When to partner instead of hire: staffing AI and automation work you can't deliver yet

A client you have served for years asks whether you can build an AI agent to handle their support queue, or automate the process their operations team runs by hand. You have good developers. None of them has shipped this kind of system. The client would prefer to buy it from you, and if you say no, they will buy it from someone else, who will then be in the account.

You have three options: decline, hire, or partner. This article is about choosing between the last two. It belongs to the Team hub, which covers the other ways an agency adds capacity.

Why this work is hard to hire for

For ordinary development work, the usual answer to a capacity gap is a hire or staff augmentation. Specialist AI and automation work breaks that answer in three places.

You cannot evaluate the candidate. If nobody at your agency has built a production AI system, nobody can tell a strong practitioner from a confident one in an interview. You would be hiring on trust for the most expensive role on your payroll.

You cannot manage the work. Augmentation hands you a person to direct. If you do not know what good looks like, how to test it, or which failure modes to plan for, you cannot direct them. The pillar guide on augmentation makes this point: rent a skill only if you already know how to manage it.

The demand is lumpy. One client has asked for one project. Perhaps two more will follow this year. A specialist needs full-time work.

The arithmetic: hire versus partner

Take a real shape of problem. The client’s project is worth $90,000 and will take about three months. Here is what hiring a specialist to deliver it involves.

To carry that person at a 50 percent gross margin, you need $225,000 divided by 0.50, which is $450,000 of specialist work a year. You have $90,000 of it. The hire only works if you sell another $360,000 of similar projects within twelve months, in a service line you have never sold before. And the client is waiting three to four months before anything starts.

Now the same project through a white-label delivery partner.

A 30 percent margin is lower than you earn on your own team’s work. Compare it with the alternatives that are really available: a $171,000 shortfall on a speculative hire, or zero revenue and a competitor inside your account.

The comparison side by side:

How a white-label arrangement works

You sign the contract with your client. You sign a separate subcontract with the partner. The client pays you and you pay the partner. The points below are the ones to settle in writing before the first project.

Margin

There are two common structures.

Wholesale rate. The partner charges you a discounted hourly or daily rate and you bill the client your own rate. If the partner charges $110 an hour and you bill $160, you keep $50 an hour, or 31 percent. This works for ongoing or loosely defined work, and it leaves you exposed if hours run over on a fixed price you quoted.

Fixed fee per project. The partner quotes you a fixed price for a defined scope and you add your margin. This is the structure in the table above. It is the safer one for a first project, because the partner carries the estimating risk on work you cannot estimate yourself.

Whichever you choose, involve the partner in scoping before you give the client a number. The most common way to lose money on white-labeled work is to quote the client first and ask the partner second. Change requests should flow the same way: the partner prices the change, you add margin, the client approves. Managing scope creep matters more here, since every unbilled extra comes out of a thinner margin.

Agree payment timing too. If the client pays you at 45 days and the partner expects payment at 15, you are financing the project. Match the terms, or tie the partner’s payment to the client’s milestone payments.

Who talks to the client

There are three workable models. Pick one deliberately.

The first model protects the relationship most and slows the work most, because every question passes through someone who cannot answer it. For AI and automation projects, where requirements get refined through conversation, I prefer the second or third.

Two rules apply to all three. Check whether your client contract requires consent for subcontractors, and get it if so. And do not tell a client that the partner’s staff are your employees. Presenting a blended team under your brand is normal practice. A direct false statement, when the client asks, puts the relationship and possibly the contract at risk.

IP

The client expects to own what they paid for. Ownership has to pass from the partner’s people to the partner, from the partner to you, and from you to the client, with each step in writing. Your subcontract should assign all project work product to you on payment, or on creation, so that you can meet the promise in your client agreement.

Specialist partners usually bring existing tools, frameworks, and components they have built before. They will keep ownership of those and grant a license. That is reasonable, provided the license is broad enough for the client to use, modify, and maintain the finished system without the partner. Make sure your client contract describes this accurately, since promising the client full ownership of something that contains licensed parts is a breach waiting to be noticed. The article on who owns the code walks through the chain.

Also settle who holds the accounts. Model provider accounts, automation platform workspaces, and hosting should sit in the client’s name or yours. If they sit in the partner’s name, a change of partner becomes a migration.

Non-solicitation

You are introducing a specialist firm to your client. Protect the introduction.

During this Agreement and for twenty-four (24) months after the end of the last project for a given client, Partner will not solicit or accept work from that client, directly or indirectly, except through Agency.

The partner will want a matching protection for its staff: you agree not to hire its people for a period. Accept that, and negotiate a clear route to bringing the work in-house later, such as a conversion fee or a time limit, so the clause does not block the plan described at the end of this article. Enforceability of these clauses varies by jurisdiction, so have counsel review the wording.

Liability, quality, and support

Your client’s contract is with you. If the system fails, the client’s claim is against you, whatever the partner’s role. Three protections follow.

• Back-to-back terms. Whatever you promise the client on warranties, acceptance, confidentiality, data protection, and security, the partner promises you in the subcontract. Your master services agreement with the client is the starting document.
• Acceptance criteria you can check. For AI systems, define how quality is measured before the build starts: a test set, a target accuracy on that set, response times, what happens on failure. Without this you cannot tell the client, or yourself, whether the work is done.
• Support after launch. AI and automation systems need attention after delivery. Models get updated, connected systems change, and edge cases appear. Agree a support retainer with the partner and sell a matching one to the client. This is often the most profitable part of the arrangement, and the guide to retainer pricing covers how to structure it.

How to choose a partner

You are staking a client relationship on this firm. Check these things before the first project.

They have built this specific kind of system. Ask to see two or three in operation, with a walkthrough of what went wrong and how they fixed it. A partner who cannot describe failures has not run anything in production.

They have worked white-label before. Ask for a reference from another agency owner. Ask that owner two questions: did the partner ever approach your client, and what happened when a project went badly?

You meet the people doing the work. Some firms sell with senior staff and deliver with juniors or a further subcontractor. Name the lead in the subcontract and require consent for any subcontracting.

They scope before they price. A partner who quotes a fixed fee from a two-line brief is guessing, and the guess will become a change request. A good one asks about data, existing systems, volumes, and what success means.

Their terms fit yours. Non-solicitation, IP assignment, confidentiality, and insurance should be offered without a fight. Hesitation over the non-solicit tells you something.

They explain things to you. You need to sell and support this work. A partner who helps your team understand the system is building a long relationship. One who keeps it opaque is building a dependency.

Then start small. The guide to hiring remote developers recommends a paid trial for individuals, and the same logic applies to firms. A paid discovery phase or a first project under $30,000 tells you more than any amount of checking.

When to bring it in-house

Partnering is the right structure for uncertain demand. Once demand is proven, the margin you give up starts to exceed what a specialist would cost.

Using the numbers above, a specialist costs $225,000 a year loaded. At a 30 percent partner margin, you earn $30 on every $100 of this work. On your own team at a 50 percent margin, you would earn $50. The extra $20 per $100 pays for the hire when volume is high enough:

In this example the two options cross at about $320,000 a year, and the hire is comfortably ahead by $450,000. The in-house column assumes one person can deliver all of it, which stops being true somewhere above $450,000 at typical rates. Run the table with your own figures.

I would want revenue plus all four of these before hiring:

1. Two consecutive quarters of this work at a run rate above the crossover point.
2. Signed or near-signed pipeline for the next two quarters.
3. Someone at the agency who has learned enough from the partner projects to interview and manage a specialist.
4. A support load from delivered projects that gives the new hire a base of steady work.

The transition can be gradual. Many agencies keep the partner for peaks and for the hardest builds while a first hire takes on support and smaller projects. If the practice keeps growing, the AI hub’s guide to running an AI consulting business covers what it looks like as a service line of its own, and bench management covers how to keep an expensive specialist from sitting idle between projects.

The decision in one pass

Say yes to the client. Bring a partner into scoping before you quote. Use a fixed fee for the first project, with back-to-back terms, written IP assignment, and a non-solicit that runs both ways. Put your own project manager on it and have them learn everything they can. Track the revenue from this kind of work each quarter. When it has stayed above the cost of a specialist for six months and the pipeline says it will continue, hire.