AI Goals let Sift read every incoming case across your channels and take the actions you allow — drafting replies, tagging, routing, and escalating — all inside the guardrails you set.
Think of a goal as a teammate you brief once. You tell it which conversations to watch, what you want it to achieve, and exactly which actions it's allowed to take. Then it works every matching case, around the clock — and shows you everything it did.
Point a goal at one of your saved searches — like Lost item or App store — so it only acts on the cases you mean.
Write what you want in plain English — or click Draft my goal and let Sift write the instructions for you.
Tick only the actions it may take. Every customer reply is drafted for one of your agents to review and send.
A live activity feed shows every case a goal touched, what it decided, and why — so you stay in control.
Everything lives in one place inside Sift. From anywhere in the app, follow this path:
Open Settings from the bottom of the left sidebar, choose the Automation section, and open Workflows. You'll see three tabs — AI Goals, Rules, and Schedules. Stay on AI Goals. Every goal you create lives here, and you can switch any of them on or off at any time.
Click + New goal on the AI Goals tab. You'll fill in a short, friendly form — here's what each part means.
What you'll call this goal, e.g. "Lost Item Assistant." Just for your team.
Choose which conversations the goal runs on. Leave it blank and it runs on every case; pick a saved search to narrow it down. We've already mapped each recommended goal below to a search you already have.
Describe the goal in plain language. Type @ to reference your tags, queues, channels, or knowledge base. Not sure how to phrase it? Click Draft my goal and Sift writes a first version for you to tweak.
The goal can only do what you check here: Draft a reply, Use saved responses, Add note, Add tag, Assign agent, Move to queue, Close case, or Notify Slack. Nothing else.
Switch this on to let the goal search your Help Center before it answers, so replies are grounded in your real policies and articles.
When a goal drafts a reply, it lands in your team's review queue. An agent reads it, edits if needed, and sends — so a person is always the one who replies to your customers.
Flip the toggle from Paused to Active and the goal starts working new cases right away. Flip it back any time to pause instantly.
Seven goals to begin with, each wired to a saved search and queue your team already has. We've grouped them into three phases so you can start safe and expand as your confidence grows.
Recognizes safety-critical situations the moment they come in, writes a clear internal summary, tags it, moves the case out of the public queue into your HQ Only queue, and pings the team in Slack. It never replies to the customer on its own and never makes promises — a person always handles these. This goal takes priority over all the others.
This goal handles safety-critical cases: anything involving physical safety, injury, an accident, assault, harassment, a minor at risk, or a serious threat. Do not reply to the customer and do not make any promises. Write one clear internal note summarizing what happened and what we know, add an urgent tag, move the case to the HQ Only queue, and send a Slack notification to the safety channel so a person picks it up immediately. This goal takes priority over every other goal.Paste into the Draft my goal box in Sift
Drafts the standard "here's how to get your item back" reply using your Help Center and tags the case. One of your highest-volume, most repetitive request types — perfect for a consistent first response every time. If a lost-item case is really about safety, it hands off to Goal 1.
This goal helps riders who left an item in a Lyft vehicle. Draft one friendly, concise reply that walks the rider through reporting a lost item in the Lyft app (open Ride history, select the trip, tap Find Lost Item, then contact the driver), using our Help Center for the exact steps. Match the rider's language. Add a lost-item tag and route the case to the support queue. Never promise the item will be found and never offer compensation. If the message is actually about a safety incident, take no action and let the safety goal handle it.Paste into the Draft my goal box in Sift
Summarizes each appeal in a clean internal note, tags it, and moves it to your driver-appeals queue so nothing sits unseen. It never promises reinstatement and never states account status — it sets your team up to make the call quickly.
This goal handles drivers appealing a deactivation. Do not reply to the driver. Write a short internal note summarizing the appeal and any details the driver provided, add a deactivation-appeal tag, and route the case to the driver support and appeals queue. Never promise reinstatement, never state or guess the account status, and never share internal reasons for the deactivation.Paste into the Draft my goal box in Sift
When someone raises a support issue out in the open — a complaint, a question, a ride or account problem in a public post or @mention — it drafts one short, friendly public reply that acknowledges them and invites them to DM the team, then tags the case and moves it into the matching DM queue so the conversation continues in private. The classic social-support move, handled the same way every time — and one of the strongest fits for how your team already works.
This goal handles public posts and @mentions where a customer raises a support issue — a complaint, a question, or a problem with a ride or their account — out in the open on social. Draft one short, friendly PUBLIC reply that acknowledges them and invites them to send a direct message with their details so the team can help privately. Never ask for personal or account details in the public reply. Add a tag, and move the case into the matching direct-message queue for that platform so the conversation continues in private. If it is a safety issue, take no action and let the safety goal handle it.Paste into the Draft my goal box in Sift
For common questions — account access, app help, basic charges, how-to — it drafts one concise reply that cites your Help Center and matches the customer's language. Anything that needs an account lookup, or mentions lost funds, is handed straight to a person. This drops into the exact draft-and-approve flow your team runs today — Sift just writes the first version.
This goal answers common rider questions such as account access, app problems, how to use a feature, and basic billing questions. Draft one concise, friendly reply grounded in our Help Center, and match the customer's language. Only say what our Help Center supports; never invent policies, refunds, or steps. If the question needs an account lookup, involves a disputed or unauthorized charge, or mentions lost funds, do not answer it: leave it for a person and add a note explaining what is needed.Paste into the Draft my goal box in Sift
Sorts each review — bug, performance, feature request, praise, or possible safety — tags it, moves it into your App review queue, and drafts a short, in-language response for negative reviews. Turns a noisy review stream into organized, ready-to-send replies.
This goal handles App Store and Google Play reviews. Read each review and classify it as a bug, a performance issue, a feature request, praise, or a possible safety or fund-loss concern, and add the matching tag. Route the case to the right place based on that classification. For negative reviews, draft a short, genuine reply in the reviewer's language that acknowledges the issue and points them to support. Never promise a fix or a timeline. If a review describes lost money or a safety issue, hand it to the safety goal.Paste into the Draft my goal box in Sift
Tags and closes conversations that are clearly off-topic or already resolved — but only when it's obviously safe to do so. It never sends a customer message; it just keeps your queues focused on what actually needs a person.
This goal keeps the queues tidy. When a conversation is clearly off-topic, spam, or already resolved, add the matching close-reason tag and close it. Only act when it is obvious; if there is any doubt about whether the issue is resolved, leave it open. Never send a message to the customer.Paste into the Draft my goal box in Sift
You don't flip everything on at once. Build trust in stages.
Safety escalation, lost items, and deactivation routing. Watch the Goal Activity feed to see exactly what each goal does on real cases — and every drafted reply waits in your team's review queue, so nothing reaches a customer without an agent.
Public-to-DM deflection, rider support answers, and app-review responses. They land in the same approve / decline queue your agents already use, so it feels familiar from day one. Public → DM is a great first one — it works well today with just a saved search on public mentions.
Let close-out keep your queues tidy in the background, and fine-tune each goal's instructions as your team learns what works. Your agents stay in the loop on every customer reply, start to finish.
Every customer reply Sift writes is a draft your team reviews and sends — Sift never messages a customer on its own.
Flip any goal from Active to Paused instantly. It stops on the next case.
The Goal Activity tab shows every run — what it did, on which case, and why.
A goal can never take an action you didn't tick — no exceptions.
Tweak the instructions or change which saved search a goal applies to whenever you like.
If a goal isn't confident, it leaves the case for your team rather than guessing.
We'll set up your first three goals together on a quick call, walk your team through the activity feed, and tune the instructions to sound like Lyft. From there, you're in the driver's seat.