Managed AI for Columbus Credit Unions and Community Banks
It is 8:15 on a Tuesday. Four adverse action notices need the reason written in words a member will understand, and the loan officer who made those calls is in appointments until eleven. Down the hall the compliance officer is on version three of a policy that goes to the board next month. Marketing is on the third rewrite of the same paragraph about the fee change, because the website, the lobby sign, the statement insert and the contact center script all have to say the same thing. ScaleSight AI runs one managed AI workspace for a Central Ohio credit union or community bank, pointed at the drafting and the summarizing. It does not score, price or decide anything.
Where it starts
The recurring work worth handing over first
Not the impressive demo. The jobs that come up every week and eat somebody’s afternoon, which is where the hours actually come back.
Adverse action notices in your own approved wording
The lender makes the decision and states the reason. A workflow takes that reason, your own notice format and your own approved language, and drafts the member-facing explanation so the letter reads the same whether it came out of a branch or off the indirect desk. It does not generate reasons and it is not built to, because the reason on the notice has to be the one the decision actually rested on and a drafted sentence is not a source of reasons. The draft lands with whoever owns notice review before it reaches the member, and that step is written into the template rather than left to habit. Nothing sends itself.
Policy drafts and the annual review cycle
Policies come around every year and most of the work is reconciling one long document against the handful of things that actually changed. A workflow drafts the update from the current version and the change you describe, then marks every line that moved so the reviewer can see exactly what is different. The person who owns the policy still reads it and signs it, and what reaches the board is what that person signed.
A rate or fee change said the same way everywhere
A change to the fee schedule on share draft accounts has to reach members in the notice your own rules require, and then it has to say the same thing on the website, on the lobby sign, in the statement insert and in the contact center script. The workflow produces that set from one approved description of the change, so the versions do not drift apart in four different hands. What has to go out and when is your compliance officer’s call, and compliance reads all of it before anything posts.
Collections and member solutions correspondence
The first letter after a missed payment, the follow-up two weeks later, the workout confirmation and the internal note for the file are four registers of the same conversation. Working from the account notes and the last exchange, the workspace drafts each one in your own wording. The review step is written into the template and the person handling the account edits and sends. Nothing sends itself, and nothing goes out that commits the institution to something it did not agree to.
Vendor documents read once, into your own format
A vendor agreement, a service description, a security summary and a set of answers to your questionnaire arrive as four PDFs in somebody’s inbox. The workflow summarizes each one with page references and pulls dates, terms, renewal windows, termination triggers and subcontracting language into a list, quoting the sentence each item came from. The risk rating and the decision to sign stay with the person who owns third-party risk. What comes back is the reading, in one place and in your own format.
Board packets and the meeting record
A recording or a set of rough notes becomes minutes in your format, with motions and votes separated from discussion. A second pass produces the one-page summary of what changed since the last meeting, and the same mechanism builds the pack the supervisory committee works from around the annual audit, or the audit committee at a bank. The CEO reviews before anything is circulated.
People here already opened their own AI accounts. What is wrong with that?
Every one of those accounts is a relationship between one person and an outside company, opened on a personal card, with no agreement your institution holds and no record it can reach. When your vendor file needs something on that relationship, there is nothing to put in it. Nobody signed on the institution’s behalf, nobody set a retention term, and nobody can say what left the building. It is a third-party relationship either way, opened one card at a time and managed by nobody. The only question is whether anything sits behind it.
A managed service produces the other version. One agreement, one workspace, one place where access is granted and removed, and a set of documents that can go into your file. We are careful about how those documents are worded. ScaleSight AI runs on a SOC 2 Type II audited platform, and the audit belongs to the platform vendor rather than to ScaleSight AI as a legal entity. We write it that way every time, including inside a vendor file, because an overstatement there is the expensive kind of error. Tell us what your file needs and we will tell you what can be provided and whose name is on each piece of it.
The second half of the answer is the part every industry gets. A workflow built straight onto one AI company’s model carries that company’s pricing and its release schedule with it, so a repricing or a retired version turns into rebuild work at your expense. Work inside a ScaleSight AI workspace sits above the models. When one changes, we re-point the work and it keeps running, and the person who opens the same template the next morning finds nothing different.
Routing is the rest of it. No single AI company leads at every task. One reads a long vendor agreement better than the others. One does live web research. One drafts bulk text cheaply, which matters when a fee change means redoing a stack of member-facing copy. Each job goes to whichever model handles it best, and the person using it never has to know which one answered. Accounts bought one at a time give you none of that, and no spend ceiling, no budget by department, and no report of what the institution is using AI for.
Our BSA team works suspicious activity cases. What has to stay out of this?
Suspicious activity reports, and anything that would reveal one exists, are out of scope for a ScaleSight AI workspace. That is our own scoping rule about what we will accept, set before the first seat is turned on. We are not explaining the rule behind it to you and we would be the least useful voice in the room on it. We are telling you where our line sits, because a vendor that will not say where its line sits has not thought about it.
Now the honest part. What holds that line is the scoping boundary, the configuration built around it, and the people who follow it. It is not a filter that reads content and rejects it. No such gate exists here, and we are not going to imply one. If somebody pastes case material into a workspace anyway, the scope did not stop them, and a vendor promising you a control that inspects and rejects is selling you something it does not have.
The practical version is a scoping session. Which system holds the alert queue. Which mailboxes carry internal referrals. Which people work cases and which never do. Whether that department takes seats at all, because the cleanest answer is often that it does not. The line gets written down, the workspace is configured around it, and the people who sit near it are told in plain words what may go in and what may not.
Confidential supervisory information sits next to that and needs its own answer, which is not ours to give. Documents tied to your exam relationship and your correspondence with your regulator are their own category, and whether any of that may reach a service provider under contract is a question for your regulator rather than for us. So ask before any of it goes in, rather than take a vendor’s word for it. Our sibling company TTS Cyber works from the same Columbus address on the managed IT and cybersecurity side. If your scoping question turns out to be a compliance question, we will say so and walk it across the hall rather than answer it here.
Most of our people are in a branch or on the phones. What are we actually paying for?
Seats are per person, and you buy them only for the people who write, summarize or explain for a living. In a credit union that is lending, including business lending, plus marketing, HR, accounting, compliance, collections and the executive team. An MSR spends the day inside the core opening accounts and posting transactions, which is a screen that produces almost no drafting, so a seat there would sit unused. Frontline and part-time branch roles mostly do not need one, and the few that do are settled in scoping rather than assumed. A community bank draws the same line in different words: the seat follows the drafting, not the desk.
Billing is per seat plus usage. The allowance rides on the seat and renews monthly, and the ceiling on total spend is a number you set, so the invoice cannot run away from you between months. Nothing here is flat and nothing is unlimited. This is a managed service, not a software subscription.
Budget is allocated by team and department, never person by person. Lending carries an allowance, marketing carries one, compliance carries one. If collections runs through its allowance in week three because a letter workflow took hold, you move budget between departments. For a not-for-profit, member-owned cooperative that answers to its members for how it spends, that shape matters in a specific way: the board can see a number by department without any report on an individual.
The published ranges are monthly totals for the whole institution. Up to 10 seats runs $250 to $500 a month. Ten to 25 seats runs $500 to $1,175. Twenty-five to 100 seats runs $1,175 to $3,800. A $250 monthly minimum applies, and we do not publish a per-seat rate.
What shows up in the workspace that we did not build ourselves?
Work that is the same in every organization gets built once and given to every client. A vendor follow-up is a vendor follow-up whether the vendor prints statements or paves the lot. A recording turned into decisions and dated actions is the same job everywhere. That is the library, and an institution that starts in March inherits everything built before March, so the workspace is not an empty screen on day one.
Anything built on your own documents, your own data or your specific process stays in your workspace and is never shared. Your policy set, your approved notice language, your house style, your product descriptions, the collections letters your team has sharpened over years. Those never enter the library and never appear in another client’s workspace, including the credit union three exits down that competes for the same members you do.
New templates, agents and workflows ship into the workspace every month out of that shared library. The library grows as more general work is built into it, so the workspace holds more in month twelve than it did in month one, at the same allowance.
Work that is specific to your institution is where AI Build Projects come in: scoped work with a start date and an end date, delivered into your workspace. A notice assembler shaped around your own approved wording is a build project. The generic meeting-notes workflow is a library item. Nothing connects to your core. The workspace works on documents, email and files, and we do not name a core system because naming one would imply an integration we have not built.
How do we know what people are actually doing with it?
You get reporting on what kinds of work AI is being used for, broken out by team. It does not include transcripts of conversations or productivity scores on individuals. The view is categorical: which departments are using it, and the kinds of jobs the reporting distinguishes. We will put an actual report in front of you before you sign, because what the categories are is worth seeing rather than described.
There is a limit worth stating plainly. That reporting cannot demonstrate that no suspicious activity case work happened in the workspace. It shows what kinds of work are being done; it cannot prove the absence of a category. What holds that boundary is the scope, the configuration and the people, which is the same answer we gave further up and the same answer we would give under questioning. Being straight about the limit is worth more than a claim that would not survive the first hard look.
Read the other way, the same reporting is a signal. If a department’s usage starts showing up in categories the scope did not anticipate, that is visible in a report and it is a reason to go ask what changed. It is a signal and not proof, for the reason above. A rule about what AI is used for holds only when somebody can see whether it is being followed.
Collaboration is the other half. When somebody in collections works out a sharper way to write the first letter after a missed payment, that phrasing becomes a template every collector opens. A method that lived in one head now sits in the workflow, and it survives two weeks of vacation and a retirement. The target for year one is plain. In a year, the best AI ideas in your institution should come from your people, not from us.
Questions
Asked by credit unions and community banks
Can we keep our BSA work off this entirely?
In practice, yes: a ScaleSight AI workspace can be scoped so BSA work stays off it, and suspicious activity reports and anything that would reveal one exists are out of scope for the managed service, so we do not build workflows that touch that material. What holds that line is the scoping boundary, the configuration built around it and the people who follow it. It is not a filter that reads content and rejects it, no such gate exists here, and we would rather say that plainly than let you assume otherwise.
Do we buy a seat for everyone on the payroll?
No. ScaleSight AI seats go only to the people who write, summarize or explain for a living, which in a credit union means lending, marketing, HR, accounting, compliance, collections and the executive team. An MSR spends the day inside the core opening accounts and posting transactions, so a seat there would sit unused, and frontline and part-time branch roles mostly do not get one. Seats are per person because that is what licensing is, while budget is allocated by team and department.
Will this be making lending decisions?
No. A ScaleSight AI workspace drafts and summarizes, and it does not score, price or decide. A licensed lender makes the decision and states the reason, and the workflow only restates that reason in your own approved notice language, because a drafted sentence is not a source of reasons. The draft goes to whoever owns notice review before it reaches the member. Information about a member’s protected characteristics does not belong in a drafting workspace and we do not build workflows around it. If anyone wants to point this at underwriting, pricing, asset liability management, CECL or alert disposition, that is a different question and it is outside what ScaleSight AI sells.
Our vendor file needs something on you. Is ScaleSight AI SOC 2 Type II certified?
ScaleSight AI runs on a SOC 2 Type II audited platform, and ScaleSight AI is not itself SOC 2 Type II certified. The workspace is branded ScaleSight AI and white-labeled from a third-party vendor, so the audit belongs to that vendor rather than to ScaleSight AI as a legal entity. We hold to that wording everywhere, including inside a vendor file, because an overstatement in a due diligence packet is the expensive kind of error. We also make no claim to be approved or endorsed by any regulator, and we are not the party that answers to yours. Tell us what your file needs and we will tell you what can be provided and whose name is on each piece of it.
What would a credit union our size actually pay?
A credit union with 45 people on the payroll will not need 45 ScaleSight AI seats, because seats go to the people who write, summarize or explain rather than to everyone at a screen. Take out the frontline and part-time branch roles and the seat count can sit in the 10 to 25 band, which runs $500 to $1,175 a month as a total for the institution. Pricing is per seat plus usage, you set the ceiling on total spend, and a $250 monthly minimum applies. We settle the seat count in scoping before anyone quotes a band.
Other industries
We also work with
Tell us what your week actually looks like.
The readiness review starts by finding where AI is already being used in your company, what it is costing, and which recurring work is worth handing over first. You keep the findings either way.