Second-Attempt AI Buyers: Why Agencies Are Buying AI Tools Again, and What Changed

Agencies are buying AI tools a second time because the first purchase was aimed at the wrong target. A July 2026 GoodFirms survey of 144 agencies found that only 9.1% of current AI buyers are first-timers. The other 65.9% already tried an off-the-shelf tool, an internal build, or another agency, and it failed to do the job. This time they are not buying chatbots or "AI features." They are buying workflow automation with a measurable outcome attached, priced against results instead of hours. The failure wasn't the technology. It was the target.

I have sat across the table from agency owners holding a canceled invoice for a chatbot nobody used. I have watched founders explain to their board why the "AI transformation" line item produced nothing but a Slack channel full of apologies. None of that means AI doesn't work. It means most agencies bought a feature instead of a system. The second wave of buyers has figured that out, and the data backs them up.

The First Attempt Wasn't a Technology Problem. It Was a Doctrine Problem.

Here's the anecdote, because I want you to see the dollar figure, not just the concept. A client of mine, a 22-person performance marketing shop, spent $38,000 on a custom chatbot build in 2024. The pitch was clean: deflect client emails, answer FAQs, free up account managers. Eighteen months later the bot handled 4% of inbound messages and the team had quietly gone back to answering everything by hand. The owner's reaction was the one I hear most often: "I bought a chatbot. I needed a system." That single sentence is the whole second-attempt market in miniature. He didn't need a feature bolted onto his intake process. He needed his intake process redesigned so a machine could actually run part of it. Feature versus system is the whole gap.

That gap is not a one-off. S&P Global Market Intelligence found that 42% of companies abandoned most of their AI initiatives, up from 17% the year before, a trend confirmed by CIO Dive's coverage of the same research. The average organization scrapped 46% of AI proof-of-concepts before they ever reached production. And the MIT Media Lab's NANDA project, widely reported by Fortune, found that 95% of generative AI pilots produced no measurable P&L impact. Those are not small-sample flukes. Those are the receipts from an entire market that bought the wrong thing.

Watchstanding is the right metaphor here. On a ship, a watchstander doesn't get to say "I generally paid attention." He logs what he saw, when he saw it, and what he did about it. Most first-attempt AI purchases had no watch log. Nobody could point to a number and say "this tool moved that metric by this much." When the invoice came up for renewal, there was nothing to defend.

What Second-Attempt Buyers Are Actually Purchasing

The GoodFirms data draws a sharp line between what failed and what agencies are buying now. 75% of agencies say clients are prioritizing workflow automation over chatbots or generic generative AI features. That is a complete reversal from the first wave, which was sold almost entirely on novelty: a chatbot here, a content generator there, a "co-pilot" bolted onto an existing tool stack. None of it touched the actual bottleneck in the business.

Second-attempt buyers ask a different question before they sign anything. Not "can it write copy" but "which specific step in my delivery process does this remove, and how do I measure that it's gone." That is a systems question, not a features question. It is also, not coincidentally, the same question that separates an agency that compounds value from one that just burns budget on tools.

The autonomy numbers tell you how early we still are, even with sharper buyers. Only 2.3% of agencies report reaching a fully autonomous AI operating model, per GoodFirms. The overwhelming majority are running supervised or assisted setups, meaning a human still checks the output before it ships. That's not a failure. That's a casualty drill. You practice the procedure under supervision until the team trusts it enough to run without a hand on the wheel. Agencies skipping straight to "fully autonomous" on the first attempt are the ones who ended up back in this survey as second-attempt buyers.

The Pricing Shift Is the Real Story

Here is the number that should change how you think about your own P&L: 56.8% of agencies are shifting toward outcome-based pricing, according to the same GoodFirms research. That means agencies are pricing AI-driven delivery by result, not by hour or by seat. If a workflow automation tool cuts your time-to-deliverable by 40%, you don't bill 40% less. You bill for the outcome, and the AI becomes part of your margin structure instead of a cost center you have to justify every quarter.

This is the single most important structural change in the whole dataset, and most agency owners are missing it because they're still asking "what tool should I buy" instead of "what should I be charging for now that the delivery mechanics have changed." A tool is an expense. A repriced deliverable is an asset. One shows up on your income statement as a line item you defend. The other shows up on your balance sheet as a reason a buyer pays more for your agency at exit.

Buy Beats Build, and the Data Isn't Close

If you are debating whether to purchase a vendor tool or have your team build something in-house, the MIT NANDA research settles it more decisively than most owners expect. Purchased or vendor-partnered AI tools succeed roughly 67% of the time. Internally built tools succeed at roughly 22%, about a third as often, according to reporting on the same MIT study by Softed. The report's lead author put it bluntly: enterprises kept trying to build their own tools, and the data kept showing purchased solutions delivered more reliable results.

I see agency owners fall into this trap constantly. There's a founder-operator instinct that says "we're smart, we can build this ourselves and own the IP." Sometimes that's the right call, when the workflow is genuinely your moat and nobody else can build it for you. Most of the time it's ego wearing a business case as a costume. Buying the boring, proven tool and integrating it into your actual delivery system beats building a custom one that nobody maintains after the developer who built it leaves.

The Adoption-Impact Gap Hasn't Closed

Zoom out to the broader economy and the pattern holds. McKinsey's State of AI research found that 88% of organizations now report using AI in at least one business function. But only 39% report any measurable EBIT impact from it, and most of those attribute less than 5% of EBIT to AI. Adoption is nearly universal. Profit impact is still rare. That gap is exactly why the second-attempt market exists: everybody adopted something, most of it didn't move the number that pays the bills, and now they're back shopping with better questions.

The ATLAS Model: Why Point-Solutions Keep Failing

I built the ATLAS Model for Growth because I got tired of watching agency owners solve the wrong layer of the problem. ATLAS stands for Acquisition, Tech-stack, Labor, Automation, and Sales/pricing, treated as one integrated system instead of five separate line items you shop for independently. An AI tool bought in isolation, disconnected from how you acquire clients, staff delivery, and price the result, is a point-solution. Point-solutions are exactly what failed in the first wave. They automated a task nobody had redesigned the surrounding process for, so the task got faster and the bottleneck just moved somewhere else.

Under ATLAS, you don't ask "what AI tool should I buy." You ask where the system's current bottleneck sits, whether automation actually removes that bottleneck or just relocates it, and whether your pricing model captures the value once the bottleneck is gone. That's the difference between the 65.9% who are back here for round two and the agencies that got it right the first time. The tool was never the strategy. The system was.

Doctrine Connection: Systems Beat Slogans

Every failed AI rollout I've reviewed had a slogan behind it: "AI-powered," "cutting into busywork," "the future of the agency." None of them had a system behind the slogan. The ATLAS Model for Growth exists because slogans don't compound and systems do. If your AI purchase isn't wired into how you acquire, staff, automate, and price, in that order, you're not running a system. You're running a slogan with a subscription fee attached. Systems beat slogans every time the invoice comes due.

What This Means for Your Balance Sheet

An agency that treats AI as a bolt-on feature is compartmentalizing risk in the wrong direction: all the downside sits with the owner who signed the contract, and none of the upside compounds into something sellable. An agency that treats AI as part of an integrated operating system, acquisition tied to automation tied to pricing, builds something a buyer will actually pay a multiple for at exit. Buyers don't pay for tools. They pay for systems that keep producing results without the founder standing over them.

That's the real test of whether your second AI purchase will work where the first one didn't. Not "does it have good reviews." Not "did a competitor buy it." Ask instead: if I removed myself from this workflow for thirty days, does the system still produce the outcome I'm charging for? If the honest answer is no, you haven't bought a system. You've bought a slogan again, just a more expensive one.

Frequently Asked Questions

Jeff Barnes, MBA has no personal position in any company, fund, or platform named in this article. demg.ai has no current commercial relationship with any party mentioned. demg.ai provides marketing strategy and education for owner-operators, not investment advice.