel says the Agent Id is required / not configured | The Agent Id property is blank | Set Agent Id to your recommendation agent and save. |
| “No active Context Field configuration found” | The agent can’t read the record’s fields | Add Context Fields for the object, or enable Allow Degraded Record Context on the agent. |
| An error instead of recommendations | The user lacks REDA AI access, or the agent is inactive/not Internal | Assign the REDA AI permission set; confirm the agent is Active and on the Internal channel. |
| Recommendations look thin or generic | Little recent activity, or key sources turned off | Turn on the relevant scan toggles; recommendations are richest on records with recent emails, tasks, and events. |
| Recommendations are too long / read like a summary | Agent instructions aren’t action-focused | Tighten the agent prompt to lead with a short takeaway and one-line action points (see Step 1). |
| Recommendations are slow | Very large amount of activity being scanned | Turn off sources you don’t need, or set Generate on Load off so they run on demand. |
| Wrong language | The user’s Salesforce language setting | Recommendations follow each user’s Language (Settings › Language & Time Zone). |
Frequently Asked Questions
No. Recommendations are advisory only. Configure the agent as read-only (remove action tools) so generating recommendations never writes data.
Yes — place it on any record page, standard or custom, and point it at your agent. No per-object development is required.
Yes. Create a separate agent per object (for example, one for Opportunities and one for Cases) and point each object’s component at the matching agent.
Yes. Turn off Generate on Load; users then generate recommendations on demand with a button.
Each user’s own Salesforce language, automatically. See Multi-Language Agents.
Any user who can open the record page and has REDA AI access. The agent only reads data the running user is already allowed to see.
No. Everything is point-and-click — creating the agent, adding the component, and configuring options.
Yes — set the Panel Title property. Leave it blank to use the default.
It records whether a recommendation was helpful, stored with the conversation so you can review how recommendations are landing and refine the agent over time.
Next Steps
Set up the agent in Creating an Agent and Configuring Agent Settings, control what it can do in Configuring Tools, and give it the record data it needs in Adding Context Fields.