AI for ecommerce
Patterns, no client yetDirect-to-consumer brands and online retailers where the support queue grows faster than the team.
Ecommerce operations rarely break at the storefront. They break in the queue behind it: the same forty questions answered by hand, the refund nobody decided on, and the churn signal that was visible three weeks before the cancellation. Those are the patterns worth looking at first.
No client yet
We will not pretend otherwise
Patterns
What we would look at first
Start here
A workshop, not a pitch
Where the time goes
We have not built a system for an ecommerce brand yet, so treat this as a considered read on where the work usually waits rather than as a claim about your operation. These are the patterns we would go looking for on day one.
Where is my order, can I change the size, does this ship to Canada, what is your return window. A support person answers each one as if it were new, because nothing in the stack knows that it is the fortieth time today and that the answer never changes.
The same answers, typed again
There is a policy. There is also what you actually do for a good customer with a damaged item and a reasonable story. That gap lives in a senior person's judgment, so every borderline case escalates, and the customer waits two days for a decision the business already knows how to make.
Borderline cases wait on one person
The subscriber skipped twice, opened nothing for a month, and had one support ticket that went sideways. Each of those lives in a different tool. Together they are a cancellation you could have seen three weeks out, and instead you saw it on the day.
The signals are visible and scattered
The promotion worked, so the support queue tripled on the same day the orders did. The team is drowning during the exact window when the customer experience decides whether the acquisition cost was worth paying.
Support spikes when it matters most
Descriptions, attributes, and the details that decide whether a product is findable. It is real work, it is repetitive, and it is always the thing that slips, so the catalog quietly gets worse as it gets bigger.
The catalog decays as it grows
A helpdesk, a subscription tool, an email platform, a reviews app, and a spreadsheet holding them together. Each one is fine. Nothing reads across them, so the person who knows what is happening is the one with all five tabs open.
Five tools, one person holding them together
The machine
The operational surfaces behind a direct-to-consumer brand. The storefront is rarely where the drag is.
01
02
03
04
The split
The rule that matters most in this category: money never moves without a person. A support system that can issue refunds on its own has a bad day exactly once, and it is expensive and public. The system drafts the decision, cites your policy, and waits. Approving a queue of pre-decided refunds takes minutes. Undoing an autonomous one takes a week and a customer.
What we build
01
The queue sorted before a person opens it: what this is, how urgent, what the order context is, and whether it is one of the forty questions that never change. The volume that never needed a human stops reaching one.
02
The reply written with the order, the history, and the policy already in front of it, for a person to read and send. Not a bot talking to your customer. A support person going four times faster with the research already done.
03
The recommendation, the policy line it came from, and the reason, queued for approval. Your actual decision logic written down once instead of living in one senior person's head, with the money still moving only when a human says so.
04
The skip, the silence, the sideways ticket, read together instead of separately. The output is a list of subscribers worth a human touch this week, with the reason attached.
05
Descriptions and attributes drafted from what the product is, in your brand's voice, for a person to approve. Repetitive, real, and always the work that slips.
06
Your policy, your voice, your product knowledge, and what you actually do for a good customer, written down in your own cloud. Every system above reads from it. Without it they produce generic ecommerce support, which your customers have already met.
Under the hood
Storefront
Support
Subscriptions
Lifecycle messaging
This is the software this category runs on, and we would build into it rather than replacing any of it. To be straight with you: we have shipped systems into contractor and insurance stacks, not into this one. The pattern of reading a system of record and writing back to it is the same, and the roadmap is where that gets scoped honestly rather than assumed.
Principles
So this page does not have stat chips pretending to be results. Every number on this site belongs to a real system we built and can point at. There is no ecommerce one yet. When there is, it will be on the work page with a client name on it, the way the others are.
Refunds, credits, and anything that reaches a customer's card. The system drafts and recommends, a person approves. This is the one place in an ecommerce build where autonomy is genuinely tempting and genuinely wrong.
The forty repeat questions are not a support problem to be answered faster. They are your product, your shipping page, or your sizing chart telling you something. A system that only makes the answering cheaper has hidden the report. It should also be reading it back to you.
A support reply that sounds like generic ecommerce support is worse than a slow one from a person, because your customers can tell and it costs you the brand you paid to build. That is what the company brain is for, and it is why we build it first.
Same approach, different language
Professional services
Agencies, consultancies, and expert firms where the product is judgment and the constraint is the founder's calendar.
We speak the language →Home services
Roofing, siding, gutters, windows, and the exterior trades that sell from an inspection instead of a price list.
Live system →Financial services
Advisory firms, lenders, and finance teams where senior people spend their day gathering what they need to think about.
Patterns, no client yet →Questions
No. We build for contractors, roofing and exterior companies, and insurance brokers, and those pages have real systems and real numbers on them. This page is a considered read on where the work usually waits in a direct-to-consumer operation, and nothing more than that. If you want proof before you talk to us, read the construction page, because that is where the receipts are.
It can handle the part of it that was never really a conversation. Where is my order, what is the return window, does this ship there: those are lookups with a friendly sentence around them, and they are usually most of the volume. What it should not do is talk to an upset customer or decide a refund on its own. The useful shape is triage plus a drafted reply, with a person sending it.
It can decide them and it should not execute them. The system reads your policy, reads the case, recommends the call, and cites the line it came from. A person approves. Approving a queue of pre-decided refunds takes a few minutes a day. An autonomous refund system has one bad day and it is expensive, public, and hard to undo.
That depends entirely on whether it has a company brain, which is a written record of your voice, your policy, your product knowledge, and what you actually do for a good customer in an edge case. Without that, a model produces the average of every ecommerce support reply ever written, and your customers can tell instantly. That is why it is the first thing we build rather than the last.
Usually two things. The first is that the built-in features answer from a help center article and stop at the edge of it. The second is that they cannot read across your subscription tool, your messaging platform, and your order history at the same time, which is where the useful answers actually live. If your helpdesk AI is working for you, that is a good outcome and you may not need us.
Honestly, with a workshop, because we would be learning your category while you are learning what AI can do in it. One session mapping one real workflow end to end, and you leave with the map and a plan whether or not you build anything with us. In a category where we have no case study, paying us to guess would be the wrong trade for you.
You do, from day one. The code sits in your repository, the knowledge base with your policy and your voice sits in your cloud, and the accounts are in your name. Nothing we build stops working because you stopped working with us.
Read next
Explainer
What an AI-legible business actually looks like, the five-point readiness checklist to score your own operation, why one person is usually compensating for every broken handoff, and the 90-day path from messy to legible.
Strategy
Why businesses should start thinking in workflows instead of tools when evaluating AI consulting, automation, and operations redesign.
Operations
How to reduce admin overhead without adding staff by redesigning intake, follow-up, and reporting workflows into systems that run without constant manual attention.
Where to start
Find the bottleneck, price the fix, build the system, then keep compounding it. The four engagements are how you buy it, and you can start on any rung.
See the four engagements →Get started
Book a 30 minute call. We find the one place your work waits on you, and you get a straight answer about whether AI is the fix. No pitch, no deck.