We bought an AI sourcing tool six months ago and nothing changed. What went wrong?

You bolted the tool onto a workflow that was built for people. The recruiter still does the old steps, plus a new one to check the AI. That adds work instead of removing it. Fix the workflow first. Decide which steps the AI owns end to end, then delete the human steps it replaces. If a recruiter still touches every candidate the AI surfaces, you did not redesign anything. You just added a tab.

How do we measure AI ROI when it touches so many parts of the process?

Pick one metric per tool before you turn it on. For sourcing, measure qualified candidates per recruiter per week. For screening, measure time from application to first human touch. For a scheduling bot, measure interviews booked without a coordinator. One number, one baseline, one review date 60 days out. If you cannot name the number the tool should move, you are not ready to buy it.

Should we wait for AI tools to mature before we invest?

No. Waiting is not the problem and buying is not the problem. Firms that get ROI are not using better tools. They redesigned the work around the tools they already have. Start with one workflow you run hundreds of times a week. Redesign that. You will learn more in 60 days of real use than in a year of watching the market.

Why Staffing Firms Buy AI But Get No ROI

Nearly every staffing firm has bought AI. Almost none can point to earnings from it. The gap is not the tool. It is the workflow redesign step that gets skipped every time.

Nearly 80% of companies deployed generative AI in 2025, and most reported no material impact on earnings. McKinsey calls it the gen AI paradox. In staffing, the numbers are worse. Around 92% of staffing leaders have adopted some form of AI, yet almost none can point to a line on the P&L and say the AI put it there.

The money is not lost because the tools are bad. It is trapped between the tools. Firms buy a sourcing engine, a screening bot, a scheduling assistant, and a resume parser. Then they hand all four to recruiters who still run the same process they ran in 2019. The AI produces output. The output goes nowhere. That is shelfware with a login.

The step everyone skips is workflow redesign. Not training. Not change management slides. The actual rebuilding of who does what, in what order, and which human steps get deleted.

Why the Tool Never Delivers on Its Own

Here is what a normal AI rollout looks like at a staffing firm. Leadership buys a sourcing tool. IT sets it up. Someone runs a lunch-and-learn. Recruiters get a login and a nudge to use it.

Now watch what a recruiter actually does. She opens the AI tool. It surfaces 40 candidates. She does not trust it, so she opens LinkedIn and her ATS and checks the candidates herself. She just did her old job plus a new one. The tool added a step instead of removing one.

Multiply that by every recruiter and every tool. You did not speed up the work. You gave your team more screens to check. That is why the earnings never move. The AI is running. The workflow is not.

ROI shows up only when the AI owns a step from start to finish and a human step disappears. If the recruiter still touches every candidate the tool surfaces, you bought a suggestion box, not a worker.

The Redesign Step That Separates Gains From Shelfware

Redesign means you sit down with one workflow and rebuild it around the tool. Sourcing is a good place to start because you run it hundreds of times a week. Small gains compound fast.

Walk through the four questions that make redesign real.

Which step does the AI own completely? Not assists. Owns. Example: the AI sources and ranks candidates, and no recruiter opens LinkedIn to double-check the top 10. If you cannot say a step is fully owned, you have not redesigned it.
Which human step gets deleted? For every step the AI takes, a person stops doing something. If nobody stops doing anything, you added cost. Write down the deleted step. Make it official.
Where does the human re-enter? AI hands off to the recruiter at a clear point, like the first live conversation. Draw that handoff line. Everyone should know exactly where the machine stops and the person starts.
What number proves it worked? Pick one metric before launch. Qualified candidates per recruiter per week is a good one for sourcing. Set the baseline today. Check it in 60 days.

Notice what is missing from that list. No new software. This is the free part, and it is the part firms skip because it is uncomfortable. Deleting a human step means someone changes how they work. That is where rollouts stall, and it is the whole reason the Build. Change. Adopt. framework exists.

Build. Change. Adopt. Is Where the ROI Lives

Most firms nail Build and quit. They buy the tool, wire it up, and call the project done. Build is the easy 20%. The tool works on day one.

Change is the redesign. It is the four questions above. It is deleting steps, redrawing handoffs, and telling a recruiter that the thing she did for eight years now belongs to a machine. This is hard, and it is where the money is.

Adopt is what makes it stick. A recruiter will quietly go back to the old way the first time the AI surfaces a weak candidate. That is human. If nobody watches the metric and coaches the team through the rough patch, the old workflow creeps back in 30 days. The tool keeps running. The people stop using it the way you designed.

I have watched a $40M firm buy a strong AI screening tool and get nothing for six months. We did not swap the tool. We rebuilt one workflow. The bot screened every inbound applicant and booked the first phone screen with no coordinator. We deleted the coordinator step for that req type. Time from application to first human touch dropped from 26 hours to under 2. Recruiters got their mornings back. Same tool. Different workflow.

That is the whole game. The firms getting ROI are not using better AI. They redesigned the work around the AI they already own.

What This Costs You If You Skip It

Skipping redesign is not neutral. You pay for the license every month. You pay for the recruiter time spent checking output nobody trusts. And you pay in credibility, because after two dead rollouts your team stops believing the next tool will work either.

PE-backed portfolios feel this hardest. The deck said AI would cut cost per placement. The tools are deployed across every portfolio company. The cost per placement did not move. The board wants to know why. The answer is always the same. Nobody redesigned the workflow. They bought the Build and skipped the Change.

Do This Week

Pick one workflow you run at least 100 times a week. Sourcing or first-touch screening are the best candidates. Block 90 minutes with the recruiter who runs it most and the person who owns the tool.

Answer the four questions on a single page. Which step does the AI own. Which human step gets deleted. Where does the human re-enter. What one number proves it worked. Write the baseline for that number today.

Set a review date 60 days out and put it on the calendar now. That single page is your redesign. Do it once, watch the number move, and you will never buy another tool the old way again.

Why Staffing Firms Keep Buying AI and Not Getting ROI