This real estate investment manager already had AI in use without anyone deciding on it, since one team member built a deal-screening tool and then paused it over data-security questions, so the firm had AI in use with no way to decide what was safe to run or how to explain it to an examiner. New Clarity ran an AI audit over about three weeks and interviewed people across acquisitions, asset management, fund operations, investor relations, accounting and legal. The result is a strategic plan the firm owns that maps six functions and ten workflows, and each workflow carries a build-or-buy call, an estimate of the time it would save with the assumptions shown, and human review, recordkeeping, access and retention controls designed in from the start. The plan also settles the data foundation question in favor of an ordinary database with scripts around it, since a small team with no IT department needs an organized record with controls around it, and a full data platform is a system someone has to administer. The firm now has a three-phase roadmap over about nine months, a safe path into production for the paused screening tool, and a document it can put in front of an examiner or an investor showing which workflows it chose, why, and what sits around each one.
Most private fund managers have AI in use with no decision behind it, since someone on the team builds a tool that works and someone else pastes a document into a chat window, so the firm ends up with AI in production with no record of what it touches, who reviewed it, or where the data ends up. That is the real problem for a regulated manager, since the firm has to explain to an examiner or an investor how any of it is controlled, so a registered manager cannot adopt AI the way a normal small business does.
The two approaches most firms consider do not work here. Buying an enterprise platform and a data warehouse assumes a firm with an IT department to run it, and it front-loads a large, heavy build before anyone knows which workflows are worth automating. Letting each team pick whatever tool it likes is faster, but it spreads client and fund data across services nobody approved and leaves no record behind. Both leave open the question a regulated manager has to answer, which is which workflows are worth doing and what controls sit around them.
This registered real estate investment manager was a clear example. The firm runs two funds with an investing team, operating partners, and an outside compliance consultant, so it carries the obligations of a much larger institution with a small team doing the work. One team member had built a deal-screening tool and then paused it over data-security questions, which was the right call and also a sign the firm had no way to decide what was safe to run.
The day-to-day work sat in the usual places, and that was the second half of the problem:
Deal, fund and investor information spread across email, shared folders and spreadsheets, with no single record anyone could point to.
Operating-partner reports arriving in whatever format each partner chose to send.
Work moving between acquisitions, asset management, fund operations, accounting, investor relations and legal through handoffs that depended on someone remembering.
None of that is unusual, and it matters because you cannot put controls, review steps or retention rules around a process that lives in nobody's system, so any AI built on top of it inherits the same gaps. The firm needed to know which work to automate first, what to build versus buy, what each step would save, and what controls had to exist from day one.
New Clarity ran an AI audit and delivered a strategic plan the firm owns. The short version is that instead of picking a platform or a tool, the team started with how the firm runs and worked out which parts of the work are worth automating, what to build versus buy, what each one is worth, and what controls have to sit around it before anyone turns it on.
The work took about three weeks. It started with one-on-one interviews across the firm, including the managing partner, capital formation, general counsel and compliance, fund operations and accounting, acquisitions, asset management and administration, so the plan reflects how the work gets done rather than how an org chart says it should. The team also reviewed the tools already in use, including the deal-screening tool the team member had built and paused, and looked at where information lives today across email, shared folders, spreadsheets and operating-partner reporting.
Out of that came a map of six functions (deal sourcing and screening, asset and portfolio management, capital formation and investor relations, fund operations and accounting, legal and compliance, executive support) and ten workflows inside them. Each workflow is written up as a decision the leadership team can make, not a recommendation to take on faith.
Build or buy on each one. Every workflow carries a call on whether the firm should configure something off the shelf, connect systems it already pays for, build something custom, or leave it alone for now. The point is that a lot of this work does not need custom software, and saying so up front keeps the firm from paying for a platform to solve a problem a spreadsheet replacement would handle.
Named components. Each workflow lists what it would actually be made of, so leadership can see the scope of a thing before agreeing to it rather than after.
Financial impact with the assumptions in the open. The plan shows the estimated value of each workflow along with the math and the assumptions behind it, so the firm can change an assumption it disagrees with and watch the number move. To be clear, these are estimates the firm can adjust, not promises.
An impact-versus-cost view. All ten workflows are placed against each other on impact and cost, which is how the sequencing in the roadmap was decided instead of by whoever asked loudest.
This is the part that separates a regulated manager from a normal small business, and it is why an off-the-shelf plan does not work here. Every workflow in the plan carries three things from the start: a human review step before anything leaves the firm or drives a decision, a record of what the system did and what it touched, and controls for who can see the data and how long it is kept.
An independent compliance consultancy reviewed the plan and wrote compliance considerations on every workflow, and every current-state finding carries a rating of how much exposure it represents today. The plan also includes a section mapping current examination priorities on AI against what is being proposed, so the firm can see where each workflow lines up. New Clarity is not a compliance consultancy or a law firm, so the compliance review sits with the firm's own consultant and counsel, which is the right place for it.
The plan lays out nine months of work in three phases, sequenced so the firm builds the foundation before the things that depend on it:
Phase one puts a real deal record in place and moves broker-email intake and deal screening onto it, which also gives the paused screening tool somewhere safe to run.
Phase two covers operating-partner reporting, loan and property drift checks, fund reporting and a legal playbook.
Phase three covers investor records, outreach and investor materials.
These phases are the road ahead, not work already done. What the firm has now is the plan, which is a document it owns and can hand to an examiner or an investor to show which workflows it chose, why it chose them, and what sits around each one. In other words, when the question comes about how AI is controlled at the firm, there is an answer on paper instead of a conversation.
This engagement was an audit and a plan, so the technical work went into specifying how each workflow would be put together, what it would be made of, and where it would break. A few of those calls are worth spelling out, since they are what makes the plan something a regulated manager can act on.
The same property shows up in the firm's email under a broker's subject line, in a folder under a street address, and in a spreadsheet under an internal shorthand. The plan calls for a canonical record (one golden record per deal) plus entity resolution rules that decide which incoming email, document or partner report belongs to which deal, including aliases for the same property or sponsor across sources. This is phase one for a practical reason, since any screening or reporting automation built before it produces output that attaches to nothing and has to be filed by hand anyway.
The audit included a walkthrough of the deal-screening tool a team member had built with an AI coding assistant, looking at what it did, how the team was relying on it, and what was missing for production. The plan specifies moving the tool onto firm infrastructure, putting credentials and API keys into managed secret storage, logging every run and what it touched, and adding a defined human review step before a screen reaches the twice-weekly pipeline meeting. The person who paused it was right to pause it, since the output was useful while nobody could say which data left the firm.
Broker emails and offering materials get read for the same handful of facts every time, so the plan treats intake as structured extraction against a fixed schema. Each field comes back with the source it came from and a confidence, and anything the model is unsure about goes to a person for a look. That is the part an off-the-shelf chat tool cannot give you, since a chat window returns prose with no schema, no source and no place to put the answer.
For operating-partner reporting, the plan calls for extraction into a common set of fields plus a time-aware record that keeps each period as its own version. Drift checks then compare this period's loan and property numbers against what the record already holds and flag movement that looks off. The engineering reason for keeping history is that a flat snapshot only knows today, and a manager needs to show what was reported when, so a restatement stays traceable.
Diligence summaries, draft IC memo sections, fund reporting and the legal playbook all run on retrieval over the firm's own documents (RAG), so every answer comes out of a document the firm holds. The plan specifies hybrid search, meaning semantic vector search for the concept plus exact keyword matching, because fund documents are full of defined terms, section numbers and entity names where an approximate match is the wrong answer, with a reranker on top to put the best passages in front of the model. Every generated line carries a citation to the source document and location, so grounding becomes a review step a person can finish in about a minute.
Role-based access groups across acquisitions, asset management, fund operations, investor relations, accounting and legal, retention rules, and handling for sensitive investor and fund data are specified once at the record, so whoever builds the next workflow inherits them. The plan also documents access controls and data flows in a form the firm's compliance consultant and counsel can review. Anything applied tool by tool gets forgotten by the third tool, so a firm ends up with a control it believes in and cannot evidence.
The plan pairs each workflow with a way to check it, starting with past deals the team screened by hand as a test set, so extraction and screening output can be scored against a known answer. Consistency checks can use LLM-as-judge across runs, meaning one model scores another model's output, with human spot checks on a sample, so a person signs off in the end. Phase two also captures measured time savings once the deal record is in use, which feeds back into the estimates from the plan, so the assumptions get corrected by what the firm sees.
One of the real decisions in the audit was comparing data platform approaches, a full data platform against an ordinary database with scripts around it. For a firm with a small team and no IT department, the plan calls for the plain database, since a warehouse is a system someone has to run, and the firm needs an organized record with controls around it.
What the firm has now is a finished AI strategic plan it owns, covering six functions and ten workflows, with a three-phase roadmap over about nine months. The work took about three weeks. The roadmap is work ahead, so the outcome today is that the decisions are made and ordered, and the firm can start building in a known sequence.
The deal-screening tool a team member had built was sitting unused because nobody could say what data left the firm when it ran. The plan gives it a defined path into production, meaning firm infrastructure, managed secret storage, logging on every run, a review step before a screen reaches the twice-weekly pipeline meeting, and a deal record to write its output into, so the acquisitions team can rely on it day to day.
All ten workflows come with a build-or-buy call, the components they would be made of, an estimate of the time each one would save with the assumptions shown, and a position against each other on impact and effort. The firm can approve, defer, or drop any one of them on its own terms, and it can change an assumption it disagrees with and see the estimate move. Leadership had been judging vendor pitches one at a time with no way to compare them, so having all ten side by side is new.
Every workflow carries compliance considerations written by an independent compliance consultancy, every current-state finding carries a rating of how much exposure it represents today, and the plan includes a section mapping current examination priorities on AI against what is proposed. So when an examiner or an investor asks how AI is used and controlled at the firm, there is a document to put in front of them showing which workflows were chosen, why, and what human review, recordkeeping, access and retention controls sit around each one. New Clarity is not a compliance consultancy or a law firm, so that review stays with the firm's own consultant and counsel.
The data foundation call was the plain database, so the firm gets an organized record with controls around it without taking on a system that needs someone to run it. For a small team with no IT department, that means a plan they can carry out with the people they have.
Deal volume was capped by how many broker emails and offering documents a person can read and file, so phase one puts the deal record and intake in place, which is what lets the team screen more deals without adding headcount. Phase two adds partner reporting and drift checks, so asset management can see loan and property movement across the portfolio each period instead of reading every report one by one. Phase two also captures measured time savings once the deal record is in use, so the estimates in the plan get corrected by what the firm sees.
Other private fund managers have the same problem, which is a small team carrying the obligations of a much larger institution and tools already in use that nobody signed off on. What this firm got out of the work was a decision on which work is worth automating, in what order, and what controls have to exist before the first build starts.