AI-first, with people in charge.

Not for the strategy. The AI Blueprint, the policy and staff training work whatever systems you run today.

For uses that read your member and donor data, you need one clean record first, and for most of our clients that record is in Salesforce, where we build. If your data lives in several systems, AI will repeat the disagreement between them, quickly and convincingly. The Digital Readiness Blueprint tells you what it would take to get to one record, and which uses can start before you get there.

A language model writes by predicting what a good answer would most likely say. When the facts are in front of it, that gives a clear answer; when they are not, it gives something that sounds like one anyway. So we do not rely on it knowing anything.

Each use is grounded: it first finds the right records and your published documents, answers only from those, and shows its source. When the answer is not there, it says so and hands over to a person. We test it on a set of real questions before launch, re-run that test after every change, and read a sample of live answers every week. Most of the mistakes we find turn out to be in the data, which is why the foundation comes first.

You decide, use by use, and it is written down: what each use can see, what it can do, and who reviews it. Permissions are enforced when records are retrieved, so a staff member gets answers only from records they can already see, and a member only from their own.

Where data goes to a model, it goes under business terms that do not allow it to be used for training, and we keep it in systems you already control wherever we can. Sensitive information, such as health details or scholarship applications, is kept out unless you approve a specific use.

The one that fits the use. Inside Salesforce we use its own AI and agents where they fit the job; where another leading model fits better, we use that, connected to the same record and the same permissions. You always know which model sits behind each use, and when a model changes, the test set is run again before anything goes live.

Usually the AI Blueprint: four to six weeks with your team to find out what is already in use, build the portfolio of uses, check how ready the data is, draft the policy and agree a 12-month roadmap your leadership team can stand behind.

If your record is already in good shape, we can instead start with one low-risk use from the roadmap, run it for one cycle with a person reviewing everything, and decide on the next from the result. The fee is fixed and quoted before you commit.

Before a use is switched on, we agree one number it should move, measured the same way before and after: renewals among flagged members, time to first reply, hours to prepare the leadership report. Time saved is measured by asking staff to log it for a few weeks, not by estimating it in a proposal.

We also track whether it is right, from the test set and the weekly sample, and what it costs, including the time people spend reviewing. Each use has a decision date. If the number does not move, switching it off is the process working.

That is not what we build it for. Associations and nonprofits are rarely overstaffed; the uses that pay back take the repetitive work off people who are already stretched: the weekly list, the routine question, the report that takes three days to assemble.

The judgment stays with your team, and so do the decisions about people. What changes is how much of the day goes on the work only they can do.