A real estate investment firm that raises capital and develops and manages residential communities had outgrown the systems it started with, so operational, customer, and investor information sat across spreadsheets, email, and shared drives with nothing tying them together. New Clarity interviewed people one-on-one across leadership, sales and marketing, construction and project delivery, finance and administration, and customer service, mapped how the business runs today, and went through every tool in use, including the AI tools an outside marketing agency was operating. The result is a written AI strategic plan that ranks the workflows worth automating, says for each one whether to build it or buy it, shows the estimates behind the ranking so the team can adjust them, and lays the work out in phases. The plan also spells out the shared data foundation the firm's future AI work depends on, scoped to run on an ordinary database, so the firm can get organized, dependable data without standing up a data warehouse first. The firm came out of the engagement with an agreed order of operations it can explain to its partners and investors, and a plan it owns and can run with or without New Clarity.
Most firms that get serious about AI start by shopping for tools, and that is where the whole thing stalls. The real constraint is almost never the tool, it is that the business runs on spreadsheets, email, and shared drives with nothing underneath tying them together. In other words, there is no shared view of the company's own data, so every tool you add is reading a partial copy of the truth and every result has to be checked by hand before anyone trusts it. That blocks the things leadership actually wants, which are reliable reporting, a clear picture of the pipeline, and answers that do not depend on one person remembering where a file lives.
The two obvious responses both fail for structural reasons. Buying a point tool or standing up a chatbot gets you something to demo in a week, but it sits on the same scattered data, so it never gets past the demo and adoption dies. Going the other direction and building a full data warehouse is the heavyweight answer that looks responsible on paper, but it is a long, costly project for a lean team, and it forces decisions about what to store and how to model it before anyone has agreed on which workflows matter. To be clear, the sequencing is the problem, since you end up building infrastructure for work you have not prioritized yet.
This real estate investment firm is a good example. It raises capital, develops and manages residential communities, and answers to its investors, and it had grown to the point where systems built for a smaller operation were not keeping up. The signs were the ordinary ones:
Operational, customer, and investor information spread across spreadsheets, email, and shared drives, with no single place to look
A CRM in place alongside reporting spreadsheets and general AI use, plus AI tools an outside marketing agency was running, so no one had a full picture of what was already deployed
Reporting and record-keeping for investors that leaned on manual assembly rather than a system
Some tools that had been tried and not adopted, which makes the next proposal a harder sell internally
None of those symptoms is the actual problem. The actual problem is that leadership was being pushed toward AI without a documented view of how the firm runs, which workflows are worth automating, or what order to do them in, so any spending decision was a guess. That is why the first deliverable had to be a written plan the firm owns rather than a pilot.
New Clarity built the firm a written AI strategic plan it owns, which is a documented picture of how the business runs today paired with a prioritized, sequenced list of what to automate and in what order. In other words, the deliverable is a decision document, not software, since the firm needed a defensible basis for spending before anyone wrote code or signed up for another tool.
The work ran in a straight line from discovery to a roadmap leadership agreed on:
The team ran one-on-one interviews (roughly 60 to 90 minutes each) with leadership, sales and marketing, construction and project delivery, finance and administration, and customer service. One-on-one matters here because people describe their own workarounds honestly when their boss and their peers are not in the room, and the workarounds are where the real process lives.
Every interview fed a current-state map of operations, reporting, and how information gets passed between people, with the areas of improvement called out on the map itself. To be clear, this is the part most firms skip, and it is why their AI projects target whatever workflow the loudest person named in a meeting.
The team went through the tools in place, including the CRM, the reporting spreadsheets, general AI use across the team, and the AI tools an outside marketing agency was operating, then made a call on each one (keep, configure, upgrade, or replace). The agency piece mattered because nobody inside the firm had a full list of what was already deployed on its behalf.
The plan documents where operational, customer, and investor information sits today, who has access to it, and what investor reporting and record-keeping need from it. That assessment is what turns "we should get organized" into a specific case for a shared data foundation, which the plan recommends rather than assumes.
Each candidate workflow got written up with a recommendation on whether to buy something off the shelf, configure what the firm already pays for, or build. Some problems have a decent product on the market and some do not, so treating every item the same way is how teams end up either overbuilding or stacking subscriptions nobody uses.
The team estimated cost savings, revenue uplift, and capacity gains per initiative, and left the formulas and inputs exposed so leadership can change an assumption and watch the ranking move. Estimates from an outside firm are guesses dressed up in a spreadsheet unless the client can argue with the inputs, so the model is built to be argued with.
Everything landed on an impact-versus-cost view to guide sequencing, then into a phased roadmap, with a shared data foundation and ongoing operational AI support named as follow-on phases rather than work already done. The plan closed with a walkthrough session with the leadership team to work through questions and settle the order of operations.
This is the part that separates the approach from buying a tool or standing up a chatbot. An off-the-shelf product answers the question "what can this do," while the plan answers "what should we do first, what is it worth, and what has to exist underneath it before it works." The firm had already tried tools that did not get adopted, so a plan grounded in its own interviews, its own map, and its own numbers is what makes the next proposal something the team believes rather than another pilot to sit through.
This phase delivered a plan, so the technical work shows up as a written architecture rather than shipped code. The plan names the parts a shared data foundation needs and says why each one is there, since a handful of decisions made up front are what let an AI agent trust the data it reads.
The plan calls for entity resolution with a canonical registry, which means every spelling, alias, and partial record for a buyer, investor, vendor, or project lines up under one master record. A firm with the same investor sitting in a CRM, an inbox, a spreadsheet, and a folder of signed documents needs something to decide those four are the same party before it can ask one question across all of them.
The recommended store is bi-temporal, which tracks two clocks at once: when something was true in the real world, and when the firm learned it. Facts are never overwritten, so the earlier state stays in place and the firm can ask as-of questions ("who was the contact on this deal at closing"). Investor reporting and record-keeping both depend on showing what was known at the time, so a fund manager gets a record it can stand behind.
Retrieval in the plan is hybrid, pairing semantic search (which finds things by meaning) with keyword search (which finds the exact words), then sharpening the result with a final rerank before anything reaches a model. Semantic search on its own misses a unit number or a document title, and keyword search on its own misses a question asked in different words, so running both and reranking is how the firm gets the answer it asked for.
Redaction and separation rules get applied at ingestion instead of inside each tool, since a rule that has to be rebuilt in every agent will get missed somewhere. Handling it at capture means a new agent picks up the same protections with no extra setup.
All of this is served through a single interface, so no individual agent rebuilds retrieval or identity matching, one improvement to the foundation helps every agent at once, and a new agent plugs straight in. It runs on an ordinary database, which is the point of scoping it as a phase, and the firm gets the organized foundation its agents need without standing up a data warehouse first.
The firm finished the engagement holding a written AI strategic plan and an agreed order of operations, which is what it did not have going in. No system went live in this phase since the deliverable was a decision document, so the results here are qualitative on purpose. The plan covers all five functions, ranks the workflows worth automating, and lays the work out in phases, so leadership can make spending calls on AI with a documented basis and the team can start on the first item without another round of discovery.
How the work flows across leadership, sales and marketing, construction and project delivery, finance and administration, and customer service sits in one document with the weak spots called out. That knowledge used to live in people's heads across five functions, so answering a basic question about a process meant tracking down the person who owned it. The firm can hand this map to a new hire, a vendor, or an auditor of its own choosing without re-interviewing the whole team.
The ranked workflows, the impact view, and the phased roadmap give leadership a clear answer to what comes first and what waits. Every proposal used to be judged on its own, so the loudest idea in the room tended to win. The team can say yes to one initiative and later to the rest, and explain why in a sentence.
The impact model ships with the formulas and inputs visible, so the firm can change an assumption about volume or staff time and watch the ranking move. The finance side can re-run it without calling New Clarity back in, so the model keeps working as conditions change. Most outside analysis cannot be inspected, so it gets ignored the first time it looks wrong.
The tool review gave the firm one list of everything deployed on its behalf, including the AI tools an outside marketing agency was operating, with a keep, configure, upgrade, or replace call on each. Nobody inside the firm had that list before, so the team can hold the agency to a standard, stop paying for overlapping tools, and get more out of the CRM it has before buying anything new.
The assessment of where operational, customer, and investor information lives (spreadsheets, email, and shared drives) turned a general sense that things should be organized into a specific, scoped recommendation for a shared data foundation as a follow-on phase. The firm can scope that work, staff it, or put it out to another vendor, and it knows what investor reporting and record-keeping need from it before anyone starts building.
The plan is the firm's property and it can execute it with or without New Clarity. The team had tried tools that did not get adopted, so the next internal proposal needs to come from the firm's own interviews, its own map, and its own numbers.
Most private fund managers are in the same spot. If you run a private equity firm or a real estate investment manager, and your CFO, COO, or general counsel is being pushed toward AI while the business still runs on spreadsheets, email, and shared drives, the useful first step is a plan you can explain to your partners and your investors. Work out the order of operations first and the builds that follow are shorter and more likely to stick.