eCommerce Fastlane

Free every Thursday · 59,694 Readers

Fastlane Insider Newsletter: What's working for Shopify in the AI era?

Ecommerce Automation for Shopify: Which Back Office Jobs Should AI Run First?

Steve Hutt
Published:
November 16, 2026
Clay render of a shipping box, checklist clipboard and calendar card linked by blue lines to a gear, with a teal approval check in front
An AI agent will happily run a job nobody has written down. It will also run it wrong at 3 a.m. with your name on it.

Quick Decision Framework

  • Who This Is For: Shopify founders and operators doing roughly $10K to $1M a month who spend hours each week tagging orders, checking stock, building reports or copying data between apps, and are weighing an AI tool or agent to take it over.
  • Skip If: You want AI to answer customers (that is support, covered in the AI customer service guide linked below) or to write product copy at catalog scale.
  • Key Benefit: A one sentence test for what to automate, a free first build in Shopify Flow, and a clear line between what an AI agent can run alone and what a person approves.
  • What You’ll Need: A list of the repeat jobs your team did last week, admin access to Shopify, and the Shopify Flow app installed.
  • Time to Complete: 12 minutes to read, about an hour to write the job list, and an afternoon to build the first two workflows.

Let’s say you’re doing $40,000 a month and Monday morning looks the same every week. You export last week’s orders, tag the first-time buyers by hand, scan a spreadsheet for anything running low, email a supplier, and paste a few numbers into a report nobody reads until Thursday. None of it is hard. All of it is time, and every vendor in your inbox says an AI agent could do it while you sleep.

Some of it, an agent could. But the order matters. On episode 498 of the podcast I walked through one Shopify Flow rule that tags every first-time Black Friday buyer as the order comes in. My point was about the cost: “It’s literally only 10 minutes.” Skip it and you can’t go back in February and tag the buyers you missed. That is what good back office automation looks like. A job you already understand, written down, running every time, with no new subscription.

This guide owns back office automation for a Shopify store: which jobs to automate, Flow and Sidekick first, inventory forecasting, reporting, and where an AI agent needs a person beside it. Cash and margin live in the ecommerce accounting pillar, shipping and warehousing in the fulfillment pillar, and AI that answers customers in the AI customer service agent guide.

What ecommerce automation means for a Shopify back office in 2026

Ecommerce automation in a Shopify back office means software doing repeat operational jobs without a person starting each one: tagging, alerting, reordering, reporting and moving data between apps. In 2026 it comes in three layers, and most stores need the first two long before the third.

The first layer is rules. Something happens, a condition is checked, an action runs. An order comes in from a new customer, so tag it. Stock on a product drops below 20, so send a message to the team channel. This is what Shopify Flow does, and rules do exactly what you told them, every time, which is their strength and their limit.

The second layer is AI that builds the rules for you. Shopify’s help center says Sidekick “can build Shopify Flow workflows that do repeated work for you,” and that it presents changes for your review before applying them. You describe the job in plain words, Sidekick drafts the workflow, and you check it. The work still runs as a rule, so it stays predictable.

The third layer is AI that makes the call itself: an agent that reads your sales, decides a reorder quantity, writes the supplier email or pauses an ad. This is where most of the vendor pitches sit, and where the risk sits too, because the agent’s output depends on judgment you can’t see. Aras Kolya, Chief Revenue Officer at ShipMonk, framed the useful version on episode 410 as using AI “to operationalize and not just analyze.” I agree with the aim. I’d argue most stores should get the first two layers running before they let the third one act.

Automate the job you already do by hand, not the one you wish you had

The right first automation is a job someone on your team did by hand last week and can describe in one sentence: when this happens, check this, then do this. If nobody can say that sentence, the job is not ready to automate, and an AI agent will not fix that. It will guess.

Start with a list. Ask everyone who touches the back office, including you, to write down every repeat task from the last two weeks and roughly how long each took. Most lists I hear about from merchants and agency partners come back with the same handful: tagging orders and customers, flagging risky orders for review, hiding products that sell out, low stock alerts, weekly sales numbers, and copying orders or inventory between Shopify and another channel or app.

Then write each one as a trigger, a condition and an action. “When an order is created, if it is the customer’s first order, add the tag first-order.” “When a product’s inventory drops below 20 units, send a message to the ops channel.” If you can’t write it that way, the job has a judgment call buried inside it, and you either make the rule explicit or keep a person on it.

Here is the arithmetic that makes the list worth an hour. Illustrative example, assuming 1,500 orders a month and 30 seconds of hand work per order to tag it and check it for risk: that is about 12.5 hours a month, more than a working day and a half, spent on a job a free rule does in the background. Your numbers will differ. The point is to price each job in hours before you price any software.

The list does a second job too. It shows you what is broken underneath. Greg Tull, Director of Marketing at Classy Llama, put it bluntly on episode 424: “if your data is bad, then all you’re doing is using AI to get bad answers faster.” A tagging rule that depends on a product type field half your catalog leaves blank will quietly mis-tag half your orders. Fix the field first.

Build it in Shopify Flow first and let Sidekick draft the workflow

Build your first back office automations in Shopify Flow and let Sidekick draft them, because Flow is a free app on the Basic, Grow, Advanced and Plus plans as of November 2026 and Sidekick turns a plain description into a workflow you review. Install Flow from the Shopify App Store, give the staff who build workflows Flow permissions, and start with the two jobs at the top of your list.

Flow works on triggers, conditions and actions, the same sentence you wrote above. It also reaches into many of the apps you already run, so a workflow can tag a customer in Shopify and pass that tag to your email platform or helpdesk. If an automation touches email flows themselves, the Shopify email marketing pillar covers which flows to build first.

Let Sidekick write the first draft. Describe the job in the words you used on your list, and Sidekick builds the workflow for you to review. Read every step before you turn it on. Check the condition does what you meant (first order for this customer, not first order today), and test it on a few orders before it runs on all of them.

Know where Flow stops. Shopify’s Flow page says some features are limited by plan: the Send HTTP Request action, which lets a workflow call an outside system, is on Grow, Advanced and Plus, and Plus stores can use tasks from custom partner apps. Usage limits also follow your plan. If a job needs to talk to a system that has no Flow connector, that is a reason to look at the Shopify plan upgrade math before you buy a separate automation tool, because the plan may already cover it.

Two jobs I would not hand to Flow alone: anything that changes a price or a discount across the catalog, and anything that sends a message to a customer that a person has not written or approved. Flow will do both if you build it. Whether it should is the question in the approval section below.

When inventory forecasting software earns its fee

Inventory forecasting software earns its fee when Shopify’s own inventory tools and a spreadsheet can’t tell you what to reorder and when, which usually means hundreds of products, long supplier lead times, bundles you assemble, or several channels drawing on the same stock. Below that, Shopify’s native purchase orders, a reorder point per product and a low stock alert in Flow cover most stores.

Check the native tools first, because they changed in 2026. Shopify retired its Stocky inventory app on August 31, 2026, and its guide to moving off Stocky lists what replaces it in the admin: purchase orders, transfers, stock counts and inventory reports. The same page says Sidekick helps forecast demand, flags items to replenish and “can also create purchase orders for you automatically.” It also lists the gaps, including no native report of purchase order spend by supplier.

Dedicated software still has a job past those gaps. On episode 455, Riikka Söderlund, Chief Operating Officer at Katana, said Shopify is the world’s best selling platform, “but inventory is its own.” She works for an inventory software company, so weigh that accordingly. It is most true for stores that manufacture or assemble their own products, where the cost of a finished item depends on parts the admin doesn’t track.

Riikka made a second point on that episode that matters for AI. Inventory is product data. Google checks stock when it decides which products to show, and AI shopping assistants read the same signal, so a store whose stock levels update once a day is telling them less than a store that updates them as they change. The agentic commerce pillar covers what those assistants read.

Here is how I’d decide. If you sell a few dozen products from one location with short lead times, use Shopify’s purchase orders, set a reorder point per product, and let a Flow alert tell you when to look. The formulas are in Fastlane’s piece on the inventory math that breaks past seven figures. If you have hundreds of products, bundles you assemble yourself, or Amazon and TikTok Shop pulling on the same stock, a dedicated planning app is worth trialing. Katana is one, and the Shopify App Store has several others; trial at least two against your own sales history before you commit. Don’t wait for the spreadsheet to break if lead times run to months and minimum orders are large: one bad reorder can tie up your cash for a season, so compare the native tools and a planning app before that order, not after.

Whatever you buy, let it suggest and let a person order. A reorder is cash leaving the business, and the right quantity depends on things the software can’t see: a supplier you don’t trust, a launch next month, a cash squeeze. How much stock your cash can carry is a finance question, and the accounting pillar linked above covers it. Where the stock sits and who ships it is the fulfillment pillar’s job.

What Fastlane learned running its own back office on AI agents

Fastlane Media is a media company, not a store, but most of its back office now runs on AI agents, and the lesson so far is that the agents rarely failed on intelligence. They failed on identity and money, which is exactly where a person needs to stay in the loop.

Here is what runs. Scheduled AI tasks build a morning brief, check on sponsorship deals, and sweep the shared inbox every few hours. The sweep has standing rules a person wrote down first. WordPress plugin update notices get closed with no reply. Sitemap monitoring alerts that fire all the time get deleted. Newsletters get tagged and closed. Link swap requests, which we don’t accept, get a standard decline I wrote and approved once. Some pricing replies follow price steps I set in advance, and the agent can’t go outside them. Anything else about revenue, a podcast guest, a newsletter sponsor or a client order goes to the top for a person.

Here is what broke. Replies the agent sent through the helpdesk connector went out with a blank sender name, and the person who handles our billing pointed out that a message from lowercase “steve” looked like spam. Now the agent sends replies from my own email in the same thread and closes the ticket as a separate step. Later, she saw a run of automated looking messages with her copied in and assumed a bot had done the work on its own. Now she is copied only where billing needs her, and fix requests go back to the client alone. And once, a drafted reply pulled a price from a different client’s deal: the thread said one rate, the draft said a lower one. The rule now is that no price goes into a draft unless the thread states it.

There is a subtraction in there too. The agent can answer on a Sunday. It doesn’t, for anything that isn’t revenue, because a Sunday reply on a routine request reads as too keen. Those get held until Monday morning. Doing less was the better call.

The pattern under all of it: the agent is excellent at sorting, closing and drafting, and it needs a person wherever money, a client relationship or a name is on the line. That is the line the next section draws for a store.

What an AI agent can run alone, and what a person approves

An AI agent can run a back office job without a person in each step when a wrong result is cheap to spot and easy to undo; anything that spends money, changes a price or speaks to a customer or partner in your name needs a person’s approval. That approval comes in two forms: a person signs off each action, or a person signs off a fixed template or rule in advance, with its trigger and its limits, and the agent may only act inside it. Draw that line per job, write it down, and give the agent the narrowest access that does the job.

Back office job
Let it run alone
A person approves
Tagging orders and customers
Yes, once tested on real orders
Changes to the rule
Low stock and sellout alerts
Yes, alert and hide sold out products
Any purchase order that follows
Reorder quantities
Draft the suggestion
Every order sent to a supplier
Weekly sales and stock report
Yes, for internal readers
Numbers sent to lenders or partners
Prices and discounts
No
Every change
Inbox for vendors and partners
Sort, close noise, send approved templates
Every reply outside an approved template

Access matters as much as approval. An agent connected to your store, your inbox and your ad accounts can do anything those logins can do. Give it read access where reading is the job, write access only to what it changes, and a separate login you can switch off. Fastlane’s guide to private AI workflows for Shopify operators covers approved tools and least privilege access in more depth.

Then check the work on a schedule, not when something goes wrong. Read a sample of what the agent closed, tagged or drafted each week, and keep a log of what it did so you can undo it. If it starts doing a job you didn’t give it, narrow it. Widen it one job at a time, and only after a person has reviewed enough of its runs, including the awkward cases, to trust the next job. A quiet week in a small sample is not proof.

What to automate at each stage

Smaller stores should automate with free Shopify Flow rules and a spreadsheet, and larger stores with many products, channels and people should add forecasting software and AI agents under written approval rules. The common mistake at every stage is buying the next stage’s tool before this stage’s free rules are running. These revenue bands are rules of thumb; product count, channels and lead times matter more than revenue.

Stage
Automate this
Skip this
Starting, under $10K a month
Two Flow rules; Sidekick for questions
Paid tools a free rule could replace
Growing, $10K to $100K
Tagging, alerts, weekly report, native purchase orders
Forecasting apps, unless lead times are long
Scaling, $100K to $1M
Native planning first; planning app if gaps show
Agents that send orders unreviewed
Leading, over $1M
Agents across systems with logged approvals
One agent holding every login

If you’re just starting, you are the back office, and the job list is mostly for you. Build the first-order tag and a sellout alert in Flow, use Sidekick to answer questions about your own store data, and don’t pay for a tool that does what a free rule could. The exception is a small catalog with long lead times, which belongs in the forecasting section above, not here. The habit of writing a job down as a sentence is worth more right now than any tool.

If you’re growing, this is where Flow pays the most. Work down your job list two rules at a time, add a weekly report that lands in your inbox on Monday, and run reorders through Shopify’s purchase orders with a reorder point per product. From what I watched during my Shopify years, this is the stage where stores stack up apps that each automate one thing and overlap with Flow; check what Flow already covers before you install another.

If you’re scaling, the job list has outgrown one person’s head. Push Shopify’s native planning and Sidekick’s forecasts as far as they go, trial a forecasting app against your own sales history where the gaps show, let an agent draft reorder suggestions and supplier emails, and keep every send with a person. Write the approval rules from the table above into a shared doc your team can see.

If you’re leading, the work is control. Agents run across your store, inbox, ad accounts and warehouse system, each with its own narrow login, every approval logged, and someone reading a sample of the agents’ work each week. One agent with access to everything is the setup to avoid.

Disclosure: names on this page may link through /recommends/ links that can earn Fastlane a commission. Katana, ShipMonk and Classy Llama appeared as podcast guests.

Frequently Asked Questions

What is ecommerce automation on Shopify?

Ecommerce automation on Shopify is software doing repeat store jobs without a person starting each one, such as tagging orders, alerting the team when stock runs low, hiding sold out products, building weekly reports and moving data between apps. Most of it runs as rules in Shopify Flow, a free app on the Basic, Grow, Advanced and Plus plans as of November 2026. AI adds two things on top: Sidekick can draft Flow workflows from a plain description for you to review, and AI agents can make some decisions themselves. Start with the rules and add AI decisions only where a person still approves anything that spends money, either each time or through a rule they signed off.

What should I automate first in my Shopify store?

Automate first the job someone already does by hand every week that you can write as one sentence: when this happens, check this, then do this. For most stores that is tagging first-time customers, flagging risky orders for review, low stock alerts and hiding sold out products. Write the list of repeat jobs from the last two weeks, note roughly how long each took, and build the two biggest in Shopify Flow. If a job can’t be written as a trigger, a condition and an action, it has a judgment call inside it, so keep a person on it until the rule is clear.

Can Shopify Sidekick build automations for me?

Yes, Shopify Sidekick can build Shopify Flow workflows from a plain language description, according to Shopify’s help center as of November 2026. You describe the job, Sidekick drafts the workflow, and it presents changes for your review before applying them. Read every step before you turn a workflow on, check the condition matches what you meant, and test it on a few real orders first. The finished workflow runs as a Flow rule, so it behaves the same way every time, which makes it safer than an AI agent deciding each case on its own.

When does a Shopify store need inventory forecasting software?

A Shopify store needs inventory forecasting software when a spreadsheet can no longer tell it what to reorder and when, which usually means hundreds of products, supplier lead times measured in months, bundles assembled in house, or several channels drawing on the same stock. Check Shopify’s native tools first: since Stocky was retired on August 31, 2026, the admin covers purchase orders, transfers, stock counts and inventory reports, and Sidekick can forecast demand and draft purchase orders. If you do need an app, trial at least two against your own sales history, and keep a person approving every purchase order.

Is it safe to let an AI agent run my store’s back office?

Letting an AI agent run your back office is reasonable only for jobs where a wrong result is cheap to spot and easy to undo, such as sorting an inbox, tagging orders or drafting a report, and even those need supervision. Don’t let one spend money, change prices or message customers and partners in your name unless a person approves each action or has signed off the template or rule it follows. Give the agent the narrowest access the job needs, using a separate login you can switch off, log what it does, and read a sample of its work every week. Widen its jobs one at a time after enough reviewed runs, and narrow them the moment it does something you didn’t ask for.

FIND US ONLINE

WEEKLY DTC INSIGHTS

TRUSTED BY THOUSANDS

TRUSTED PARTNER

Choose a language