The pitch deck promised transformation. The diligence session should ask for something duller: a log, a cost per outcome, and the last thing that broke. Investing in revenue capacity means funding operating capacity that earns, and judging it by operating numbers rather than promises. The category rewards exactly that posture, because unlike most software bets, this one produces weekly evidence you can inspect before and after you commit.
What does capital actually buy in an agentic practice?
Capital converts into four things, and the list is deliberately unglamorous:
- Integrations into the systems where revenue work happens: CRM, inbox, calendar, billing. The unglamorous plumbing where finished work either lands or does not.
- Evaluation and governance, the controls described in part seven: scope documents, gates, audit trails, regression tests on real past cases.
- Deployment capacity, so more systems go live for more clients sooner. This is the scaling mechanism; the unit being scaled is a working loop, not a seat license.
- Distribution, so proven workflows reach the buyers who need them. A deployment that works and is never seen earns nothing for anyone.
Deployment skill is the scarce item on that list, and the market data agrees. Keyhole's 2026 analysis of the enterprise agentic AI market finds platforms capturing most spending today while implementation services grow fastest, because deployment is the hard part (third-party analysis of a market still forming; treat the ratios as directional). In a market where platforms are abundant and deployment is scarce, capital placed on the scarce side buys more differentiated returns than capital placed on the abundant one.
Which five numbers should a deployment report?
Returns show up in five numbers, and a practice worth funding reports them weekly per deployment:
| Measure | What it tells you |
|---|---|
| Revenue influenced per system | Whether the capacity earns |
| Cost per completed outcome | How it compares with staffing the same work |
| Payback period per deployment | How fast capital returns |
| Retention of deployed systems | Whether clients keep them running |
| Override and incident rate | Whether the controls work |
Two of these deserve a note. Cost per completed outcome, not cost per seat, is the unit that makes comparisons honest: staffing the same follow-up work with people has a number, and the system has to beat it. Override and incident rate is the honesty meter: a deployment reporting zero incidents is either not running or not reporting, and both disqualify the other four numbers.
What are the real risks?
Results depend on data quality and on how well agents are configured around a company's own sales motion, a caveat Outreach itself states about its own results. The field moves fast, so architectures age in public. Adoption inside client teams takes effort, because a system nobody steers is shelfware with a log. And any system built on third-party models depends on those providers: pricing, terms, and capabilities change on someone else's schedule, which is why systems should be designed so the model can be swapped without surgery.
None of these risks argues against the category. They argue for pricing it honestly, with a review layer budgeted like payroll, a model-swap seam designed in from the start, and a weekly reporting cadence that surfaces drift while it is still cheap.
What should you ask before committing?
Four questions, asked before the term sheet, not after:
- Show me a live system and its logs.
- What was your last failure, and what did you change?
- What does a deployment cost you, and what does it earn?
- What happens if a model provider changes its terms?
A practice with real operations answers all four from memory, with artifacts. The fourth question sorts the field faster than any other: teams that have never thought about provider dependency have built a demo with a revenue glossary.
What this is not
This is not a claim that every deployment earns, or that operating numbers replace judgment about a team. Numbers describe the machine; they do not evaluate the people steering it, and the same five figures can describe a practice worth scaling and one worth stopping, depending on trajectory. And it is not a framework for valuing platform companies. It prices one thing only: the operating capacity a specific practice runs, today, in production.
The move
Demand the five numbers weekly from anything you fund or buy, and attach a condition to the ask: the numbers come with log access. Then verify one number by hand against the log, once. The hour it takes calibrates every conversation afterward, because you will know whether the reporting and the operation are the same object.
Take-home
- Put the five measures into the diligence template, weekly cadence, logs attached, before the next opportunity closes.
- Ask the four questions in order, and treat a fast, specific answer on the failure question as the strongest single signal in the set.
- For any system you fund, confirm the model-swap seam exists by asking which component changes if pricing triples tomorrow.
FAQ
Is this a pitch for investing in agencies?
No. The frame applies to any structure, equity, revenue share, or fixed-fee builds, because the thing being judged is the same in all of them: whether deployed capacity earns more than it costs. The five numbers are structure-neutral by design.
Which of the five numbers matters first?
Retention of deployed systems. A system clients keep running is the ground truth that the other numbers are real; churn makes even good revenue figures a treadmill. Payback and revenue influenced follow, and the override rate audits the whole story.
How do you compare deployments across clients?
Normalize by the client's own baseline: the unworked-work cost from [part one](/insights/you-are-buying-revenue-capacity), and the staffing cost from [part five](/insights/capacity-is-the-unit-of-growth). Percentages against those two baselines travel between clients; raw figures do not.
What makes a practice defensible at scale?
The operating record, the governance layer, and the deployment method that reproduces both. Models get commoditized; a year of logs across live client systems does not. That is why the record, not the roadmap, is the asset.
Related reading
- Governance Is the Moat
- The First Ninety Days
- You Are Buying Revenue Capacity, Not a Tool
- Small Teams, Real Owners, Fast Executives
- The New Venture Stack: Small Teams and Agent Fleets
If you are weighing a commitment and want the diligence done with logs on the table rather than slides in a deck, book a discovery call.