AI at ScrollEngine
The last mile, handled on your rules.
Math makes the promise. AI helps you keep it. You set the rules. Slot capacity, cutoffs, routing and ETAs run on rules, optimization and live data you can inspect. AI works where judgment helps, and only as far as you allow.
2,000+ merchants trust us (as of Oct 2026)100+ merchants rate us 5.0 stars on the Shopify App Store (as of Sep 2026)
The Scroll · Order #1043 · Typed eventsfull delivery history
- 2:01:07 pmorder.slot_bookedStore 12 · 4–5 pm · 12/12Shopper
- 2:02:11 pmslot.capacity_reachedStore 12 · 4–5 pmRule
- 2:05:40 pmroute.plannedR-18 · 14 stops · windows keptOptimization
- 3:41:02 pmshipment.at_riskETA +9 min past windowAlert → your workflow
- 3:41:05 pmnotification.sms_sentnew ETA to customerWorkflow
- 3:43:30 pmassistant.answeredR-18 · #1043 · Driver 4AI · Early access
- 3:46:12 pmroute.resequenced#1043 up 2 stops · ETA 4:52 pmYou
- 4:52:48 pmpod.capturedphoto · inside 4–5 pmKept
Thesis
Local delivery is a coordination job. Coordination is what agents are for.
Every local order is a chain of small decisions: which slot to offer, which store fills it, which driver takes it, what to tell the customer when a stop runs long. Today people make those calls by phone, spreadsheet and WhatsApp, and the work grows with every store you add. Agents can take that on, but only if they work from a complete record of what was promised, inside rules you can see. That record is what we built first.
Ask Engine
Ask where it is and why it's late.
Which routes are running late right now?
AI · Early access2 of 9 live routes are behind: R-18 at Store 12, with 3 stops at risk, and R-4 at Store 3, with 1 stop at risk.
Records: R-18 · R-4
Why is R-18 late?
AI · Early accessStop 7 took 11 minutes longer than its planned service time, and the next 3 stops inherited the delay. Order #1043 is now 9 minutes past its 4–5 pm window.
Records: R-18 · #1043 · Driver 4
Move two stops to R-21.
AI · Early accessI read your data and leave changes to your team. Driver 7 on R-21 is available and 1.2 miles from stop 8. You can move stops in the route editor.
Records: R-21 · Driver 7 · Open route editor
Modes and controls
Governance, today and dated.
- Suggest: the Engine drafts, you do it. Ask Engine and AI steps work at this level today, read-only.Early access
- Ask: the Engine acts only after someone on your team taps Approve, with the customer message pre-written for review.Vision
- Rules: plain limits you set, such as never offering a slot above its capacity, applied the same way every time.Live
- Receipts: every agent action records what it did, why, the data it used and the rule it followed, on the order's history.Vision
- Undo: reverse an agent action. Irreversible actions, such as customer messages, default to Ask.Vision
- Pause: one switch per store stops every agent and AI step there.Vision
- Graduation to Act: a job moves from Ask to Act only after enough approvals at a high enough acceptance rate, and drops back if undo or override rates rise.Vision
In the product
Agents, one bounded job at a time.
Ask Engine
AI · Early access- role
- Answers operations questions about live orders, drivers, routes, exceptions and cash on delivery.
- trigger
- A question from someone on your team.
- actions
- Reads through read-only tools and answers with links to the records it used.
- guardrail
- Read-only; scoped to your account and each user's permissions; a daily cap; the standalone web dashboard only.
- escalation
- You stay in charge: it points you to the records, and your team decides what happens next.
- metric
- Time to answer, published once it's measured.
AI steps in Workflows
AI · Early access- role
- The AI Block summarizes, classifies or extracts; the AI Agent step investigates with read-only tools and returns a verdict with a step trace.
- trigger
- Any workflow trigger you choose, such as an at-risk alert.
- actions
- Returns text or a structured verdict to the next workflow step.
- guardrail
- Read-only tools; runs only where you place it; your model key or the platform default.
- escalation
- Your workflow decides: notify a person, open a task, or stop.
- metric
- Verdicts accepted versus overridden, published once measured.
Engine labels
We call a rule a rule.
- Rule
- What it is
- A policy you set, applied the same way every time.
- Example
- Close the 4–5 pm slot at 12 orders. Send ZIP 10012 to Store 3.
- Status
- Live
- Optimization
- What it is
- Math that finds an optimized plan inside your constraints.
- Example
- Plan 14 stops across 2 drivers, inside every time window.
- Status
- Live
- AI · Early access
- What it is
- A model that reads your data and advises. It can't change anything.
- Example
- Ask Engine explains why R-18 is late, with links to the records.
- Status
- Early access
- Agent · Vision
- What it is
- AI you allow to act on one named job, under your rules, after you approve.
- Example
- Exception Agent will propose moving 2 stops and draft the customer message.
- Status
- Vision
Data
Why the record comes first.
Why the Scroll is the right place to build promise intelligence.
Checkout apps see the promise but not the door. Dispatch platforms see the door but not the promise. The Scroll records both for every local order: promised, prepared, routed, the live ETA, and kept or broken. Across stores and merchants, that record shows which slots, stores, zones and routes hold, and which don't. It's the base we're building promise intelligence on, aggregated and anonymized, and only with consent.
This is our vision, not a shipped feature.
Data use
How your data is used.
AI features read your data within each user's permissions, and they can't change anything.
FAQ
Questions
Can your AI act on its own today?
Not on its own, and that's by design. Today our AI reads and advises, and nothing changes without you: Ask Engine and AI steps in workflows read your data, never change it, and leave every decision to your team. The agent we're building to act after your approval is Exception Agent, and its status and any committed quarter are on our roadmap.
What will Exception Agent do?
When an at-risk or late alert fires, it will propose a fix, such as a reroute, a new slot or a switch to pickup, draft the customer message and carry it out through your existing workflow and messaging actions after you approve. Version 1 runs in Ask mode only: nothing happens without approval, and unapproved proposals wait while the dispatcher still gets today's alert.
Is route planning AI?
It's mathematical optimization, which is exactly what route planning needs: it finds an optimized plan inside your time windows, shifts, breaks and vehicle capacity. We label it Optimization, and it works the same with AI switched off.
Is automated dispatch AI?
It runs on rules you can read. Automated dispatch (early access) offers orders to nearby drivers in waves by rule, and we label it that way.
What doesn't ScrollEngine call AI?
Route optimization. Automated dispatch (early access), which offers orders to nearby drivers in waves by rule. Live, traffic-aware ETAs. At-risk alerts. Location priority rules. Driver scorecards. The translate button in the widget settings. Each is a rule, optimization or a count of what happened, and we label it that way.
Our policy doesn't allow operational data to go to an outside model provider. Can we still use ScrollEngine?
Yes. Leave AI off and keep everything else: every rule, route and workflow works without it.
Is Ask Engine available inside Shopify admin?
Ask Engine runs in the standalone ScrollEngine web dashboard, alongside the live driver map and driver scorecards. Bringing Ask Engine inside Shopify admin is on our roadmap, listed with its status on our vision page.
How does ScrollEngine flag a delivery that's running late?
With live, traffic-aware ETAs and at-risk alerts that fire on thresholds you set. Every alert traces back to a rule you can read, so your team always knows why it fired.
Which model does ScrollEngine use?
Your choice: bring your own key for OpenAI, Anthropic, Google Gemini or OpenRouter, or use the platform default. When you use your own model key, your agreement with that provider also applies.
Does ScrollEngine train models on our data?
ScrollEngine has no model training of its own: the AI features send a request to the model you choose when someone asks a question or a workflow step runs. How each provider may handle that data, and which providers we use, is set out in our AI data-use statement and sub-processor list.
Start with rules you can read. Add AI when you're ready.
Book a demo to see rules and optimization run your slots and routes, every action labelled, and decide which jobs, if any, you'd let the Engine suggest, ask about or act on. Ready to price it? Request a quote for your stores.
2,000+ merchants trust us (as of Oct 2026)100+ merchants rate us 5.0 stars on the Shopify App Store (as of Sep 2026)