An executive director at a nonprofit health and human services provider walked us through her situation earlier this year. She wanted a CRM. What she actually had was six systems and one customer living in most of them.
She also had a staffing problem she hadn't named as one. Going through the stakeholder map with her, the same person kept landing in every row: executive sponsor, project manager, IT leadership, and the approver for security and compliance. Four roles, one director, on top of running the organization. That shapes how you have to run the project, and it is the single most common condition we see at this size.
Six systems, and the reason none of them were wrong
Salesforce held the customer, donor, and coordinated care lifecycle, but with no standardized process flow and no connection to service delivery or finance. Grants and donations were tracked separately in Monday.com, which meant the same business function was recorded in two places. QuickBooks ran accounting, disconnected from operations, reconciled by hand. Claims, payments, and eligibility checks ran through Tebra, also disconnected. Marketing was split across MailerLite, Stripe, and Gravity Forms on the website. Spreadsheets covered billing, claim reconciliation, and funded equipment inventory.
Every one of those was a sensible purchase at the time. The problem only appears when you line them up.
The finding that changed the estimate
The most consequential thing discovery turned up was not a missing feature. It was that critical information lived in PDFs and email attachments rather than in fields. Service authorizations, consents, vendor credentials. All present, all findable by a human who knew where to look, none of it reportable and none of it automatable.
That reframes the project. You are not moving data between systems, you are converting documents into structure. It also explains how the same organization can have a filing cabinet full of compliance evidence and still be unable to answer "which vendors have an expiring credential this quarter."
Pick three questions leadership asks every month. If the answer lives in an attachment rather than a field, your CRM project has a data structuring phase in it whether or not anyone has budgeted for one.
Four weeks of design, before anyone quoted a build
We did not quote an implementation off a discovery call. We ran a paid Phase 0: four weeks, fourteen workshops across every workstream, ending in a solution design document, a data dictionary and ERD, a security role matrix, a migration strategy, and a RAID log.
It was its own engagement with its own fee, credited in full against the implementation that followed. The client owns the design outright, including the right to take it to a different partner. A design phase someone can walk away with has to actually be good.
It also produced a decision log with reasons attached, not just outcomes. On eligibility, the recorded decision was a real-time API integration to Office Ally's 270/271 transaction rather than batch file exchange, fired from inside the service order record at two separate stages, because it lets coordinators decide faster and cuts downstream billing rework. Months later that sentence is still what settles the argument.
What the compliance work actually looked like
Twenty-one security roles across the applications and eight documented segregation-of-duties controls: purchase order creation separated from approval, vendor master creation separated from payment processing, invoice posting separated from payment release with dual approval above a set threshold, asset creation separated from financial capitalization, and user provisioning held outside every functional role.
Alongside that, a library of twenty-four operational reports. The ones that mattered most were the compliance ones. A vendor compliance report tracking insurance, OIG exclusion checks, and background checks with alerts at 30, 60, and 90 days before expiration. A consent and guardian compliance report showing which records are blocked and why. A service authorization report flagging missing authorization numbers before they turn into denied claims.
None of that ships in a CRM. All of it is a purpose-built table on the same Dataverse platform, with the same security model and the same reporting layer as everything else.
The recurring cost nobody scopes
Two things worth naming, because they catch people out in opposite directions.
An eligibility integration is built once and licensed forever. The clearinghouse charges annually for API access to the transaction, and that line is entirely separate from the cost of building the integration. Integration work gets scoped as implementation hours and the recurring vendor fee gets discovered after the budget is approved. Ask about it during evaluation, not after.
The AI components ran the other way. Automated vendor invoice capture through the Business Central payables agent, and service order document intake through Copilot Studio, both consume credits per document rather than per seat. That means the cost scales with actual volume and can be forecast from your real invoice and referral counts. It is a much easier number to defend than a per-user license you have to guess at.
What we left out
The document template library, roughly twenty Word and PDF forms, stayed with the client. So did survey form creation, outgoing communication templates, and Dynamics 365 license procurement. All written into the scope as out of scope, before anyone started building. We build the automation rules that fire the templates. They own the templates.
Where it stands
Design is signed off. The implementation is underway, structured as a fixed fee released against demonstrated phase gates rather than by calendar month, running through a single-wave weekend cutover, thirty days of hypercare, and knowledge transfer to internal staff.
If that shape is familiar, several systems, several teams, and one customer who exists in all of them, the first question is not which platform. It is what the system is supposed to be, written down, before anyone configures anything.