Why inbound slowed at your software development shop
Inbound inquiries at your dev shop fell 40% and your team says clients build it with AI now. That explanation covers four different structural causes, and only one of them is actually substitution.
By Stacey Tallitsch | July 13, 2026
Your software development shop booked 40% fewer inbound inquiries this year than last. Not fewer wins. Fewer people asking. The contact form is quiet, the referral calls stopped, and the pipeline you used to fill without trying now takes work to fill at all. Someone on your team has already supplied the explanation, and it sounded credible enough that you stopped looking: clients just build it themselves with AI now.
Maybe. But that sentence is doing a lot of work for a sentence nobody tested. It explains everything, which is the first sign it explains nothing. And it leads directly to the two worst moves available to you — cut marketing because demand is gone, or bolt on outbound because you need to fight harder for a shrinking pie.
Before you do either, you need to know which of four things actually happened, because three of them are not demand destruction and one of them is only partial.
The market did not leave. Check the number.
Start with the fact that contradicts the story your team told you. The Bureau of Labor Statistics projects computer systems design and related services to grow 15.8% between 2024 and 2034, adding roughly 386,800 jobs — one of the 15 fastest-growing industries in the economy, and one of the 15 adding the most jobs outright. That projection was published with full knowledge of what code-generation tools do. The people whose job is counting this market do not think the market is going away.
So the aggregate demand for building software is not collapsing. Yours fell anyway. That gap is the whole diagnosis. Something moved. Demand did not evaporate; it relocated, and it relocated out of the exact segment your inbound was made of.
Four candidates. They look identical on a lead-count dashboard, which is why the lead-count dashboard is where founders go to get confused.
Substitution, but only at the floor
Code generation did eat something real. It ate the small, fully specified, low-ambiguity job. The $20K internal tool. The CRUD app with a spec attached. The scripting work. The integration a competent product manager can now get 80% of the way to over a weekend.
That work has not become impossible to sell. It has become impossible to sell at your rate, to a buyer who now has a cheaper first attempt available. If your inbound was disproportionately made of small, well-specified projects — and for most shops between $2M and $8M, it was, because those are the ones people find you for — then your inquiry count can fall 40% while the dollar value of the market above that line does not move at all.
This is the one real substitution effect, and it is narrower than the panic. Test it directly: pull 24 months of inquiries, band them by deal size, and look at where the drop lives. If it is concentrated under a threshold and in scopes the client wrote themselves, you have substitution at the floor. If the drop is spread evenly across bands, AI is not your problem and you are about to fix the wrong thing.
The referral engine changed hands
Software shops of your size run on referral. That channel does not report why it stops. Your referral sources are individual people — a CTO, a VP of engineering, a product lead — and when those people move roles, take a job at a company with an in-house team, or start handling the small work internally, the channel goes quiet without anyone telling you.
That is not a demand event. That is a source event, and it looks exactly like a demand event from where you sit. I have written before about how to tell why your referrals slowed before you start advertising, and the diagnostic there transfers: name the specific humans who sent you work in the prior 24 months, then find out what changed about them. Usually four names account for most of it. Usually two of those names changed jobs.
You went invisible where the shortlist gets made
Buyers used to search. Now a meaningful share of technical buyers open an assistant and ask for a shortlist of firms before they run a single query, and the shortlist is assembled from what the model can see about you.
If your firm's public surface is a portfolio of logos and a services page that says "senior engineers, agile delivery, flexible engagement models," you are not describing anything a model can attach to a problem. You are describing capacity. Every one of your competitors describes capacity. The result is not that you rank badly. The result is that you do not appear in the consideration set at all, and no inquiry is ever created for you to lose. Nothing shows up in your funnel because you never entered anyone's.
This one is brutal precisely because it produces silence rather than rejection. Rejection at least generates data.
You are still selling hours into a market that stopped buying hours
The last one is the one nobody wants. When your buyer could not write code, hours were the product. Now the buyer's junior team can generate a plausible first draft, which means the scarce thing is no longer typing. The scarce thing is knowing what to build, knowing what will break in production, and being accountable when it does.
Shops that still lead with time-and-materials staff augmentation are selling the exact input that just got cheap. The inquiry does not arrive because the buyer no longer perceives a gap you fill. This is not price pressure. It is a category error in your offer, and it manifests as the cost to win each new customer climbing right up until the inquiries stop arriving at all.
What the drop is actually telling you
Here is the turn. Everyone reads the 40% as a verdict on the market. It is a verdict on the composition of your pipeline.
The shops getting hit hardest are not the worst shops. They are the shops whose inbound was built on the most easily specified work — which, historically, was a sign of a healthy, frictionless funnel. Easy-to-describe projects generate more inquiries. They convert faster. They made your lead count look excellent for a decade. That same characteristic is what made them the first thing a code-generation tool could take.
Your inbound decline is a lagging indicator of what fraction of your revenue was replaceable. Read it that way and it becomes the most useful number you have gotten in years. It is telling you, in one figure, how much of your business was built on the part of the work that was always going to commoditize. A shop that lost 40% of inquiries but held revenue learned that its lead count was mostly noise. A shop that lost 40% of inquiries and 40% of revenue learned something much more serious, and needs to act on it this quarter, not next year.
The same structural mistake shows up in adjacent businesses. A managed services firm signing clients every month while recurring revenue stays flat is looking at the same class of error: a top-line count that stopped correlating with the thing that actually pays.
The move this week
Do not cut marketing. Do not buy outbound. Do not rewrite the website yet.
Pull the last 24 months of inquiries into one sheet. Four columns: date, estimated deal size, who wrote the scope (you or them), and how they found you. Band the deal sizes. Then answer one question — is the drop concentrated in small, client-specified scopes, or is it evenly distributed?
Concentrated means substitution at the floor, and your response is to stop competing for work the tools now do and rebuild the offer around ambiguity, integration, and accountability, which is where the 15.8% is going. Evenly distributed means this was never about AI, and you have a visibility or positioning problem you have been misfiling under a technology story for a year.
You cannot fix a problem you have not named. And the AI explanation is not a name. It is a place people stop looking.
— Stacey Tallitsch, Stronghold CMO
About the Author
Stacey Tallitsch is the President of Stronghold CMO, a Fractional AI CMO service operating under Talisman Capital, Inc. He is a 30-year tech veteran and the author of 21 books on systems thinking, operator-grade decision-making, and personal sovereignty, with more than 30,000 students across his Udemy course catalog.