Should each portfolio company pick its own AI tools?

No. That is the trap. When seven firms each choose their own tools, you get seven vendor contracts, seven security reviews, and zero shared learning. Set the operating model at the fund level. Let each firm run its own reqs, but standardize the AI layer underneath. You buy once, you train once, and every win at one firm shows up at the others in weeks instead of never.

How long before a shared AI operating model shows up in EBITDA?

Plan for 6 to 9 months to see it in the numbers. The first 90 days is picking the shared stack and running it at one or two firms. Months 4 through 9 is rolling the proven playbook to the rest. Most funds see the first measurable margin lift in recruiter productivity and cost-to-fill before the second quarter of deployment closes.

What if my portfolio companies use different ATS platforms?

That is common and it does not block a shared model. You do not need one ATS to share an AI layer. You need one standard for how AI plugs into each ATS, how data is handled, and how results get measured. Pick tools that connect to the systems you already own. Standardize the process and the metrics, not the underlying software.

The PE Portfolio AI Trap That Kills EBITDA Gains

FTI Consulting's 2026 data shows only 7% of PE-backed portfolio companies have AI running in daily operations. Funds treating AI as separate per-company projects are leaving 200 to 400 basis points of EBITDA on the table. Here is why one shared operating model beats seven pilots every time.

Only 7% of PE-backed portfolio companies have AI fully integrated into day-to-day operations. That number comes from FTI Consulting's 2026 data, and it should stop you cold. Most funds have been talking about AI for two years. Almost none have it running where the work actually happens.

Here is the part that costs you money. The funds that treat AI as a separate tech project inside each portfolio company are leaving 200 to 400 basis points of EBITDA expansion on the table. Not chasing it. Leaving it.

If you own or advise a staffing portfolio, this is the difference between a good exit multiple and a great one. And the mistake that causes it is one most funds make without noticing.

The trap: seven firms, seven pilots, zero leverage

Picture a portfolio with seven staffing firms. Each one has a smart ops leader. Each one reads the same headlines about AI. So each one goes and does something.

Firm A pilots an AI sourcing tool. Firm B builds a resume-screening bot with a different vendor. Firm C signs a contract for AI note-taking in interviews. Firm D is still in a committee about it. By the time you look up, you have five vendor contracts, four security reviews, and three ops leaders who all learned the same lesson the hard way, separately.

That is the trap. You paid for the same mistakes seven times. You bought the same capability at five different price points. And not one win traveled from one firm to the next.

This is what happens when AI gets treated as a per-company IT decision instead of a fund-level operating decision. The math does not work. Staffing runs on the same core motions everywhere. Source, screen, submit, place, redeploy. When seven firms solve those motions in isolation, you throw away the one thing a portfolio gives you: scale.

Why firm-by-firm quietly kills your margin

The cost of the firm-by-firm approach is not just the wasted software spend. It is the compounding that never happens.

In a shared model, when Firm A figures out how to cut cost-to-fill by 18% with AI sourcing, that playbook ships to the other six firms in weeks. Six firms get the gain without paying for the learning. That is compounding. That is where the 200 to 400 basis points live.

In the firm-by-firm model, Firm A's win stays at Firm A. Nobody else knows it happened. Firm E is still writing a business case for the thing Firm A proved nine months ago. You are running seven separate clocks when you could be running one.

Here is what the firm-by-firm approach actually costs you:

Duplicate spend. You license the same category of tool five times instead of negotiating one portfolio-wide deal at a better rate.
Duplicate risk. Five security and data reviews, five sets of vendor contracts, five ways your candidate data could get handled badly.
Lost learning. Every firm relearns what another firm already knows. Nobody teaches the group.
Slow rollout. The winner in one firm never becomes the standard across the portfolio, so your best result stays your rarest result.

None of this shows up as a line item. That is why it survives. It hides inside a hundred small decisions that each looked reasonable on their own.

The fix: one operating model, not seven projects

The funds getting the EBITDA lift do one thing differently. They set the AI operating model at the fund level, then let each firm run its business on top of it.

That does not mean one ATS for everyone. It does not mean stripping local leaders of control over their reqs and their clients. It means the AI layer underneath is shared, standard, and owned by the fund.

A shared operating model has four parts:

1. One standard stack

Pick the core AI capabilities the whole portfolio will use. Sourcing, screening, and candidate matching are the obvious first three for staffing. Buy them once. Negotiate portfolio pricing. You will pay less and get better support because you are a bigger account.

2. One data and security standard

Decide once how AI touches candidate and client data. Run one security review, not seven. This is the thing that keeps you out of trouble and it is far cheaper to do centrally than to fix firm by firm after something breaks.

3. One measurement standard

Define the metrics that prove it is working before you deploy anything. Cost-to-fill, time-to-submit, recruiter productivity, and redeployment rate are the ones that hit EBITDA. If every firm measures the same way, you can compare, rank, and copy the winners. If they all measure differently, you are guessing.

4. One rollout playbook

Prove the model at one or two firms first. Then ship the proven playbook to the rest. Do not run seven simultaneous experiments. Run one, get it right, then multiply it.

This is the Build, Change, Adopt framework applied to a portfolio. Build the shared model once. Change how the firms operate on top of it. Make the adoption stick so the gain shows up in the numbers and stays there.

What this looks like in real numbers

Say your portfolio does $400 million in combined revenue. A 300 basis point EBITDA improvement is $12 million. That is not a rounding error. That is a materially better exit.

The firm-by-firm approach might get you 50 basis points at two firms and nothing at the rest. The shared model gets the full gain across all seven because the winning playbook travels. Same tools, same effort, wildly different result. The only variable is whether you decided at the fund level or let seven firms decide alone.

The 7% who have AI fully in operations are not smarter than everyone else. They stopped running pilots and started running one model.

Do this week

Pull a list of every AI tool your portfolio companies are currently paying for or piloting. Ask each firm's ops leader two questions: what tool are you using, and what has it done to your cost-to-fill. You will almost certainly find duplicate spend and zero shared measurement inside an afternoon. That list is your business case for a shared operating model, and you can build it before Friday.

The PE Portfolio AI Trap: Why Deploying AI Firm-by-Firm Destroys the EBITDA Gains You're Chasing