Lightbloom AI vs a consultancy vs an AI agency vs hiring in-house: which one actually changes the P&L?
A consultancy is paid for the plan, an agency for the build, a hire for their time. The option that changes the P&L is the one that starts from how the company actually works.
The option that changes the P&L is the one that starts from how the company actually works, not from a plan, a build or a hire. A consultancy is paid for the plan, an agency for the build, and an in-house team for its time, and each is the right choice in a specific situation. Lightbloom AI starts every engagement with Transcript and the world model it builds, which is why it is worth considering and why it is not for everyone.
Most AI spend is committed before anyone knows whether it worked
BCG's survey of 1,000 executives (BCG, 2024) put companies yet to show tangible value at 74 percent. MIT NANDA's 2025 study, as reported by Fortune (2025), found that about 95 percent of pilots had little or no measurable P&L impact.
Every one of those projects was paid for. In three of the four models below, the invoice arrives before the result does.
A consultancy is the right call when the answer is a decision
Hire a consultancy when the deliverable is a decision: which business to exit, how to restructure a function. You are paying for pattern recognition across many companies.
The engagement ends at the recommendation, and the fee was earned when the deck was accepted. If the plan does not fit how the work actually runs, you find out after they have left.
An AI agency is the right call when you already know exactly what to build
An agency is a good buy when the scope is precise: a document pipeline for one named workflow, say. The MIT researchers, in Fortune's account, found that purchasing from specialised vendors succeeded about 67 percent of the time, while internal builds succeeded a third as often.
An agency builds what it is asked to build. Choosing the process is your job, and the base rates above say that is where programs go wrong. The project is done at go-live; nobody verifies the P&L effect, and running it is your problem from that day.
Hiring in-house is right when AI is the product, and slow when it is the tool
If AI is what you sell, hire. If it is a way to run finance or operations cheaper, the maths changes. Among EU enterprises that considered AI and did not adopt it, 71 percent cited a lack of relevant expertise (Eurostat, 2025). Deloitte's survey of 3,235 senior leaders (Deloitte, 2026) names insufficient worker skills as the biggest barrier to putting AI into existing workflows. In the same survey, 74 percent hope AI will grow revenue and 20 percent see it doing so already.
In a mid-sized company the first hire has to be strategist, engineer and change manager at once. Payroll starts on day one. The verified number, if it comes, comes quarters later.
The four options side by side
| Consultancy | AI agency | In-house hire | Lightbloom AI | |
|---|---|---|---|---|
| What you pay for | A recommendation | A specified build | Salaried time | Value your team has verified |
| When you pay | Monthly, from week one | Milestones, from kickoff | Payroll, from day one | After verification; no upfront fee, no retainer |
| Who decides what to build | You, from their recommendation | You, before the brief | The hire and their sponsor | Together, from discovery; every opportunity gets a number first |
| Who builds | Usually a separate firm | The agency | The hire | Lightbloom AI's engineers, with your team |
| Who stays to run it | Your team | Your team | The hire, if they stay | Lightbloom AI, until it runs; Maintain phase optional |
| How value is verified | Rarely; the report is the deliverable | At go-live, if at all | Self-reported by the builders | Your team counts it, before any invoice |
| Best for | Strategic decisions | A precise scope | Companies where AI is the product | Companies from about $20M revenue that want P&L change without carrying the risk |
| Not for | Getting things built and run | Choosing what to fix | Results needed this year | A report, or a build without verification |
Our view
Our view, and a CFO should press us on it: the market is structured so that everyone else gets paid whether or not the P&L moves. Seventy-four percent of companies with no tangible value is a lot of paid invoices.
The alternative is to start with the world model. Transcript hears everyone in the company, writes up every problem with a number, and the company chooses the visions worth building before anything is built. The partner then builds with the company's team and stays until it runs. How that works, in full.
Questions people ask next
Why not a consultancy for strategy and an agency for the build?
Many do. You then own the gap: the plan that does not fit the real process, and the build nobody verifies.
Does paid-from-value mean you only pick easy wins?
It means we pick wins that survive your finance team's count. Often unglamorous ones in finance and operations, because those are the ones that move EBITDA.
What if we already have an internal AI team?
Then discovery is usually what they are missing, and it is where we start. Verification stays with your finance team either way.
What size of company does this work for?
From about $20M in annual revenue. Below that, a well-chosen agency build is usually the better buy.
Lightbloom AI is the option here that starts from the company itself: the world model first, then the visions the company chooses, built with its team. The full picture is at /service-offering.
References
- BCG, AI Adoption in 2024: 74% of Companies Struggle to Achieve and Scale Value (2024), https://www.bcg.com/press/24october2024-ai-adoption-in-2024-74-of-companies-struggle-to-achieve-and-scale-value
- Fortune, MIT report: 95% of generative AI pilots at companies are failing, reporting on MIT NANDA, The GenAI Divide: State of AI in Business 2025 (2025), https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/
- Eurostat, Use of artificial intelligence in enterprises (2025), https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises
- Deloitte, State of AI in the Enterprise, 2026 report (2026), https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html
Keep reading.
All notes- How do you capture tribal knowledge before someone leaves? Tribal knowledge is how the company actually runs, held by the people who run it. Why handovers fail, what works, and why it is the first layer of a world model. Read →
- Why AI agents need a world model of your company AI agents stall inside companies because nobody has described how the work runs. What a company world model gives an agent, and why to build it before you deploy. Read →
- What is a company world model? A company world model is a living picture of how a company really works: every process as it runs, what hurts most, and who everything depends on. Why AI agents need it. Read →