This week we hit the same wall twice with two different vendors, Neon and GiveSmart, both rationing our client’s own data back to us through narrow APIs while …This week we hit the same wall twice with two different vendors, Neon and GiveSmart, both rationing our client’s own data back to us through narrow APIs while keeping the full history behind admin-panel exports. That friction is the tax that breaks AI for most organizations, because the value of AI is bounded by the completeness of the data underneath it. This is why Prospus runs Autonomy Engineering as a service line, and why we now assign every vendor an Autonomy Score during assessment, so leaders can see exactly how much of their own record they actually control.

---

A client engagement this week put us in the same fight twice in two days, with two different vendors, over the same thing: our client’s own data. The first vendor was Neon, their CRM. We connected the API, authenticated cleanly, and started pulling. The system handed us back the latest note on each contact and nothing else. Hundreds of contacts. Years of relationship history. Every donor conversation, every meeting recap, every internal context note that someone on the team had spent time writing. All of it sitting inside Neon, and the official API would only surface the most recent entry per record.

The second vendor was GiveSmart, their fundraising platform. We generated the API key, the connection worked, the data read correctly. Then we saw what was actually coming through: a small handful of campaign names and the most recent people, going back to roughly about one year. Everything before that, gone from our view. Years of donors, events, and donations sitting on their servers and not coming down the pipe.

In both cases the data exists because our client can log in and see it. They funded the labor that produced every note and every donation record. But the vendor still gets to decide which slice of it they hand back.

Why this is a category problem, not a vendor problem

Both Neon and GiveSmart, to their credit, do offer a full CSV export through their admin UIs. So we are doing the obvious thing. We are asking the client to run the exports, send us the CSVs, and we will load them into Kaamfu Mind alongside the live API feed. Within a day the full history sits on every profile next to the recent data. Problem solved for this engagement.

But sit with the shape of that workaround for a moment. The “solution” is that a human being inside the client organization has to log into a vendor admin panel, find the right report, run an export, download a CSV, and email it to us. To get their own data. From a system they pay for. On data they generated. So that we can put it into an environment where an AI can actually use it.

That is the friction tax of operating in the current era of SaaS, and it is the tax that breaks AI for most organizations. You cannot automate chaos, and you cannot draw real-time insights from a record you have to email yourself in pieces.

Neon and GiveSmart are actually the cooperative end of the spectrum because they do let you out, even if the door is narrow. Many vendors do not. We have run into systems where the API gives you a fraction, the export is locked behind an enterprise tier, and the only remaining options are to scrape the interface or open a support ticket and beg. None of those are ideal and all of them are costly. All of them require ongoing human labor to maintain something that should be a continuous, structural feed.

Why this lives inside Autonomy Engineering

This is exactly the work Prospus does under our Autonomy Engineering service line, and it is why we put it there rather than calling it a generic data integration project.

A data integration engagement assumes the data wants to come out, but Autonomy Engineering assumes it does not. It assumes the vendor’s business model depends on keeping their grip on their record of work, and that recovering that record is a structural fight, not a technical configuration step. Our job in an engagement is to look across every place a client’s data lives, score it honestly, and then do the work of getting that data into an environment the client actually owns. That work ranges from clean API ingest on the cooperative end, to scheduled CSV imports in the middle, to scraping and manual reconstruction on the hostile end.

We are doing this for our client because the alternative is that they continue to operate on the slice each vendor decides to show them, and AI on that slice produces nothing of value. The full record is what makes the AI useful. The full record is what we are wrestling back.

The Autonomy Score

Because this pattern repeats on every engagement, we now formalize it. Every place a client’s data lives gets an Autonomy Score, assigned by us during the assessment phase of an engagement.

The Autonomy Score measures how well a given vendor supports your data sovereignty across the dimensions that actually determine whether you can act on your own information: do you have direct, unfiltered, real-time access to every attribute, or only filtered dashboards; is the full record available through the API, or only a recent slice; can you export everything yourself, or is the full export gated behind an enterprise tier; do the APIs return raw records, or stripped, summarized, vendor-shaped versions; can you legally access the categories that matter, or are entire categories walled off behind privacy claims that protect the vendor more than the worker; and does the live feed update continuously, or in batched delays that arrive days after the decision was needed.

A high Autonomy Score means the vendor treats your data as yours. You see everything, in real time, through clean interfaces, with no negotiated tiers in the way. A low Autonomy Score means the vendor treats your data as their asset to ration. You see a slice, in batches, through filters they designed, with the full record behind a paywall, a support queue, or a refusal.

Most of the systems sitting inside the average mid-market organization right now score low. That is not an accident. It is the business model of the last era of SaaS, where the vendor’s leverage over the customer was the vendor’s ongoing control over the customer’s record. In an AI-first world that leverage flips from a nuisance into a structural disadvantage, because the value of AI is bounded by the completeness of the data you can feed it.

The Workstack is the asset. Vendor control is the obstacle.

This is the practical version of the argument we made in our Workstack Control guide. Your Workstack, the complete record of work activity that surrounds and shapes your decisions, is your most valuable data asset. It captures who does what, how decisions get made, where time goes, and what outcomes cost. With AI on top of it, the quality of insight available to a mid-market organization for the first time matches what only the largest companies used to be able to afford.

But none of that becomes real if the Workstack is scattered across vendor tools that each ration their slice of it back to you. The average organization can directly access less than 20% of the data they generate. The other 80% is sitting on someone else’s servers, behind someone else’s pricing tier, accessible only through someone else’s API decisions. That is the gap Autonomy Engineering closes.

This week’s two engagements with Neon and GiveSmart are small examples of a structural shift that every founder-led firm is going to have to make. You either keep renting your Workstack from the vendors who own it today, or you take steps to reclaim it. Reclaiming it is not a single project. It is an ongoing engineering discipline, because the vendors are not going to make it easy, and the AI on top is only as good as the data underneath.

If you want to know what your own Autonomy Score looks like across the systems your team uses every day, that is the assessment we run at the start of every Prospus engagement. It is usually the first time the leader sees, in one place, exactly how much of their own organization’s record they actually control.

…