Technology function · Private equity
built, proven, and handed over.
A portfolio company between $10M and $100M of revenue carries the technology dependencies of a much larger business and has nobody to run them. The function is usually outsourced to an MSP whose job is keeping the lights on, not building anything.
So nothing gets built. Every initiative — AI or otherwise — arrives at the same four constraints.
AI has made this visible rather than different. Everyone adopted chat; almost nobody moved past it, because the constraint was never a shortage of ideas. Pick up any single use case and it's an iceberg — the automation is the tip.
An agent gives you an answer that's probably right.
An engineered workflow gives you the same answer every time.
That distinction is fine to ignore for research and drafting. It is not fine in finance, where the tolerance for error is zero. The common failure is building an agent where the business needed a workflow — it demos beautifully, and then doesn't survive contact with a CFO.
So we default to code. Deterministic, testable, and owned by the portco afterwards. The AI is in how fast it gets written, not in what answers the question at runtime.
Every engagement has a fixed price and a fixed date — the numbers are in the menu
Below is the scope, the clock, and where each one stops. These are the ones we're asked for most; the menu goes further.
Four stages, in the order they're usually bought. You don't have to start at the first — but each assumes the ones above it are true.
Architecture, scalability, technical debt, team, spend, and what the first hundred days would cost.
Where it stops: We assess, we don't remediate. Fixing what we find is a separate engagement with its own number and date.
Roadmap credibility, product-market fit, and the question every software asset now faces: does AI make this product a moat or a commodity? A buyer will ask that at exit. Better to answer it at entry.
Where it stops: An assessment. We don't build the roadmap we recommend — that's a separate engagement with its own number and date.
What the technology function costs, what it's worth, and a hundred-day plan with sequencing.
Where it stops: The plan, not the execution. We'll tell you which parts we'd run and which you should give to someone else.
Source systems replicated into a warehouse before deciding exactly what it's all for — because insight starts as soon as the data is out, well before everything is linked up. We surround the ERP rather than replace it: replacing one is usually a twelve-month project that outlives its own justification.
Where it stops: The source systems named in the pre-flight inventory. Anything discovered later is a second sprint with its own number and date — not a change order against this one.
Data enrichment, mapping and vision models chained together to do in hours what a research team does in a week. The value is in deepening relationships with accounts you already know, not adding leads.
Where it stops: We build the machine, we don't run your sales team. And we don't write your ICP — if that's unresolved, this is the wrong engagement.
Month-end close, consolidation across entities, AP, AR, board pack assembly, variance analysis. Built as workflows, for the reasons above.
Where it stops: We automate the process, we don't become the finance function. No controller role, no audit, no sign-off.
Use cases into production beyond finance — operations, service, procurement. Deterministic wherever the tolerance for error is low. Plus the governance framework that survives diligence.
Where it stops: Production, not research. If a use case can't be made deterministic and the tolerance for error is low, we'll say so rather than build it.
AI in the product your customers use, not the operations behind it. The same rule applies: deterministic where the tolerance for error is low, probabilistic only where a wrong answer is survivable. AI is compressing the economics of a lot of legacy software — the question is rarely whether to respond, it's whether the response ships before the next diligence.
Where it stops: We build the capability and hand it over. We don't take ownership of your product roadmap.
Any team with engineers benefits from the current AI coding tools, and most have no idea what's currently possible. This holds whether or not the company sells software — a distribution business with four developers gets the same multiple as a platform with forty. Tooling, working practices, and the review discipline that stops speed becoming a liability. Different lever from AI Factory: that one builds workflows for the business, this one makes the business's own engineers faster.
Where it stops: We make your team faster, we don't become your team. No staff augmentation, and we don't take the roadmap off your CTO.
Board and IC reporting, roadmap and architecture ownership, hiring and building the team that eventually replaces us, vendor and spend discipline.
Where it stops: One to two days. Not an escalation path at 2am, and not the person doing the building — that's what the sprints are for.
Product leadership for assets that need the judgement and can't yet justify the hire. Roadmap ownership, pricing and packaging, and the discipline of shipping what was promised to the board.
Where it stops: One to two days, same as the CTO seat. Not the person writing specs for every feature.
The same two engines, pointed at your own operation rather than an asset's. Not a side business: a sponsor asking a portfolio company to automate its close has an easier conversation having automated its own.
GTM Engineering, pointed at dealflow. For a fund, origination is go-to-market — the targets are companies rather than customers, the intermediaries are bankers rather than channel partners. Same machinery, and every inferred fact carried as a sourced claim with a link back to where it came from.
Where it stops: We build the pipeline and the provenance. Commercial data licences stay in your name and on your bill — we're not reselling anyone's data.
Every PE firm runs on three ledgers: the GL, the investor record (transactions, transfers, capital accounts) and the asset record (marks, cashflows). Almost every question a fund answers — a capital account statement, a quarterly report, an LP request — is a join across those three, done by hand. We replicate all three, reconcile them, and build the workflows on top. Deterministic, because a capital account statement that is probably right is worthless.
Where it stops: We automate what your team does. We don't replace your fund administrator and we don't sign anything.
Six weeks is a claim until you can see where it goes
Most failed data projects die in week 0 and call it week 4. The pre-flight exists so that ours dies before you've paid for it.
The first engagement in a portfolio pays for work the next three don't repeat. A close process is a close process whether the asset is a healthcare roll-up or an industrials business.
So the second engagement costs less than the first, and the fourth costs less than the second. That compounding is the product. It's why the firm is called PortfolioLevel.
It travels sideways too. The research machinery built for a portfolio company's pipeline is the same machinery that builds the fund's own view of the market.
Five commitments
We would rather lose the work than take the wrong engagement
We are early, and we would rather say so than have you work it out.
The first two engagements are offered at a third off standard pricing. That is design-partner pricing: we are buying reference cases, and we would rather buy them with cash than with corners. The scope, the clock and the seniority are the same as they will be at full price.
Contact