The Manager Tax
A researcher at Strange Loop Canon ran an AI experiment with 15 tasks and three setups.
- An LLM model working alone.
- A “manager” model that split the work and delegated to AI specialist agents.
- And a market where models bid for the job and the best bid won.
The manager cost four times more than the market and produced worse output.
This manager model is predominant. An account director at the top, splitting the brief across channels (direct mail, email, telemarketing, major gifts) and sending the pieces out for revision. The donor at the other end gets a fragmented, over-frequent experience. You pay quadruple for the orchestration and the orchestration is the problem.
Why does the model break? The manager has to nail two things before anybody else can do anything. Know what the subtasks are. Know what good reassembly looks like. Get either wrong, and individually excellent specialists produce a mess.
Coase, Nobel laureate economist, said firms exist when using a market is too expensive: too much haggling, too much measurement. Hayek, another pro market economist, said the magic of a market is that prices reveal what people actually know. Fundraising got the Coase half. Big integrated agencies and plenty of overhead. It skipped Hayek. There’s no real price signal on what matters. So the sector grades on what’s legible: sends and gross response. Activity becomes the proxy because activity is cheap to count.
That is the volume trap in a sentence, optimizing the legible thing whether it’s helpful or hurtful. Then we wonder why retention keeps drifting south and acquisition has to run harder to fill the bucket.
The fix isn’t a smarter manager, it’s pricing on the right outcome. Pay for net retained value, three years out, not the campaign that closes Friday. Until the price signal is right, you will keep paying four times more for worse work. The AI agents at least have the excuse of being new at this.
Kevin
Behavioral Science Q & A
Thanks so much for raising this. Yes, capturing donor information can be helpful for stewardship like newsletters, thank-you letters, impact updates. But how you ask matters. Forcing full data capture introduces friction that can significantly depress conversion, many donors may simply abandon the process. Beyond the friction itself, required fields also shift the emotional experience […]
Read Full Answer
Unlike holidays that everyone already knows, Giving Tuesday is a created event. Many donors recognize the name but not the exact timing, so referencing it becomes a helpful cue. It serves as a reminder and taps into social norm activation (“everyone’s giving today”), which boosts response. However, we still want it paired with the mission, […]
Read Full Answer
When a subject line leads with the match (“Your gift matched!”), it risks triggering market-norm thinking: the sense that giving is a financial transaction rather than an act rooted in values, identity, and care. This shift reduces intrinsic motivation and, over time, can weaken donor satisfaction and long-term engagement. It also makes the email indistinguishable […]
Read Full Answer
There’s no evidence that QR codes suppress mid-value giving; all available research suggests they either help or have no negative effect. In fact, behavioral and usability research consistently shows the opposite: reducing friction at any point in the donation process increases completion rates and total response. And that has nothing to do with capacity and […]
Read Full Answer
What you’re experiencing is very common. Resistance often isn’t about capability, but about motivation quality. If board members feel pushed into fundraising, that triggers controlled motivation (low quality motivation) i.e. obligation, guilt, or fear of judgment, which often results in avoidance. Instead, we need to create conditions for volitional motivation (high quality motivation) by satisfying […]
Read Full Answer
That’s a really thoughtful question, and you’re not the first to raise it. Many of our clients have been cautious about placing the ask at the very end. To address their concern, we’ve tested both approaches, and the results are clear: when the ask comes last, even if that means it appears on the second […]
Read Full Answer