How Much Should a 20–100 Person Business Budget for AI?

Budget for AI in three lines: access for the people who will use it, one funded workflow, and the work required to keep it useful. At Applied Margin, we recommend starting with one team and a defined decision about expansion. Your employee count alone does not determine the budget. A small approved investment with an owner and a review date is more useful than a company-wide allocation nobody can explain.

Price the first step before reserving money for the rest. That might be a $1,500 AI Workflow Sprint, a defined Implementation from $3,500, or an improvement to software you already own. Allow separately for subscriptions and staff time. Review the evidence at an agreed checkpoint, such as 60 days after regular use begins. That is a management decision date, not a promise that the investment will pay back within that period.

What belongs in the three-line budget?

The budget should show what you are authorizing now, what recurs, and what needs another decision. Mixing those categories makes a small pilot look cheap while quietly committing the business to several recurring costs.

Staff time belongs beside the service invoice even when payroll will not change. Someone has to find representative examples, check permissions, attend working sessions, review outputs, and maintain the operating instructions. If that time is unavailable, reducing the external fee will not make the plan feasible.

For a comparison of the services themselves, read what AI consulting prices should buy. The budgeting question here is different: how much should you authorize now, and what evidence should unlock the next amount?

Should every employee get a paid AI seat?

Start with roles and tasks, not a headcount multiplier. An account lead who drafts from approved meeting notes may use an assistant throughout the week. A field technician who only needs a checked work order may not need a separate general-purpose account for that workflow. Neither person is more committed to AI; their jobs require different access.

List the people who will operate the first workflow, the people who will review it, and the administrator. Check the vendor's current licensing rules for each role. Do not share an individual login to avoid buying appropriate access. Also check whether an existing business subscription already includes a capability you need before adding another service.

A small starting group still needs appropriate controls. Choose the plan based on the data, administration, access, and contractual requirements of the work. Buying fewer seats does not justify using an unsuitable account. Customer-facing work remains human-reviewed, and consequential decisions stay with qualified people.

Expansion should follow a job that has been defined and tested. “The sales team wants the same tool” is a request to understand their work, not automatic permission to duplicate the first team's setup.

What could a first budget look like for a 30-person HVAC company?

Consider a hypothetical HVAC business with 30 employees. The first participating team is five office staff. They want a consistent way to prepare internal summaries and find approved administrative procedures. The starting question is whether practical team workflows are enough, or whether searching agreed company knowledge later warrants a focused system.

Hypothetical math: assume the five participants each need one paid seat for two months. Let P be the actual monthly price per seat after checking the vendor's current pricing and terms. Assume all five seats are incremental costs; if appropriate access already exists, remove that incremental charge. The example excludes taxes, usage charges, hosting, integrations, and internal staff time, which must be added when applicable.

The first authorization could be only the Sprint and its associated access and staff time. The Sprint covers one team of up to five people, a short leadership kickoff, two hands-on working sessions, two to three practical workflows, reusable instructions and verification guidance, a prioritized opportunity map, and one follow-up session. Session lengths and scheduling are agreed before booking. Custom software, integrations, and subscriptions are separate. There is no automatic credit toward a later build.

If the team subsequently needs a source-grounded administrative knowledge assistant, the Implementation entry scope starts at $3,500 for one agreed use case, materials, and users. It includes source-linked answers, permissions, evaluation questions, a correction workflow, a team walkthrough, and a 30-day correction window. This is not the price of connecting dispatch, invoicing, and customer communications. Those needs require a separate fixed proposal.

Hypothetical math: a Sprint followed by an Implementation that fits the $3,500 entry scope would total $5,000 in service fees before the separate costs above. That is one possible sequence, not a package price, required path, or return forecast. The second purchase should depend on what the first work reveals.

How much should you reserve for upkeep?

Write the maintenance job before buying a maintenance agreement. A team using a simple drafting workflow may need someone to update instructions and review examples. A connected system may require additional technical support for authentication, changed fields, failures, or model updates. Those are different workloads.

Ongoing Partnership starts at $1,500 per month with a three-month opening term. Its starting scope includes two 45-minute working sessions and one lightweight skill or workflow improvement each month. It does not include unlimited engineering, embedded delivery, system monitoring, broader rollout, or a new build every month. Those are separately scoped and priced.

You may decide to maintain the workflow internally. In that case, name the owner, protect time for the work, and establish when they should seek help. Internal upkeep is not free simply because it has no separate invoice. External upkeep is not automatically necessary simply because AI is involved.

Keep an uncommitted allowance for a known uncertainty only when you can describe it. For example, an integration awaiting technical investigation should have a separate scoping decision. Do not hide an undefined “AI contingency” inside the monthly service amount and assume it buys whatever happens next.

What should happen before you add another subscription?

Run existing and proposed tools through the money, time, and customer filter; the Three Outcomes Filter playbook provides the working exercise.

For the budget, record each tool's owner, renewal date, actual use, and whether another approved tool already performs the needed job. Check dependencies before canceling: an apparently unused account may own an integration or a shared resource. Removing a cost should be a deliberate change with an owner, not a surprise that breaks someone else's work.

Separate committed spend from requested spend in that register. A trial that converts to a paid subscription needs a decision before renewal. The same principle applies to annual contracts: compare the commitment with evidence of sustained use, not just a lower advertised monthly equivalent.

How should the first year unfold?

Think in funding decisions rather than a predetermined sequence of purchases. A useful path can be Sprint, then one Implementation, then Partnership if recurring help is needed. It can also stop after the Sprint, move directly to a defined Implementation, or remain entirely inside existing tools.

Before each next step, ask:

1. What changed in the work, and what evidence supports that conclusion?
2. Can the operator use, verify, and recover from the workflow without improvising?
3. What new problem does the proposed spending solve?
4. What will remain a separate cost or responsibility?
5. When will we decide to continue, revise, or stop?

The optional Margin Scan is available at $1,500 for seven business days when a complex operating area needs deeper discovery. It is not a required stage in the budget. Its full $1,500 credit applies to Applied Margin implementation authorized within 14 days; the Scan's credit and Clarity Guarantee do not apply to the Sprint. Use the assessment decision guide when that distinction matters.

At the review date, compare complete work, including review and exceptions. A draft produced faster is not a useful gain if checking it takes longer. Recovered staff time is capacity; it is not cash savings unless the business avoids a cost or uses that capacity to produce a real financial benefit. For the measurement plan, read how to judge whether AI is paying off.

Let’s talk about your business. Bring the work you want to improve and the limits you need the budget to respect. We can identify a useful first commitment without asking you to fund every possible next step. See the current service scopes and prices before deciding.

FAQ

Is there a sensible AI budget per employee?

Employee count alone is a poor starting point. Budget around the people using a defined workflow, the service or internal work needed to establish it, and the upkeep it requires. A 30-person company might begin with one five-person team. That example is a scope choice, not a recommended spending ratio or an industry benchmark.

Should we fund a full year before testing anything?

Approve the commitments you understand and make later spending conditional on evidence. Check minimum terms before buying subscriptions or services. Applied Margin's Ongoing Partnership has a three-month opening term; the Sprint is a separate fixed-price engagement. Neither requires you to pre-authorize a year of new builds. Establish a review date and the decision you will make there.

Does the $1,500 Sprint cover everyone in our company?

The AI Workflow Sprint covers one team of up to five people, not every employee in a larger company. It includes a kickoff, two working sessions, two to three workflows, instructions, an opportunity map, and one follow-up. Session lengths and scheduling are agreed before booking. Software, integrations, subscriptions, and broader rollout are separate; there is no automatic build credit.

How do we budget for software without guessing its price?

Check the vendor's current plan, licensing terms, and usage charges for the actual users and task. Keep the per-seat amount as a variable until verified. Record what is already included in your existing subscriptions, and distinguish incremental cost from money you already spend. Add any hosting, connector, or model usage charges separately rather than treating the service fee as all-inclusive.

What if the workflow saves time but does not reduce payroll?

Record the time as recovered capacity and decide where it will go. It may absorb growth, improve response quality, reduce overtime, or create room for another task. Those outcomes have different financial consequences. Do not call the full loaded value of every saved hour cash savings, and do not use that assumption to justify an otherwise unsupported recurring commitment.

Put it to work

Install the steps, prompts, human approval rules, and scoreboard in Tool Stack & AI Spend Audit.