// Field notes
Massmarket field notes
Short answers to the questions owners, controllers and operations leads ask before they test AI on one job, using documents their team already finished: what to send, what to keep out, how to choose the sample and who should judge it. Each note names the public sources it leans on.
// Notes
Questions, answered
What can we safely send for an AI test, and what should stay out?
Send copies of finished work after an NDA, with only what the job reads. Card numbers, health records, logins and identifiers the job doesn't need stay out. How to redact properly, and when to use look-alikes.
Read the noteHow do I test PO automation on our own old orders before buying a tool?
Pick 20 to 30 customer POs your team already entered, pair each with the order they keyed, and score the drafts field by field. How to choose the sample, who judges, and how to write success criteria.
Read the noteWhat should an agency test first when it tries AI on certificates?
Issuing certificates, checking contract requirements against the policy, or tracking the certificates your insureds receive: which one an agency should test first, why, and how to pick and score 20 to 30 past requests.
Read the noteHow can a property manager test AI on vendor invoices without exposing resident data?
How a property or HOA manager tests AI on vendor invoices while keeping resident data out: the five fields the coding job reads, where tenant details hide on invoices, what to cut or replace, when to build look-alikes, and who checks the set.
Read the noteShould a distributor buy an order-entry tool, or test on old POs first?
What a test on your own entered POs shows that a demo or a vendor's accuracy figure can't, the questions to ask any order entry vendor, and when buying straight away is reasonable.
Read the noteWhich construction AP job should we test AI on first?
Invoice coding, lien waivers or pay applications: what each test needs, who reviews it and how its mistakes come to light, for commercial specialty contractors choosing a first AI proof of concept.
Read the noteShould we start with an AI readiness assessment or test AI on one job?
An AI readiness assessment scores the whole company and ends in a roadmap. A proof of concept on one job scores drafts against work your team already finished. What each produces, when each should go first, and how to run them in either order.
Read the noteWhat is the difference between an AI proof of concept and a pilot, and which comes first?
A proof of concept tests AI on copies of past work. A pilot runs it on live work for a limited time, with a person checking every item. What each stage answers, who takes part, what data it touches, and the exit criteria between proof of concept, pilot and production.
Read the noteDoes testing AI on our documents mean connecting it to our ERP?
Testing AI on copies and exports needs no connection to your ERP. What that test can prove, what only a connected live run can, and the questions IT asks first: access scope, service accounts, read versus write and audit logging.
Read the noteWhat happens after an AI proof of concept says yes?
After a yes, the same job runs on new work with a person approving every draft. Who owns that month, what to log each day, what pauses or stops it, how your reviewer's day changes, and when to widen the scope.
Read the noteWhich back-office jobs are a bad first test for AI?
Five signs a back-office job is the wrong first test for AI, the jobs that shouldn't be tested at all, and how to find the job next door that will tell you something.
Read the noteWe already take EDI orders. Where does AI fit?
EDI covers the partners who send X12 orders. How to size the email, PDF, portal and spreadsheet orders left over from your own order history, and why your EDI rules still apply to AI drafts.
Read the noteOur documents are scanned, faxed or handwritten. Can we still test AI on them?
Faxes, poor scans, stamps and handwriting belong in a first AI test. What each does to extraction, how to pick a sample with the hard pages in it, what your reviewer should score, and what to leave for later.
Read the noteDoes putting AI in accounts payable weaken our internal controls?
A test on copies of posted invoices touches no control. A live run has to fit the ones you have: segregation of duties, a person who approves, evidence of each review, and change control over prompts and models. What auditors will ask.
Read the noteHow do we tell the team whose work we're testing AI on?
Who to tell before AI is tested on your team's own work, and when. What to say about why, what the reviewer decides, how finished work becomes the answer key, what a yes changes, and how to collect objections.
Read the noteCan AI sort our shared orders@ or ap@ inbox before anyone reads it?
What sorting a shared inbox means: document type, customer or vendor, urgency, repeats, forwards and attachments. How to test it on one exported month scored against its real routing, what stays out, and why bank changes go to a person.
Read the noteWhat should we ask any AI vendor before we send it documents?
Seven questions for any AI vendor before a file leaves your hands: where it is processed and kept, who can open it, training, subprocessors, deletion, incident notice, and what a SOC 2 report or ISO 27001 certificate covers. How to get the answers in writing.
Read the noteWhich month-end close task should a controller test AI on first?
Accruals, bank and card reconciliations, intercompany, journal entry support or flux explanations: which close task has finished examples and a clear answer key, who should review the drafts, and what stays out of a first test.
Read the noteWhat happens when the AI test says no?
When your reviewer's call is no, the scoring sheet says why. How to tell a no about the job from a no about the documents or the success criteria, what you keep, when a second test makes sense, and why a no found on copies costs least.
Read the noteDo we need to clean up our data before we try AI?
No full cleanup first. A document job reads a few master files: the item master, customer part number cross-references, the vendor master, the chart of accounts or cost codes. A test on old documents shows which records fall short.
Read the noteWho on our team should judge the AI test?
Why the person who does the work today should judge an AI test, not a manager or IT: what they need, when a second reviewer helps, how blind scoring keeps the marks honest, and what the owner hears back.
Read the noteSome of our orders and invoices come in Spanish or French. Can AI handle them?
Orders and invoices from Mexico, Quebec and Latin American suppliers: what has to come out in English, what must stay as written, how to test AI on old documents in each language, and what stays with a person. Not certified translation.
Read the noteWe don't have an IT department. Can we still use AI on our back office?
An owner-led company can test AI on its back office without an IT department. What you provide, what a test never needs, who owns the job after a yes, what to ask about support, and when your IT provider or ERP reseller should be involved.
Read the noteWhich 20 to 30 documents should we pull for the AI test?
Pull a recent stretch of ordinary work, not the cleanest examples: the awkward weekly cases, your main customers, vendors and layouts, and the finished work as the answer key. What to leave out, how look-alikes stand in, and why the list is fixed before any draft.
Read the noteWhat does an AI proof of concept cost us?
The proof of concept itself is free. What your company spends is your reviewer's time and the effort of pulling 20 to 30 old documents. Why that costs less than a pilot on live work, when you see a price, and what to ask any provider a test leaves out.
Read the noteWhat happens to the documents the AI can't handle?
Every document job has exceptions: a new customer layout, a handwritten change, a missing PO number, a price that doesn't match. How a well-built job holds them for a person instead of guessing, and how a test finds them in your own old documents.
Read the noteWill the AI make up numbers on our orders or invoices?
An AI drafting from orders and invoices can fill a field with a value that isn't on the page. What invented prices, quantities, part numbers and dates look like, why a draft should show its sources, how a test catches them, and what to ask about blanks.
Read the noteCan we send our customers' documents to an outside AI test?
Before a customer's PO, drawing or spec goes to an outside AI test: the confidentiality terms you agreed to, trade secret and export rules, personal information, where an NDA, redaction and look-alikes fit, and who decides. Not legal advice.
Read the noteCan AI read the spreadsheets our team built?
Price lists, takeoff sheets, rate tables and trackers hide meaning in merged cells, colour, hidden tabs and links to other files. How to test AI on copies, with your team's finished work as the answer key, and what stays with a person.
Read the noteOur orders come in through portals, faxes and email at once. Can one AI test cover all three?
The same order looks different by email, fax and customer portal. Why portal downloads and fax images are copies like any other, what to bring from each channel, how one order arriving twice by different routes gets caught, and what a person still checks.
Read the note// How these are written
The answer first, then the sources
Every note opens with the short answer: two or three sentences you could act on today. The rest explains it. Where a note relies on an outside fact, such as a federal guide, an industry standard or a statute, the text names where it comes from so you can check it yourself.
The notes describe how a test would run and what we would ask you for. They report no results, because a test only proves something on the documents it ran on. None of it is legal advice: when a file falls under a law or a customer contract, your counsel makes the call.
Rob Eason, the founder, signs each note, and he is the one you talk to when you book a call.
Rob Eason, founder
// Book a proof-of-concept call
Bring the question these notes haven't answered yet.
Thirty minutes with Rob. Come with your question and the one job you'd test first. Hold every file until an NDA is in place.
- Co-owned a 60-employee financial services firm and sold his share to his partner.
- Ran GOSO Technologies, a Boston solar company rated 5/5 across 24 Google reviews.
- 5/5 across 83 reviews in HighLevel's Certified Admin Directory.
- 4.3/5 from 6 Google reviews of Massmarket AI.
- Every directory and Google review, critical ones included, is shown word for word at massmarket.ai/reviews.
- Takes every proof-of-concept call himself.