Choosing an AI consulting firm.
Ask who runs it after launch.
Much AI consulting ends at a recommendation or a launch. The ten questions below separate firms whose AI is still in use a year later from firms that leave you a deck or a demo. We wrote this as operators, and we say plainly where we fit and where we do not.
To choose an AI consulting firm, judge three things first: production proof, who does the work, and who runs the AI after launch.Production proof means systems still in use a year later, with adoption numbers, not a portfolio of pilots. Who does the work means the seats those people have held at companies your size. Who runs it after launch matters because models turn over roughly every year and a vendor can retire one with as little as 60 days' notice. Then check speed to first value, what the firm diagnoses before it builds, how it drives adoption, its governance guardrails, who owns the IP, and whether the first engagement is scoped and priced before it starts.
Four kinds of firm you will meet.
Each is the right answer for someone. Know which one you are talking to.
Strategy and consulting firms
You get: A recommendation, a roadmap, a deck.
The catch: Execution and risk stay with you. The deck ages the day it is delivered.
AI engineering and development shops
You get: A working build, often fast.
The catch: They build on whatever operating model exists, broken or not, then hand over the keys.
Platform vendors' services teams
You get: Deep knowledge of one product.
The catch: Every answer is their platform, whether or not it fits the job.
Operator-led advisory
You get: The operating model fixed first, then the AI built and managed after launch.
The catch: Fewer firms work this way, and it asks more of your leadership team at the start.
What to ask, and what a good answer sounds like.
- 01
Show me AI you built that is still in production a year later. What is its adoption rate?
Named systems, real adoption numbers, and a straight answer about the ones that did not last.
- 02
Who will actually do the work, and what seats have they held?
The people on your account, with operating experience at companies your size, not only the partner in the pitch.
- 03
What will be measurably different in 30 days?
A specific metric and a specific date. Months of discovery before anything changes is a warning.
- 04
What do you check before you build?
Decision rights, ownership, the workflow the AI plugs into, and the data. Automating a broken process scales the problem.
- 05
Who owns the operating decisions the AI depends on?
A named owner on your side, agreed before the build starts.
- 06
How will you get people to use it?
Users inside the build from the start, trained before cutover, and adoption measured at 90 days.
- 07
What guardrails does every agent have, and who approves changes?
Written limits on what each agent can and cannot do, with people approving every change.
- 08
Who owns the IP and the code?
You do, with no licensing strings on what was built for you.
- 09
What happens when the model we built on is retired?
A plan and an owner. Models turn over roughly every year, and a vendor can retire one with 60 days' notice.
- 10
How is the work scoped and priced before it starts?
A defined first engagement with its price agreed up front, not an open-ended meter.
Where KeyDelta fits, and where we do not.
KeyDelta is operator-led advisory: a Managing Partner and six Principals who, between them, have held the CEO, COO, and CRO seats. We advise, build, and manage.
A good fit
- Companies from $50M to $1B in revenue, including divisions of larger enterprises
- AI pilots that stalled, or AI that shipped and nobody uses
- An operating model that has to be fixed before AI can pay back
- Leaders who want the AI built and kept current, not just recommended
Probably not us
- You want a strategy document and plan to execute it yourselves
- You need contract engineers to staff a team you already run
- You have committed to one platform and want its implementation partner
- The work is a single tool configured for a small team
See how we work across AI advisory, AI implementation, AI agent development, and managed AI services.
Common questions
How do I choose an AI consulting firm?
Judge three things above all: production proof (AI the firm built that is still in use a year later, with adoption numbers), who actually does the work (the seats they have held, not the partner who pitches), and who runs the AI after launch. Then check speed to first value, what they diagnose before they build, how they drive adoption, their governance guardrails, who owns the IP, and how the work is scoped and priced before it starts.
How should a CEO evaluate AI implementation partners for a $200M company?
At $200M, the risk is less the technology than the operating model it lands on. Ask each partner what they fix before they build, who from their side has run a company your size, what will be measurably different in 30 days, how they measure adoption at 90 days, and who re-tests the system when the model it runs on is retired. A partner who cannot answer the last question is selling you a launch, not an asset.
What is the difference between AI consulting and AI advisory?
The terms are often used interchangeably. In practice, much AI consulting ends at a recommendation or a pilot. Operator-led AI advisory, the model KeyDelta uses, starts with the operating model, then builds the AI and manages it after launch, so the firm stays accountable for the outcome rather than the deliverable.
Is KeyDelta an AI consulting firm?
KeyDelta is an operator-led advisory firm. We are not consultants who leave a deck. We are operators who have run companies: a Managing Partner and six Principals who, between them, have held the CEO, COO, and CRO seats. We advise leadership teams on how to execute, build the AI systems that make execution faster, and manage them so they keep getting better.
What are the red flags when hiring an AI consulting firm?
A portfolio of pilots and no production systems. Junior staff doing the work after senior staff win it. Building before diagnosing. A plan that ends at handover. No answer on what happens when a model is retired. Adoption never measured. And pricing that cannot be stated before the work begins.
Ask us the ten questions.
Start with a 30-minute call and put these questions to us directly. If we are not the right firm, we will tell you who is.
Operator to operator. No deck, no obligation.