How much should a PE-backed staffing portfolio spend on an AI pilot?

Spend less than you think you need. A useful pilot in one operating company runs $75K to $150K, not $500K. If the data foundation is not ready, a bigger budget just buys a more expensive failure. Prove the workflow in one office first, then fund the rollout once you have real numbers.

What data do we actually need before starting an AI project?

You need clean, connected records in three places: your ATS, your CRM, and your pay-bill system. That means consistent job titles, deduplicated candidate records, and matching client IDs across systems. If two portfolio companies use different ATS platforms with different field structures, fix that mapping before you touch AI. The model is only as good as the data feeding it.

Why do AI pilots stall after a merger?

Because two firms rarely share the same tech stack, the same job code taxonomy, or the same definition of a placement. A pilot built on one company's clean data breaks the moment you point it at the combined entity. Sequence the data integration first. Run the AI pilot second, on the merged data set, so what you prove actually scales.

The $2.1M AI Pilot Trap in PE Staffing Portfolios

PE-backed staffing portfolios spend an average of $2.1M per company on AI, and 87% of those projects die at the pilot stage. The problem is not the tech. It is fragmented data and bad sequencing. Here is how to fix both.

PE firms are spending an average of $2.1M per portfolio company on AI, and 87% of those projects never move past pilot. That is not a technology problem. That is a sequencing problem, and it is costing sponsors real returns.

I see the same pattern across staffing portfolios. A sponsor gets excited about AI. They fund a pilot in one operating company. The pilot demos well. Everyone claps. Then it dies quietly six months later when someone tries to roll it out across the portfolio and nothing works.

The money was never the issue. The order of operations was.

Why the Pilot Looks Great and the Rollout Falls Apart

A pilot runs on a small, clean slice of data. Someone hand-picks one office, one recruiting team, one clean data set. The AI matches candidates to reqs and it looks like magic.

Then you try to scale it. Now you are pointing the same tool at three or four operating companies that came into the portfolio through separate acquisitions. Each one runs a different ATS. Each one codes job titles differently. One calls it a "placement" the day the candidate signs. Another counts it the day the candidate starts.

The AI cannot read that mess. It was trained on order and it just hit chaos. The pilot did not fail. It just never touched reality until the rollout.

This is the trap. You prove something in a controlled environment, then discover the environment was the only reason it worked.

The Data-Readiness Prerequisites Nobody Wants to Fund

Data cleanup is boring. It does not demo well. No sponsor gets excited about deduplicating candidate records. So it gets skipped, and the AI budget gets spent on the shiny part instead.

That is backward. Before you fund a single AI pilot across a portfolio, three things need to be true.

Consistent taxonomy across companies. Job titles, skill codes, and industry categories need to mean the same thing in every operating company. A "Senior RN" in one system and a "RN II" in another need to map together. If they do not, your AI is comparing apples to fire trucks.
Deduplicated candidate and client records. When two portfolio companies both worked with the same candidate, you have two records. Maybe five. The AI treats them as separate people. Clean this before you scale or your matching accuracy tanks.
Connected systems, not siloed ones. Your ATS, CRM, and pay-bill data need to talk. If a recruiter has to check three screens to see one candidate's full history, so does your AI, and it cannot.

None of this is glamorous. All of it decides whether your $2.1M survives contact with the real portfolio.

The Sequencing That Actually Works

Here is the order I run with portfolio ops leaders. It is not complicated. It is just disciplined.

Step 1: Integrate before you innovate

If you acquired companies with different tech stacks, pick one system of record per function. One ATS. One CRM. One pay-bill platform. Migrate everyone onto it and map the fields as you go. This is the least fun part and the part that determines everything downstream.

Do not run an AI pilot on a portfolio mid-integration. You will prove something on a data set that will not exist in six months.

Step 2: Pilot small and cheap

Run the pilot in one operating company, on the newly integrated data, for under $150K. Pick one workflow with a clear metric. Candidate-to-submittal time. Recruiter screening hours per week. Something you can measure in numbers before and after.

Give it 90 days. If the metric does not move, you learn that cheaply. If it moves, you now have a real number to build a rollout case on.

Step 3: Scale on proof, not hope

Once the pilot shows a hard result on integrated data, roll it to the next operating company. Then the next. Each rollout is faster because the data foundation already holds.

The firms that skip Step 1 and Step 2 are the ones funding $2.1M projects that die. The firms that do the boring work first are the ones getting AI into every office in a portfolio inside a year.

What This Costs You If You Get It Wrong

Say you have a five-company staffing portfolio. You spend $2.1M on an AI initiative that stalls at pilot. That is not just $2.1M gone.

It is a year of operating leaders losing trust in the sponsor's tech decisions. It is recruiters who tried the tool once, watched it break, and now refuse to touch the next one. It is the internal story that "AI does not work here," which is expensive to unwind.

The real cost of the pilot trap is not the wasted budget. It is the credibility you burn with the people who have to actually adopt the next thing.

Do This Week

Pull a data inventory across your portfolio. One page per operating company. List the ATS, CRM, and pay-bill system each one runs. Then flag every place where two companies use different platforms or different field structures for the same function.

That single page tells you whether you are ready to fund an AI pilot or whether you have integration work to do first. Most portfolios discover they have integration work to do first. Better to learn that from a one-page inventory than from a dead $2.1M project.

If your inventory shows more red flags than green, that is your real first project. Build the foundation. Then buy the AI.

The $2.1M Pilot Trap: Why PE-Backed Staffing AI Projects Never Scale