Shopify Flow or a custom AI worker for order exceptions

The short verdict

Start with Shopify Flow if your order exceptions follow rules you can write down in advance. A rule such as "tag any order over a set value as priority" or "notify the warehouse when a fulfilment is late" suits Flow well. Move toward a custom AI worker when the work starts with a customer's sentence rather than a store event. A message that says the box arrived crushed, with a photo attached and a question about a refund, is a different kind of task. Flow can be told that an order was fulfilled. It cannot read what the customer wrote and decide what to do next.

Most businesses need both. Flow handles the mechanical steps inside Shopify, and a custom worker handles the reading and replying that sits outside it.

Side-by-side comparison

The table below compares the two options on the points that usually decide the choice. Shopify's plan names and prices are taken from the 2026 figures that third-party pricing guides publish, because Shopify's own pricing page changes often. Confirm the current figures on Shopify's site before you budget.

What Shopify Flow does well, and where it stops

Flow is Shopify's own automation tool. It watches for events in your store and runs a chain of steps when an event happens. Shopify's help documentation describes it as a tool for automating tasks inside your store and across your apps. Each workflow starts with one trigger, then adds conditions and actions. Flow is free for stores on the Basic, Grow, Advanced and Plus plans, and most features are available on every one of those plans.

Flow's limits matter when exceptions grow. In the Shopify documentation, a store can hold up to 1,000 workflows, counting active and inactive ones. A workflow can include up to 40 wait steps. Get-data actions return lists of no more than 100 items. A scheduled trigger can run as often as every 10 minutes. Manual bulk runs cover at most 50 orders at a time. Third-party reviewers also report that run history is kept for 14 days, so you will want your own record of exceptions if you need an audit trail beyond that window.

Flow also cannot read a message from a customer and understand it. Its triggers are events the store generates, not the content of an inbox. The AI assistant built into Flow, known as Sidekick, helps you write a workflow from a description. It does not answer a customer or check a courier's tracking page on the customer's behalf. For a team that spends its day reading "my parcel came broken, what now?", Flow can tag the order after the fact but cannot resolve the case.

What a custom AI worker does differently

A custom AI worker starts from the message. It reads the email or chat, identifies the order by name, email or tracking number, and pulls the order record, the fulfilment status and the courier scan history. It then compares what the customer says with what the systems show. If the courier shows delivery and the customer reports damage, the worker can draft a reply that asks for a photo and points to your returns policy. If the parcel is stuck at a depot, it can draft a delay notice with the courier's latest status.

The useful part is the control you set. A well-built worker has refund limits, a list of cases that always go to a person, and a log of every draft and action. Richpanel's public materials describe the same idea of guardrails, such as refund limits, approval steps and escalation rules, although that is one vendor's approach and not a standard. You decide what the worker may send on its own. For a first version, most owners keep it in draft mode, where a person reviews each reply before it goes out.

Consider two stores with the same volume. A homeware retailer shipping from the United Kingdom to Germany gets many "where is my order" messages that depend on the carrier's scans. A supplement brand based in the United States gets damage claims with photos that need a judgement about whether a refund is due. Flow can route both kinds of order to a queue. A custom worker can answer the first kind and prepare the second for a person to decide.

Who each option is right for

Shopify Flow suits a store whose exceptions are few and predictable. If your rules fit on one page and rarely change, Flow is the cheaper and faster path. It also suits a team with no developer, because the editor is built into the admin and Shopify provides a template library. Flow is a sensible first step before you consider anything else, and it does not lock you into another vendor.

A custom AI worker suits a business where the inbox is the bottleneck. That usually means a steady flow of mixed-language or free-text messages, several systems to check before answering, and rules that change with the season or the carrier. It also suits businesses with more than one sales channel, where order data sits in more than one place. A five-person firm and a group with thousands of staff can both have that problem. The size of the team matters less than the share of work that starts with reading.

Total cost of ownership beyond the sticker price

Flow's visible cost is close to zero on top of your Shopify plan. Its hidden cost is staff time spent on workflows that break or fall out of date, and on exceptions Flow cannot see. The cost of a custom worker is mostly up front and in upkeep. The build, the integrations, the testing on real messages and the review of drafts all take time. Those items are not published by any vendor in a form you can compare, so ask for a written scope with each one priced separately.

Breakage is the cost owners most often miss. Shopify and carriers change their interfaces, and any integration that depends on them needs a person to fix it. A custom worker also needs its instructions updated when your refund policy changes, and its answers need periodic sampling. Plan a monthly review of a sample of drafts, and budget someone's time for it.

Vendor lock-in is a risk in both directions. Flow lives inside Shopify, so leaving Shopify means rebuilding your workflows elsewhere. A custom worker avoids lock-in to a SaaS helpdesk, but it creates dependence on whoever built it. Insist that you own the prompts, the rules, the logs and the integration code, and that you can export them.

Data handling adds cost and obligations. Shopify's help documentation says it stores some merchant and customer personal data in Europe, and it moves data across Google Cloud regions to keep the service running. Under the General Data Protection Regulation (GDPR) in the European Union and the United Kingdom, and under the California Privacy Rights Act (CPRA) in the United States, you remain responsible for how customer data is processed. A custom worker sends message text to a model provider, so you need a data processing agreement with that provider and a clear record of where the data is hosted.

The custom-built alternative, and when it wins

A custom build does not have to mean a large project. A practical first version connects three things: your helpdesk inbox, your order system, and one courier's tracking data. It drafts replies for a set group of exception types, such as late delivery, damaged item and wrong item. It sends nothing until a person has reviewed a few hundred drafts and confirmed that they are accurate.

A custom worker wins in three situations. The first is when exceptions are frequent and need reading, not only tagging. The second is when the rules change often enough that rebuilding Flow workflows each time costs more than maintaining one set of instructions. The third is when the business runs on systems outside Shopify, so that Flow cannot see the full order. Where those conditions are missing, a custom build spends money on capability you do not need.

Be clear about what automation cannot do. A worker cannot decide a disputed claim with a legal or safety angle, and it should not promise a refund that your policy does not allow. It can misread sarcasm, a rare language or an order that matches two customers. Build the escalation path first, and keep a person on every case with a legal, safety or high-value element. If your exceptions are few and each one needs a judgement call, a custom worker will not save much, and a clear written process with Flow for the routine steps is the better use of money.

Migration notes

If you already use Flow, keep it running while you add a worker. Flow can keep tagging orders and routing alerts, and the worker can read the messages and reference those tags. Do not delete working workflows until the new path has run side by side for a few weeks. Export your current workflow list so you have a record of what each one does.

Move the rules first, then the messages. Write down each exception type, the decision rule, and the reply your team sends today. Those written rules become the worker's instructions. Then connect the worker to a test inbox, and compare its drafts with what your team actually sent.

Check three things before you go live. Confirm which personal data the worker stores and for how long. Confirm which person approves refunds above a set amount. Confirm how a customer reaches a human when they ask. Keep an exception log that records the order, the message, the draft and the final action, so that you can answer a dispute later.

Where automation is not worth it

If you handle fewer than a few dozen exception messages a month, a custom worker rarely pays back its build. If your rules change every week, a worker still needs a person to maintain its instructions, so the saving may be small. If your order data lives in a paper or spreadsheet process, fix that first, because neither option can read data that does not exist in a system.

How to decide this week

Pull the last 200 order-related messages from your inbox and sort them into three groups: those Flow could handle from a store event, those that need reading and checking across systems, and those that need a human decision. If most fall in the first group, use Flow and stop there. If the second group is large and steady, price a custom worker against the hours your team spends on it today. Write your current answer to each exception type before you talk to any vendor or builder, because that list becomes the scope.