Every board wants AI results. Most companies won’t get them and the reason has far less to do with the models than with what sits underneath them.
By Ed Pederson VP of Commercial Operations, PitchMe
There isn’t a board of directors on the planet right now that isn’t asking its leadership the same question: what are we doing with AI, and how is it going to change the business in a tangible way? They’re after the familiar things: efficiency, incremental revenue, a real competitive edge. And the market has answered with enthusiasm, pouring tens of billions of dollars into AI and automation tools.
Yet only a small fraction of those investments by most estimates (somewhere around five percent) ever produce measurable ROI. That gap is worth sitting with, because it isn’t a story about technology that isn’t ready. The models are remarkably capable. It’s a story about everything that surrounds them.
Here is what I’ve learned after fifteen years buying and deploying technology on the operator side and now building on the vendor side at PitchMe.
Bad data in, bad AI out
Start with a simple truth about how AI actually works. It is exceptional in the middle of a process, but it still needs someone to hand it good inputs and someone to act on its outputs. Remove either bookend, or feed it bad data, and it will fail no matter how strong the underlying model is.
Most of the disappointment I see traces back to exactly this. Organizations bolt AI onto processes that were broken to begin with, then act surprised when the outcome is broken too. Worse, they run it on stale, incomplete data: the wrong phone number, the dead email address, the candidate who moved states six months ago.
The data intelligence layer isn’t a nice-to-have. It’s the thing that decides whether the AI on top of it works at all.
This is why the data layer is so tantamount to getting value out of AI. If you deploy a tool to run candidate outreach, its usefulness depends entirely on whether it can actually reach the person on the other end. That’s the problem PitchMe exists to solve: keeping candidate profiles accurate and current in real time so whatever you build on top of that data has a genuine chance of working. Get the data right, reshape the process around it, then let the loop feed itself. That sequence is what produces the results boards are actually asking for.
Give ROI a single owner
When I ask companies who is measuring the return on their AI investments, the honest answer is usually “several people, none of them completely.” Operations owns throughput. Finance owns revenue and cost. Everyone sees a slice; no one sees the whole picture.
That fragmentation is the enemy of ROI. You need one operational owner, someone with real accountability for a P&L in that space, who can trace the entire path from the moment a lead arrives, through sourcing, to placement, to revenue. Only that person can see the true cost of delivering talent to a customer, and only that person can pull the levers to bring it down.
And there are real levers to pull. Better data at a lower cost means you are re-engaging talent you already know. That can translate into lower recruitment marketing spend, or fewer LinkedIn Recruiter seats. Someone has to be watching those dials and pulling the levers, otherwise the savings the technology makes possible never actually show up on the statement.
Why great tools still stall on adoption
There’s a recurring pattern where the board bets big on a tool and the team barely touches it. Two dynamics usually explain it.
The first is distance. When a decision is driven purely from the top down, the people making it are often disconnected from how the work actually gets done. Someone at the top gets sold a whiz-bang piece of technology, and the people on the front line needed something entirely different.
The second is fear. AI can be genuinely unsettling if your job is the one it might change. Will this make me redundant? Is my manager going to use it to shrink the team? Those questions are real, and pretending they aren’t, guarantees low adoption.
My answer to both is to stay human-forward. Solve a real pain point for the individual first, so they experience the tool as a benefit rather than a threat. Then build the path for them to come along: redefine roles (e.g., “AI-enabled recruiter”), make the new expectations and the upskilling path explicit, and in many cases pay people more for the higher-value work. There is a genuine equilibrium here, where the human workforce moves forward alongside the technology instead of being left behind by it.
It’s worth remembering how new all of this is. Anthropic published research not long ago measuring how capable AI has become at knowledge work, and the ceiling is enormous. But when they looked at how much of that capability is actually implemented inside organizations, it was a small fraction. That is the gap. Pick where to start, bring people with you, and reinvent the process to be enabled by the tool rather than threatened by it.
A disciplined way to choose technology
The AI space is loud. You wake up to a new release every morning, and it is tempting to chase every shiny penny. Discipline is what separates the companies that get value from the ones that accumulate a zoo of half-used tools.
It starts with strategy… a digital strategy that flows from a business strategy. Maybe you want to be more competitive, enter new markets, or hit a specific EBITDA target. That is your north star. From there, look at your core stack: in staffing, that’s usually an ATS or CRM such as Bullhorn, Avionté, or LaborEdge. Then look at the integration layer and what plugs in cleanly and matches your ambitions.
Recruiting is fundamentally the same everywhere: a customer has a need, you do an intake, create a job description, conduct a search, shortlist candidates, interviews, and then offer. So, the real question isn’t “what can this technology do?” it’s “where are our actual pain points?” I’ve made the mistake of jumping straight to the end, dazzled by everything a platform could theoretically do, only to find the organization, the recruiters, and even the candidates weren’t ready for it. Solve the real pain point and adoption follows naturally.
Say the word “AI” and fifty vendors across the room will claim they can do it, so narrow the field. One of the fastest ways is to work from your core platform’s marketplace where those partners have already been vetted and integrated. Then run a focused RFI or RFP against the things that genuinely matter to you, cut to two or three finalists, and (this is the part people skip) try before you buy. Give them a pilot. Make them prove it.
Think of the stack like a Lego system: a solid platform lets you plug technology in and pull it out as your needs, budget, and the tools themselves change.
You cannot be latest-and-greatest across every category in real time; nobody has the budget for that. Pick what matters, do your diligence, and accept that it will change. If something tempting appears mid-implementation, you had better have a serious business case, because the sunk cost of what you already started is real and the experience you gain from finishing a pilot is itself valuable.
Red flags to watch in a pilot
“It’s on my roadmap.” Translation: it’s an interesting idea, we haven’t built it, but we don’t want to say no. Those rarely materialize. I want demonstrable experience, not aspiration.
A failed data privacy assessment. When you’re dealing with AI and sensitive data, run a structured privacy review. If a vendor can’t pass it, that is a hard stop.
A weak grip on AI risk and compliance. Does the vendor actually understand the regulatory environment they operate in, and can they show you the certifications to back it up?
Performance they can’t defend. Set clear thresholds up front for whatever your measure is and hold the vendor to what they sold you. If there’s a legitimate reason a target was missed, that’s a conversation worth having; there’s often real nuance. But the burden is on them to deliver what they promised.
Breaking the LinkedIn Recruiter habit
Reducing dependency on LinkedIn Recruiter is the multi-million-dollar question for a lot of firms, and the answer isn’t to take the tool away, it’s to give recruiters a better alternative. They went to LinkedIn because that’s where the most current talent information lived. Change that, and you change the behavior.
It starts with your talent purpose. If you genuinely want lasting relationships with the same candidates (e.g., “we’ll place you for six months, and we’ll have your next role lined up before this one ends.”) then your systems must support it. This is where AI does the grunt work: keeping your ATS or CRM continuously updated so you know the moment someone comes off assignment, earns a new certification, moves states, or changes numbers. Do that consistently and your own system of record becomes the first place recruiters shop, with LinkedIn filling the gaps around the edges rather than sitting at the center.
Make IT a partner, not a gatekeeper
None of this works without IT, and the best IT leaders I’ve worked with don’t gatekeep, they set guardrails. Let your team experiment with large language models, but inside the enterprise version, with clear dos and don’ts. Give people room to build their own agents, but in a firewalled environment with the right boundaries. And monitor usage, because you will need to retire tools over time: some hit a compliance wall, some simply stop being needed.
When it works, it’s a real meeting in the middle: “You know the business, I know the systems, now let’s write the rulebook together.” The more each side embraces the other, the faster and more creatively the whole organization moves. And when the dynamic is less than ideal (e.g., when leadership imposes tools that users won’t adopt) the way through is data. Show the business metrics. Show what adoption actually looks like. I’ve never met an IT person who doesn’t speak data, and nine times out of ten, once you put the numbers in front of them, they come around.
The bottom line
The companies that land in the five percent won’t be the ones with the flashiest tools. They’ll be the ones that got the fundamentals right: clean, current data beneath the AI; a single owner accountable for the return; a modular, disciplined approach to what they buy; and a human-forward path that brings their people along.
At PitchMe, we think of ourselves as that data intelligence layer. The real-time foundation that makes everything you build on top of it actually work. Get the data right, and the rest of the AI conversation gets a great deal more productive.
Why not start there?
Ed Pederson is VP of Commercial Operations at PitchMe, the AI-powered candidate data enrichment platform built for staffing and recruiting firms. To see what real-time, accurate candidate data can do for your AI stack, visit pitchme.co.