Skip to content

Private Demo Mode

The 2026 E-commerce operations Playbook

The short version

Most of e-commerce operations is follow-up, and follow-up is the work agents do best. An order ships late, a customer asks where it is, a return needs a label, a SKU runs low, a supplier misses a date. Each one is small. Together they eat the week of every ops lead we spoke to for this playbook.

You do not need a new platform to change this. You need a clear line between the work that follows a rule and the work that needs a relationship. Give the first kind to agents. Keep the second kind with people, and give those people more time for it. The five changes below are the ones every contributor made.

  1. 01Answer every “where is my order” question from live carrier data, before the customer asks twice.
  2. 02Run returns and exchanges from a written policy that an agent applies, with a person on each exception.
  3. 03Turn inventory alerts into decisions: a reorder draft, a paused ad, or a back-in-stock list.
  4. 04Let an agent chase suppliers and wholesale buyers on a fixed cadence, in the channel each one answers.
  5. 05Keep people on VIP customers, supplier negotiations, and every call during peak season.

Why this changed

Brands used to scale operations with headcount, and every new channel added a person. A brand at 8,000 orders a month had one ops generalist. At 40,000 it had a support team of six, a returns coordinator, an inventory planner, and someone whose whole job was email to the 3PL. The work did not get harder. It got wider.

Agents change the math because almost every task in this list starts with a lookup in a system that already has an API. The order is in Shopify. The tracking events are with the carrier. The stock level is in the warehouse system. The customer history is in the helpdesk. A person used to open four tabs to answer one question. An agent reads all four in two seconds and writes the answer in your brand voice.

“We did not cut the team. We stopped hiring for tab-switching. The four people we have now spend their days on wholesale accounts and on the customers who are about to leave us.”

The companies below build the systems that most of our contributors run on. Each one now ships its own agent features, and each one exposes the data that an outside agent needs to do the follow-up work across all of them.

  • ShopifyCommerce platform for online stores and retail
  • KlaviyoEmail, SMS, and customer data for consumer brands
  • GorgiasHelpdesk and AI support for e-commerce brands

Where is my order

The single largest category of support volume, at every brand we spoke to, is some form of “where is my order.” At Lumen & Loam it was 46% of all tickets in the fourth quarter. At Saltmarsh Skin it was 38% across the year. These questions have one correct answer, and that answer is already in a carrier feed.

The first move is to answer the question when the customer asks it, from the live tracking event, not from the order status. “Shipped” is not an answer. “Your box left the Reno warehouse Tuesday night and reached the Denver hub this morning, and UPS expects it Thursday” is an answer. The second move is to send that message before the customer asks, the moment a shipment misses its expected scan.

Set a hard limit on what the agent can promise. At Saltmarsh Skin, the agent can resend an order under $60 when the carrier marks it lost, and it can offer free shipping on the next order. It cannot promise a delivery date the carrier has not given, and it cannot refund an order that is still moving. Those two rules removed almost every complaint about the agent in the first month.

“The customer does not want a tracking link. They want to know if the gift arrives before the birthday. Answer that question and the ticket closes.”

Returns and exchanges

Returns cost money in three places: the refund, the shipping, and the time it takes to decide. The third one is where agents help most. A return that waits nine days for a person to approve it becomes a chargeback, a one-star review, or both. A return that an agent approves in nine minutes, from a written policy, becomes an exchange about half the time.

Write the policy as rules an agent can apply, not as copy for the help center. Our contributors use the same structure, in this order:

  1. 01Check the window: days since delivery, not since the order date.
  2. 02Check the item: final sale, opened, personalized, or hazardous goods cannot come back.
  3. 03Check the customer: more than three returns in 90 days goes to a person.
  4. 04Offer the exchange first, with the size or color in stock, and a bonus credit of 10% if they take it.
  5. 05Issue the label, and refund only on the first carrier scan, not at the warehouse.
  6. 06Send every case that fails a check to a person, with the reason written at the top.

At Fernhollow Goods this moved 82% of returns out of the queue. Exchanges rose from 19% of returns to 44%, which kept about $310,000 of revenue in the first year. The other 18% are the cases that need judgment: a jacket that failed on a trip, a customer who bought the wrong tent for a wedding, a box that arrived wet.

“The agent is strict and fast. The person is generous and slow. You want both, and you want the customer to meet the person only when generosity is the point.”

Inventory alerts that act

An inventory alert that only sends a message is noise. Every brand we spoke to had a Slack channel full of “low stock” warnings that nobody read after the first month. The change is to make each alert end in a draft decision, so a person only says yes or no.

When a SKU crosses its reorder point, the agent should calculate the days of cover from the last 28 days of sales, check the open purchase orders and the supplier lead time, and then write one of four things. Kofi Asante’s team at Tidewell Coffee uses this short list:

  • A reorder draft, sized to the lead time plus a two-week buffer, sent to the buyer for approval.
  • A request to pause or cut paid ads on that product until stock recovers.
  • A back-in-stock list in Klaviyo, so the people who wanted it hear first.
  • A transfer between warehouses, when one location has more than 60 days of cover and the other has less than 10.
“Green coffee has a twelve-week lead time. An alert that says “low” on the day we run out is a eulogy. Our agent tells us eleven weeks early, with the order already written.”

Tidewell cut stockouts on its top 40 SKUs from 23 days a year to 4. The agent never places a purchase order by itself. That rule is the same at every contributor: a person approves every commitment of cash.

Wholesale and suppliers

Wholesale orders arrive in the worst formats a business can receive: a PDF purchase order, a photo of a handwritten form, an email that says “same as last time but double the blue.” Before agents, someone typed each one into the system by hand. Brightfold processes about 260 wholesale orders a month from 190 stores, and typing them took one full-time person.

Now an agent reads each purchase order, matches the store to its account, maps the buyer’s product names to SKUs, checks the price against the store’s tier, and creates a draft order. Anything that does not match, such as an old price, an unknown SKU, or a ship date the warehouse cannot meet, goes to Hana Petrosyan with a one-line reason.

“I used to spend Mondays typing. Now I spend them on the phone with the 20 stores that order the most, and those 20 grew 34% last year.”

Supplier follow-up works the same way. The agent keeps a list of every open purchase order with its promised ship date. Five days before that date it asks the supplier to confirm. If the supplier does not answer in two business days, it asks again, in the same thread, in plain words. If the date moves, it updates the inbound plan and tells the buyer. Our contributors chase between 30 and 120 open orders at any time, and the agent does all of it.

The tone matters more than the cadence. A supplier who gets a polite, specific question (“Can you confirm the 1,200 oak drawer fronts still leave on May 14?”) answers it. A supplier who gets a vague reminder ignores it. Brightfold’s on-time supplier confirmation rate went from 58% to 91% after the agent started to ask one exact question per message.

  1. 01Keep one record for each open purchase order: supplier, items, promised date, last reply.
  2. 02Ask for confirmation five days before the promised date.
  3. 03Follow up after two business days of silence, in the same thread.
  4. 04Update the inbound plan and alert the buyer when a date moves more than three days.
  5. 05Hand the thread to a person after the second missed date.

Where people stay

Every contributor drew the same line, and they drew it around relationships. An agent can apply a policy, read a carrier feed, and chase a late reply. It cannot rebuild trust with a customer who spent $4,000 with you and feels ignored, and it cannot negotiate a price with a factory owner you have known for six years.

  • VIP customers: anyone in the top 2% of lifetime spend gets a named person, and the agent only drafts.
  • Supplier relationships: price, terms, quality problems, and any missed date after the second one.
  • Peak season calls: from Black Friday to the end of December, every phone call goes to a person.
  • Damage and safety: a broken product that could hurt someone is never a policy decision.
  • Public complaints: a review or a social post that names the brand gets a reply from a person.

Peak season deserves its own rule. Volume at most of our contributors rises three to five times between mid-November and Christmas. That is when the agent does the most work, and it is also when a mistake costs the most. Saltmarsh Skin keeps a person on every call and every refund over $150 during that window, and it doubles its weekly review of the agent’s closed tickets.

“In December the agent answers 4,000 tickets a week and we answer 300. Those 300 are the ones people remember, so those are the ones we take.”

The tools, ranked

We asked each contributor to score the operations tools they have run in the last two years. Scores reward two things: complete data that an agent can read and write through an API, and a clear record of every change an agent makes. A tool that hides its data behind a dashboard scored low, however good its features are.

E-commerce operations tools, rankedMean of five contributor scores out of 10
  1. 01Shopify · Best for the order and customer recordIts API reaches orders, fulfillments, inventory, and customers, and every other tool reads from it. Start here.9.1
  2. 02Gorgias · Best for support for shopify brandsOrder data sits next to the ticket, and its rules and macros give an agent a clean way to act.8.6
  3. 03Loop Returns · Best for returns and exchangesThe exchange-first flow paid for itself in a quarter. Policy rules are easy for an agent to read.8.2
  4. 04ShipBob · Best for outsourced fulfillmentGood inventory and shipment data across warehouses. Exceptions still need an email to a person.7.7
  5. 05Airtable · Best for purchase orders and wholesale accountsThe fastest place to keep an open-order list that agents and people both edit. It needs care past 50,000 rows.7.4
  6. 06Stripe · Best for payments outside the storefrontClean refund and dispute data for wholesale invoices and subscriptions, and a full record of every refund.7.1

So what do you do?

Start with the question your customers ask most, which is almost certainly “where is my order.” It has one correct answer, the data is already available, and the result is easy to measure. Most contributors had their first agent answering tracking questions in under two weeks, and they used that win to earn the trust for the rest.

  1. 01Count last quarter’s tickets by type, and find the share that is about order status.
  2. 02Connect an agent to your store, your helpdesk, and your carrier data, and let it draft replies for one week.
  3. 03Review 50 drafts a day, fix the policy where the agent was wrong, then let it send.
  4. 04Write your return policy as numbered rules, and move returns next.
  5. 05Add inventory alerts that end in a draft decision, and supplier follow-ups on a fixed cadence.
  6. 06Write down who owns VIP customers, supplier talks, and peak season calls, and protect their time.

Measure three numbers each week: the share of questions closed with no person, the second-contact rate, and the hours your team spends on relationships. The first two tell you if the agent works. The third tells you if the change was worth it.

“The goal was never fewer people. The goal was a team that knows every café owner by name and never has to look up a tracking number again.”