Your team just finished a project in half the usual time. The client's thrilled, but asks why the invoice is the same when the project took half the time.
For decades, tech services firms existed with steady work and solid revenue, but never the margins of a product company. They played a key role in the life cycle of every new technology, yet never captured the returns a product company would have for the same impact.
Then AI happened.
Venture firms are now betting billions on what they call the "service as software" thesis: that AI can push services businesses toward the 70-85% gross margins software companies have always had. Knowledge, in other words, is starting to behave like software.
But, as it turns out, there's a catch. When AI powers the work, client perceptions change even when the output is of the same quality as before. This isn't a few one-off experiences. It's a documented pattern, with roots in behavioral economics.
The Effort Heuristic
Software services are intangible. When people can't verify quality directly, they substitute a proxy they can see. That proxy is usually effort. Kruger, Wirtz, Van Boven, and Altermatt (2004) found this heuristic gets stronger as ambiguity increases. Add to this the fact that most B2B buyers have little to no experience buying AI-native software services. That's uncharted territory compared to traditional software, and uncharted territory is exactly what the effort heuristic feeds on.
But AI effectively eliminates effort as a barometer of output quality. What used to take hours now takes minutes. That doesn't necessarily mean it's worse quality, but if you were used to evaluating something based on hours of effort it took to get it right, what do you do now?
If you're like most people, you assume that the thing you don't understand that also seems hard to evaluate must not be worth very much after all.
Algorithm Aversion
People like people. In fact, people like people so much that we tend to devalue something from an algorithm that we might have accepted without hesitation from a person. But Castelo, Bos, and Lehmann (2019) found that this affinity is also task-dependent. It's strongest for subjective work requiring judgment, and weakest for objective, easily-verified tasks.
You can probably see where we're going with this.
The Jevons Paradox
It may sound like it's all bad news, but it's actually not. All signs point to software markets being in the middle of experiencing a Jevons paradox: cheaper, faster delivery may shrink customer willingness to pay more for AI-powered software, but the decreased costs will in turn lead to increased demand and total market expansion. Falling cost unlocks demand that was previously priced out.
Our own findings uphold this observation. Repeatedly, founders of tech services companies have told us the same thing: leads and existing clients are coming back with new kinds of AI-related demands:
- A project that felt too expensive to justify a few years ago suddenly looks doable.
- A vibecoded app needs a professional review, or an engineering team to keep it running once it's live.
- An enterprise client needs a compliance expert to make sure their AI-integrated software meets industry and legal requirements.
This means the client asking why a faster project still costs the same and the client reconsidering a project they shelved two years ago as unaffordable can be the same client at two different points in the same sales cycle.
How to Close the Gap
The truth is it is now cheaper than ever to grab a more complete piece of the pie. You can solve the problem more completely, using AI, and software to create recurring value for the customer and tread the line between software and service. So, meet the market where it is.
That means your first step could be to review your pricing models. The hourly model no longer works, but hybrid pricing, outcome pricing, and usage-based pricing are going strong. This is because they don't rely on the old (broken) equation between hours and effort. The effort shows up in the output, human judgment, or accountability.
This also means a critical part of overcoming the perception tax is making your human judgment visible. This can look different for different types of prospects. SME clients may need to see verifiable, objective output indicators such as revenue generated or dollars saved. Enterprise clients will need evidence of the human in the loop and accountability for decisions the system makes. The named expert and the sign-off still command a scarcity premium AI can't touch.
On top of all this, services companies have a unique advantage that product companies struggle to build. They put human relationships first by default. They have the individual buyer's needs at the core of their DNA; they don't fit customers to their solution but build solutions to fit their customers. This has such powerful potential in the new AI-driven reality of Customization at Scale™ that product companies have started putting forth the role of the Forward-Deployed Engineer. Services companies need to respond to this challenge by becoming more human, more intense in building relationships, and with a deeper understanding of the customer's problem.
Wrapping Up
Human bias is real and to some extent, it's just part of being human. It's also par for the course when working with humans, whether as clients or as coworkers. But a bias doesn't need to stump us. It doesn't need to stop us from doing good business. The same shift compressing price on AI-powered work is expanding what's being paid for and who's paying. At the same time, it's important to remember that now that implementation is easier than ever your actual competitive edge will come from the depth of the understanding you offer. Human to human conversations on top of data analysis will go a long way towards serving customer needs, building relationships based on delivering long-term value.
Join tech services founders on September 24 for this month's roundtable: Proof Over Promise. We're pulling together real case studies, the ROI math customers actually respond to, and the patterns showing up across AI-native deployments. RSVP here.