The counter question Shopify just made slightly easier
Buy this instead of hiring
A woman walks in on a Saturday. On Thursday night she spent twenty minutes on your website, put two things in a cart, and closed the tab because she was unsure about the size. Your weekend floor staffer knows none of that, so he asks what she is looking for and she says she is just browsing.
Helping her looks like one action. It is four questions. Who is this — clerical, and the hardest to get. What did she already want — a records question your systems can answer. Why did she stop — judgment. What do you do about it — judgment again.
As of September 3, 2026, Shopify POS answers the second one: authorized staff can look up an identified customer and see her abandoned online checkout. It ships in POS v11.14 behind a per-role permission. Only the cart, not her browsing. The only sources are Shopify's changelog and the help page it links to; nobody independent has reported it yet.
The conditions are the real story. Four things must be true before a cart appears: the customer is identified, that customer has an email address, there is an eligible abandoned checkout, and the register session has location context. The last two your systems hand you. The first two you have to earn, because this feature does not identify anybody — it rewards you for having done it already, which is a habit performed by a nineteen-year-old on a Saturday, not a setting.
So before turning anything on, count what share of last month's in-store sales carried a customer record with an email. Under a third and this fires almost never, and the work is at the counter, not in the settings.
Close every month and the January reconstruction disappears
Reorganize and the job disappears
Every tax season starts the same way: a year of loose books lands on somebody's desk, and weeks go into reconstructing March before the real work starts. Most firms call that a client-discipline problem. In CPA Practice Advisor, September 2, 2026, Steph Conley — who runs Steph's Books, an outsourced bookkeeping firm, and so sells the service she recommends — argues in "Split the Year With Your Bookkeeper So Tax Season Isn't a Cleanup" that it is a calendar problem, a different thing entirely, because calendars can change.
Her before is the single handoff. One deadline matters, so every error compounds for ten months — a miscategorized owner draw in February distorts the summer P&L, produces a wrong estimated payment in September, and arrives in January as a reconstruction by the most expensive person at the most expensive time of year. She says cleanup billed at tax-season rates routinely runs 30 to 50% above the same corrections handled in-period.
The after is four checkpoints, each ending in a deliverable, not a status update. The second is worth stealing: within thirty days of filing, every adjusting entry the preparer made is a process defect, not just an entry. Reclassified $18,000 of miscellaneous expense? Fix the chart of accounts. Then January's close is a formality, because every month behind it already closed.
Note what did not happen, though she never says so. Nobody bought software; nobody got faster. The reconstruction stopped existing.
You are not a CPA firm, but you run one of these: job closeout at year end instead of at completion, an annual inventory count instead of cycle counts, W-9s chased in January rather than collected before the first check goes out. Same shape, same fix.
Nobody reads what the customer sends back
Nothing off the shelf fits
You won a customer in March; the first job is on April 2nd. Before anyone can go you need a signed contract, a certificate of insurance naming you correctly, a W-9, a credit application, a tax exemption certificate if they claim one, and a PO number their system will accept. Your office coordinator emails a list and waits.
What comes back is the problem. The COI is last year's. The signature page is there but page four is not initialed. The W-9 carries a trading name, not the entity that will appear on the invoice — which nobody notices until a payment is rejected in July. She emails again, and the crew goes out on April 2nd anyway because the customer is waiting.
Collecting a document is three different questions
Did it arrive? Clerical, and any request-list portal answers it.
Is it the right document? That takes reading — somebody opening the file to see it is a certificate of insurance, not a policy declaration page, and current.
Does it say what it has to say? That is the real judgment, and nothing off the shelf touches it. Does the COI name your entity as additional insured, or merely list the customer's coverage? Are the limits the ones your contract requires? Does the W-9 match the remit-to on the invoice you are about to send?
Your portal knows a file was uploaded. It has no opinion about whether that file answers the question. That is why the second and third emails exist, and those emails are the cost of this job.
The nearest thing anyone shipped
The audit profession has this job under another name. As of September 1, 2026, AuditFile's Agentic AI Orchestrator Suite — available now, the announcement says, to customers on its Pro Plus! and Enterprise plans — runs agents that generate and review the request list, with partners and managers approving, rejecting or redirecting through a sign-off queue and every action captured in an evidence-linked trail.
That answers who owns the list and who signs off. It does not answer this job: generating a request list is not the same as the client sending the documents, and nothing described opens what came back and judges it.
What would have to be built
Something that opens the file. It takes each arriving document, works out which item it answers, and checks it against what that item requires: current certificate or last year's, your entity named as additional insured, limits at or above the contract, signature page signed rather than merely present, W-9 entity matching the name you will invoice. When the answer is no it goes back naming the defect — not "your document was rejected" but "this certificate expires on the 14th and we need one through the end of the contract." Only real ambiguity reaches a person.
Nobody sells this because the standard is yours. What counts as complete was written by your own past disputes: the customer who withheld payment over an entity-name mismatch, the sub whose coverage lapsed mid-job. A template protects you from the average business's problems, not yours.
The honest no, and it is a real one
Write the list before building anything. One page, per document, saying what makes it acceptable — not "COI required" but the entity name, the limits, the additional-insured wording, the dates. That page is worth more than the automation, and if you cannot write it, nothing can be built: you have just discovered the rules do not exist yet. Most businesses should stop here.
Below a handful of new customers a year there is no case: no volume, and no corpus of past submissions to calibrate against. And where the question is whether a clause actually protects you, that is a lawyer reading a contract.
Who will resist
Your office coordinator, and she has the better argument. The chasing is a relationship she is managing — she knows which customers answer a firm email and which need a call from someone they recognize. An automated rejection landing on a new customer's bookkeeper in week one can cost more than the missing certificate.
She will not object. She will quietly exempt her important accounts and keep handling those herself, leaving the system the easy ones. The dashboard will look excellent, and the accounts where a missing certificate would actually hurt you are the ones it never touched.
Sales will push the other way, wanting crews on site while the paperwork catches up. They will win often enough that whatever you build has to record who waived what, and when.
If one of these is your Thursday afternoon, the ten agent roles are in the Library.