Contents
- Overview
- First: check what your existing stack already does
- The six workflows worth building first
- What an order-status deflection build actually looks like
- Which systems this connects to
- Card data, customer data, and messaging consent
- Which retailers this fits
- Working out whether it pays
- Questions owners ask
E-commerce is the most tool-saturated category on this site, and the honest starting point for any retailer is that a large share of what an automation agency might sell you is already sitting switched off inside software you pay for every month. Shopify Flow, Klaviyo's flow builder, and a modern helpdesk cover an enormous amount of ground between them. Any consultancy that does not open by asking what you already own is planning to bill you for it twice.
So the useful question is not "what can be automated" - almost everything can - but "what falls between your tools". That gap is where the work is, and in a small store it is remarkably consistent. Support volume is dominated by a small number of repetitive questions, most of them about order status. Returns arrive as unstructured email and get processed by hand. Suppliers and 3PLs are chased manually. Marketplace channels, wholesale enquiries, and the retail shop floor sit outside the e-commerce stack entirely and get reconciled in a spreadsheet.
The revenue side has a similar shape. Abandoned checkout and browse-abandonment flows are standard and most stores have some version running. What is usually missing is everything harder to template: back-in-stock demand that nobody captured, a first-time buyer who never got a second-order nudge tuned to the actual consumption cycle of what they bought, wholesale enquiries that went to an inbox nobody watches, and post-purchase problems that quietly turned into a refund rather than an exchange.
One structural note before the workflows. Where-is-my-order enquiries make up a substantial share of e-commerce support volume - it is the single most cited number in the category and one of the few widely-agreed facts about it. We are not going to attach a precise percentage to your store, because it varies enormously by shipping profile and product type. But you can measure it in your own helpdesk in about ten minutes, and if it is a big number then order-status deflection is the first thing to build.
First: check what your existing stack already does
This is a strange thing to lead with on a services page, but it is the fastest way to save a retailer money and it is the test we would apply to any agency pitching us. Before scoping anything, we go through the stack and mark off what is already covered by a feature you have licensed.
Shopify Flow handles a great deal of internal, event-driven store logic - tagging, inventory triggers, fraud holds, fulfilment routing - and it is included on plans a lot of merchants are already on. Klaviyo covers the email and SMS lifecycle flows comprehensively, and most abandoned-cart and post-purchase sequences belong there rather than in anything custom. A helpdesk built for commerce will already pull order data alongside the ticket and will have some form of automated resolution for the simplest ticket types, usually as a paid add-on.
What genuinely falls outside those tools is cross-system work: anything that has to reconcile your store with a 3PL, a supplier, a marketplace channel, an ERP, or a physical shop's point of sale. Anything involving a channel the marketing platform does not own, like a WhatsApp enquiry or a wholesale phone call. Anything requiring a decision that spans systems, like whether a delayed order should be reshipped before the customer complains. And the internal operational work nobody has ever automated because it never belonged to one tool.
That is a narrower brief than most agencies offer. It is also the part where a build has a defensible return, because it is not duplicating a subscription you already hold.
The six workflows worth building first
Order-status deflection across every channel
- Trigger
- A customer asks where their order is - by email, chat, contact form, social message, or a reply to a shipping notification.
- What it does
- The enquiry is matched to the order automatically, the current carrier tracking state is fetched, and the customer gets a specific answer: where the parcel is, what the expected date now is, and what happens next. Anything that is not a clean status question - a claimed non-delivery, a parcel stuck at a facility past a threshold, an address error, a customs hold - is escalated to a human with the full context attached rather than answered. Proactively, orders that stall in transit past a threshold trigger a message before the customer has to ask.
- Why it matters
- Order-status questions are the largest single category of support volume in most stores, they are almost entirely mechanical, and they arrive disproportionately outside working hours. The proactive branch is the underrated half: telling somebody their parcel is delayed before they chase you converts a complaint into reassurance.
Returns and exchange triage
- Trigger
- A return request arrives, through a portal, an email, or a message.
- What it does
- A structured capture of order, item, reason, and condition, checked against your policy window and your exclusions, generating the label or the RMA where it qualifies and routing to a human where it does not. Reason codes are captured consistently so the data is usable. Where the reason is size or fit, the flow offers an exchange before a refund. Where the reason is a fault, it escalates immediately and captures photographs. Return reasons aggregate into a report that goes to whoever owns merchandising.
- Why it matters
- Returns are high-volume, rule-governed, and usually handled as free-text email, which is slow for the customer and produces no usable data. Converting refunds to exchanges where the customer was going to buy a different size anyway is the direct revenue case; the reason data is the compounding one.
Back-in-stock and pre-order demand capture
- Trigger
- A customer views or requests notification on an out-of-stock variant, or a variant returns to stock.
- What it does
- Demand is captured against the specific variant rather than the product, so somebody wanting a size 9 does not get told the size 12 arrived. On restock, notification goes out in the order interest was registered, with a hold window if your platform supports it. Internally, accumulated interest on out-of-stock variants is reported to whoever does buying, so the demand signal reaches a purchase order rather than dying in a notification list.
- Why it matters
- A stockout is demand you generated and then failed to serve, and unlike most marketing it costs nothing to capture. The internal reporting half matters more than the customer notification: most small retailers have no systematic view of what they failed to sell.
Post-purchase sequencing tuned to the product
- Trigger
- An order is delivered, with timing set per product category.
- What it does
- A short sequence that matches how the product is actually used: setup or care guidance shortly after delivery, a review request timed to when someone would genuinely have an opinion, and a replenishment nudge at the point the item would plausibly run out. Consumables and apparel get completely different clocks. The sequence suppresses anyone with an open support ticket, an open return, or a delivery problem, because asking for a review from someone whose parcel is missing is the classic own goal.
- Why it matters
- Repeat purchase is the cheapest revenue a store has, and generic post-purchase flows underperform because they fire on a calendar rather than on how the product is consumed. The suppression rules are what stop the whole thing embarrassing you.
Supplier, 3PL, and purchase-order chasing
- Trigger
- A purchase order passes its expected date without a delivery confirmation, or a fulfilment batch is not acknowledged by your 3PL within an agreed window.
- What it does
- An automatic chase to the supplier or provider naming the specific PO and expected date, escalating to a human on your side if unanswered, and flagging the downstream consequence - which SKUs go out of stock and when - so the escalation carries a reason. Inbound confirmations update the expected date and clear the chase. Stock levels crossing a reorder threshold raise a purchase task rather than an automated order.
- Why it matters
- This is unglamorous internal work that nobody owns, and it is where small retailers lose margin quietly. It is also the clearest example of something no single tool in your stack covers, because it spans your store, your supplier's email, and your fulfilment provider.
Wholesale and B2B enquiry routing
- Trigger
- A wholesale, stockist, trade, or bulk enquiry arrives on any channel.
- What it does
- Immediate acknowledgement and structured capture: business name, type, expected volume, territory, and whether they hold a resale certificate or trade credentials where you require them. Qualified enquiries route to whoever owns wholesale with a task and a deadline. Enquiries outside what you accept get a fast, courteous decline. Nothing about pricing is disclosed automatically.
- Why it matters
- For a lot of small product businesses this is the highest-value enquiry type and the worst-handled, because it arrives in the same inbox as consumer support and gets triaged as though it were. Separating it out is often a one-week build with a disproportionate return.
Done for you
Not sure which of these to build first?
That is what the free audit is for. We look at how enquiries, bookings, and follow-up actually move through your business today, and tell you which single workflow is worth building first - including if the answer is none of them yet.
What an order-status deflection build actually looks like
This is the build with the clearest payback in most stores, and it is also the one most likely to go wrong in a way customers notice. Assume a store on Shopify with a helpdesk, one or two carriers, and a support queue that is mostly order questions.
1. Measure the ticket mix before building anything
Pull a month of tickets and classify them. What proportion are genuinely status questions with no complication? What proportion are non-delivery claims, address errors, or customs problems dressed up as status questions? The second group is the one that damages you if it gets an automated answer, and its size determines how conservative the routing rules have to be. This measurement takes an afternoon and it decides the whole design.
2. Get order and tracking data into one place reliably
Order state comes from the store platform, but shipment state comes from the carrier, and the gap between the two is where the customer's actual question lives. That means either a tracking aggregation layer or direct carrier integration, and it means deciding what to do when the carrier data is stale or contradictory - which is not an edge case, it is a weekly occurrence. A system that confidently reports what the carrier said three days ago is worse than one that says it is checking.
3. Write the escalation rules before the answer templates
Anything marked delivered that the customer says did not arrive. Anything static past a threshold you set. Anything with a customs or address exception. Anything from a customer who has already asked twice. Anything where the order value is above a line you draw. All of it goes to a person, immediately, with the history attached. The confident-sounding wrong answer is the only genuinely expensive failure mode here, and these rules are what prevent it.
4. Build the proactive branch, not just the reactive one
Orders that have not moved in transit for a set number of days should generate a message before the customer chases. This is the half most stores skip and it is where the goodwill is: the same information delivered unprompted reads as attentive, and delivered after three days of silence reads as an excuse. It also reduces the ticket volume the reactive branch has to handle.
5. Make the handoff to a human seamless
When the automation escalates, the human should open a ticket that already contains the order, the tracking history, what the customer said, and what the system already told them. If your agent has to ask the customer to repeat themselves, the automation has added a step rather than removed one - and customers reliably describe that experience as the worst part of dealing with a store.
6. Ship it on one channel, watch it, then widen
Email first, usually, because it is the most forgiving of a slightly slow or slightly formal reply. We watch deflection rate, escalation rate, and reopened tickets for a fortnight. Reopened tickets are the number that matters: a high deflection rate with a high reopen rate means the system is closing conversations rather than resolving them, which looks good on a dashboard and is bad for the business.
The engineering here is straightforward. The care goes into the escalation rules and into resisting the temptation to automate the ambiguous cases because the deflection number looks better if you do.
Which systems this connects to
The e-commerce stack is well-documented and broadly integration-friendly, which makes this the easiest category on the site technically and the hardest commercially - because the question is nearly always whether a build is worth it next to a subscription you already hold.
Shopify is the platform we see most, and its APIs plus Shopify Flow cover a great deal of internal store logic natively. WooCommerce is common at the smaller end and is completely automatable, with the caveat that reliability depends on your hosting rather than on a vendor's uptime. BigCommerce and Squarespace Commerce also appear; both are workable.
On the marketing side, Klaviyo dominates the segment and its flow builder should own your lifecycle email and SMS. Omnisend, Attentive, and Postscript come up regularly too. Our position is consistent: build lifecycle flows there, not in a custom system, and use custom work only to feed those flows data they cannot otherwise reach.
On support, Gorgias is the commerce-native helpdesk we see most often with Shopify stores, and it has its own automation and AI resolution features - typically as paid add-ons, which is worth pricing against a custom build rather than assuming either way. Zendesk and Freshdesk appear at the larger end. Where the store also has physical retail, Shopify POS, Square, or Lightspeed enter the picture, and reconciling online and in-store inventory and customer records is frequently where the genuine unmet need turns out to be.
Feature sets, add-on pricing, and API entitlements in this category change constantly. Everything above is a scoping starting point verified against your actual accounts and plans before any design.
Card data, customer data, and messaging consent
None of this is legal advice, and your payment provider and your own advisor are the authorities. These are the structural constraints worth being firm about with any vendor, including us.
Cardholder data first, because it is the one where the answer is absolute. No automation system we build sees, stores, transmits, or logs card details. Payment lives inside your platform's payment flow, which is built and assessed for it. If a customer service workflow needs to take a payment - a made-to-order deposit, a shipping upgrade - it sends a link into the platform's own checkout and learns only whether the payment succeeded. Any vendor proposing to capture card details through their own form, chat, or phone system is creating exposure that attaches to you as the merchant under PCI DSS, not to them.
Customer data is the broader issue. A store holds names, addresses, order history, and behavioural data, and under GDPR, UK data protection law, and an expanding set of US state privacy statutes that carries obligations on lawful basis, retention, access and deletion requests, and cross-border transfer. Two practical consequences for automation design: a deletion request has to propagate to every system in the chain, not just the store, which means the data-flow diagram has to exist before you need it; and any tool that receives customer data has to be inside your processor list and covered by the appropriate contractual terms. Where a language model is involved, ask specifically for zero data retention and a written commitment against training on your data.
Messaging consent is the third. Marketing SMS is consent-based and enforced with real penalties in some jurisdictions, and transactional messages sit under different rules from marketing ones - a distinction it is very easy to blur when the same system sends both. Keep the two streams separate in design, honour opt-out globally and instantly rather than per campaign, and register your business messaging with the carriers before volume sending.
- Never let the automation layer see, store, or log cardholder data.
- Keep transactional and marketing message streams technically separate.
- Make deletion requests propagate to every system, not just the store.
- Maintain a processor list covering every tool that receives customer data.
- Ask for zero data retention in writing wherever a language model is used.
- Set and enforce a retention period for support conversations and exports.
Payment, privacy, and messaging obligations vary by jurisdiction and by your merchant agreement. Treat this as an agenda for your processor and your own advisor.
Which retailers this fits
The strongest fit is a store past the point where the founder handles support personally - roughly where ticket volume justifies at least one dedicated person - running a real stack, with either a physical or wholesale side that sits outside that stack. That combination produces the cross-system gaps where custom work has a defensible return.
It fits well for stores with complex fulfilment: multiple warehouses, a 3PL, dropship suppliers, or made-to-order lead times. Complexity across systems is exactly what the platform tools handle worst and where operational chasing eats the most time.
It fits badly for a young store with low order volume. At that stage the honest advice is to configure Shopify Flow and Klaviyo properly, pick a helpdesk, and spend the money on acquisition instead. We would rather say that in a scoping call than build you something you do not need yet.
It also fits badly if product and inventory data is inconsistent - variants named differently across channels, SKUs that do not reconcile, stock counts nobody trusts. Automation on top of that produces confidently wrong messages to customers. The first phase there is data cleanup, and we will say so rather than build on it.
Working out whether it pays
This category has the most measurable arithmetic on the site, because everything is already instrumented. Run it on your own figures before talking to anyone.
Start with support. Take your monthly ticket volume, the proportion that are order-status questions, and your fully-loaded cost per ticket including the person's time. That gives you a defensible ceiling on what deflection is worth, and it is usually the clearest number in the whole exercise. Then look at returns: what share are size or fit driven, and what would converting a portion of those to exchanges rather than refunds be worth at your margin.
On the revenue side, count out-of-stock variant views over a month and the demand you never captured, and look at your repeat purchase rate against how often your product would actually need replacing. These are harder to attribute and you should be more sceptical of the numbers, including any a vendor produces for you.
Then subtract honestly. If Shopify Flow, Klaviyo, or your helpdesk's own automation covers a workflow at a price you are already paying, that workflow's value does not belong in a custom build's business case. The residual - the cross-system, cross-channel, and operational work - is the real number, and it is smaller than an enthusiastic agency will imply and usually still worth doing.
Questions owners ask
We already use Shopify Flow and Klaviyo. What is left?
Usually the cross-system work: reconciling the store with a 3PL, suppliers, marketplace channels, or a physical shop's POS; anything on a channel the marketing platform does not own; operational chasing that never belonged to a single tool; and support work that needs data from more than one place. If a workflow you want is covered by a tool you already pay for, we will tell you to switch it on rather than build it.
Will an AI answer our support tickets?
For the mechanical ones, yes - order status, delivery dates, policy questions - and only where the escalation rules say it is safe. Everything ambiguous goes to a person with the full context attached. The number we watch is not deflection rate but reopened tickets, because a system that closes conversations without resolving them looks excellent on a dashboard and costs you customers.
How is this different from what Gorgias already does?
Gorgias is commerce-native and has its own automation and AI resolution features, generally as paid add-ons, and for a lot of stores that is the right answer for ticket handling. Where custom work earns its place is the data those features cannot reach - live carrier state, 3PL and supplier systems, marketplace channels, in-store records - and the proactive and internal workflows that are not ticket-shaped at all.
Can it handle returns end to end?
It can handle the rule-governed part end to end: capturing the request, checking it against your policy window and exclusions, generating a label or RMA where it qualifies, and offering an exchange before a refund on size and fit reasons. Faults, high-value items, out-of-policy requests, and anything with photographs go to a person. It also structures return reasons so you get usable merchandising data, which is often the more valuable half.
Is customer data safe in an automated system?
That depends on the design and it is fair to interrogate. We keep each workflow to the minimum data it needs, document the full data flow so deletion requests can propagate everywhere rather than just to the store, keep every tool that receives customer data on your processor list, and require zero-data-retention terms in writing wherever a language model is involved. Then that documentation goes to your advisor before launch.
Can it take payments?
It can trigger a payment request and act on the outcome, but the payment itself runs through your platform's checkout. The automation layer never sees or stores card details. This is not caution for its own sake: PCI obligations attach to you as the merchant, so a vendor offering to capture card details in their own system is creating your exposure, not theirs.
What is the sensible first build?
Order-status deflection if support volume is your pain, wholesale enquiry routing if it is not. Both are self-contained, both show results within weeks, and neither depends on historical data quality. Post-purchase sequencing and supplier chasing are better as second builds once the plumbing and the data are trusted.
Do we own the system afterwards?
Yes. It runs in accounts belonging to your business, it is documented, and it is handed over, and we will train whoever wants to maintain it. Ask this of every vendor, particularly in a category where a lot of what is sold is access to a hosted layer you cannot inspect or take with you.
Start with a free consult
Tell us how retail & e-commerce enquiries and bookings move through your business today. We will tell you which workflow to build first, roughly what it involves, and whether it is worth doing at all. No cost, and no obligation to hire us.
