OneSeven Tech focuses on companies with $25 million or more in annual revenue because they often have two things an AI transformation needs: repeated operational work worth improving and the capacity to reinvest in the business.

Smaller companies with a clear project and budget are welcome too. A $50,000 or $100,000 project can be a sensible place to start. The budget needs to cover what happens after the build, not just the work needed to launch it.

Start with the work you want to improve

Manual order entry, document reviews, information copied between systems, and slow handoffs all consume time. The business case starts with how often that work happens, what it costs, and what changes when it improves.

A larger operation may repeat the same task thousands of times. Improving it can benefit several teams. A smaller business may have one well-defined workflow that makes an equally good starting point.

That is why we look at the operation and the investment together, rather than qualifying a project on revenue alone.

Put the investment in the context of your IT budget

Avasant’s 2025/2026 study, using first-half 2025 data, reports median IT spending of 2.9% of revenue. Its sample covered organizations with U.S. or Canadian operations and at least $50 million in revenue or more than $1 million in IT spending.

The SIM 2025 IT Issues and Trends Study, published in March 2026, reports a 5.6% mean IT-budget-to-revenue ratio among 199 organizations reporting both figures. The broader sample was predominantly U.S.-based, with median revenue of about $450 million.

These studies describe different populations and measures, not a spending range or target for a $25 million company. Industry matters too: SIM’s smaller sector samples reported averages of 1.6% for manufacturing and 7.1% for financial services.

An IT budget pays for staff, applications, security, infrastructure, and existing commitments. It is not all available for AI. If this is your first custom-software project, you will need to fund its operation alongside those commitments.

In that 2025 study, Avasant reports AI as 1.9% of the IT budget on average, a share of spending rather than an additional charge on top. In Harness’s May-June 2026 survey, 80% of 700 engineering leaders and practitioners reported rising AI spending. Those respondents worked at AI-using companies with at least 1,000 employees across five countries.

For your business, the additional cost depends on the workload and what the new system replaces.

Work backward from the cash you can reinvest

Profit and available cash are different numbers. Consider this hypothetical example, in U.S. dollars, assuming a 30% operating margin before interest and tax:

The cash column assumes that operating cash flow, after cash interest and taxes, less non-AI capital spending equals half of operating profit. Your collections, inventory, taxes, and capital needs will determine the actual amount.

Debt principal, distributions, and reserves still need funding. If management allocates 20% of the cash in this example to AI, the budgets are $300,000 and $750,000. These percentages are planning assumptions, not recommended allocations.

A $500,000 program would use about 33% of the smaller company’s modeled cash flow and 13% of the larger company’s. The smaller business could still pursue a worthwhile project, but the commitment takes a larger share of its resources.

The reverse calculation is simple:

Required cash flow after capital spending = investment ÷ chosen allocation.

At a 20% allocation, a $500,000 investment requires $2.5 million in annual cash flow after capital spending. A $1 million investment requires $5 million. Check payment dates as well as annual totals: December’s cash cannot pay an invoice due today.

Budget for the build and the operation

For a broader program, planning capacity around $500,000 for implementation and a year of operation can help frame the decision. Here is one hypothetical allocation:

That operating allowance includes:

• AWS or Google Cloud infrastructure: application hosting and servers, databases, storage, backups, and bandwidth.

• Model and token usage: API input and output tokens, embeddings, document processing, and tools.

• People and ongoing support: maintenance, evaluation, human review, security, monitoring, and staff training.

A smaller launch can have a smaller total budget. A $50,000 project does not automatically require a $500,000 reserve. Choose a scope you can fund through launch and ongoing use, then expand when the results justify it.

If you are still deciding the delivery route, see how to build, buy, or combine AI software.

Calculate model usage from the workload

Tokens are units a model processes and generates. Input and output have different prices. These standard, uncached Claude API rates were checked November 7, 2026:

The official pricing includes higher rates for longer Haiku prompts and conditions for caching and batch discounts. Select the model against the quality and speed the job requires.

For example, 100,000 Sonnet calls per month, each averaging 10,000 billable input tokens and 1,000 billable output tokens, cost:

100,000 × [(10,000 × $2 + 1,000 × $10) ÷ 1,000,000] = $3,000 per month.

If each of 100,000 business tasks needs five such calls, model usage becomes $15,000. Longer documents, agent loops, retries, and reasoning output can increase consumption. Embeddings, document processing, and tools may add separate charges.

Lower token prices do not guarantee a lower bill when usage grows. Track cost per completed task, set alerts and usage limits, and account for data-retention and residency requirements before launch.

Include networking without confusing it with hosting

Google Cloud’s network pricing lists Premium Tier internet transfer from Iowa to North America at $0.12 per GiB after the first 1 GiB monthly allowance, through 1,024 GiB.

With that allowance unused, 1 TiB, or 1,024 GiB, costs (1,024 − 1) × $0.12 = $122.76 in transfer charges. That is bandwidth alone. A decimal 1 TB is smaller than 1 TiB.

AWS has different allowances, including a shared 100 GB monthly internet-transfer allowance across services and regions, with exclusions.

Internet egress, inbound traffic, transfers between regions or zones, NAT processing, and load balancers have different billing rules. Compute, databases, backups, and monitoring are separate costs. Price the architecture you will use rather than applying one bandwidth rate to everything.

Measure the return and assign an owner

Before building, record workflow volume, time spent, errors, and current costs. Decide how improvements will become financial value. Recovered hours may create capacity without reducing payroll; additional sales contribute their incremental profit, not their full revenue.

For illustration, a $1 million investment today followed by $500,000 net cash at each year-end for three years produces 50% cumulative undiscounted ROI and about 23.4% annual IRR. This hypothetical example excludes taxes and terminal value; it is not a client result or average return.

Want to run the numbers for your business? Try our AI ROI calculator. Start by estimating the work hours your team could recover and the value of that capacity.

Custom software development with OneSeven Tech may qualify for federal R&D tax credits when the project’s activities and expenses meet the research-credit requirements. Eligibility depends on the work, not simply on hiring a developer.

A tax deduction reduces taxable income, while a credit reduces tax liability subject to eligibility and limits. Have a qualified tax professional apply the IRS research-credit rules, including coordination between deductions and credits.

For the eligibility tests, see custom software and the R&D tax credit.

The system also needs usable data, reliable integrations, a business owner, and people ready to use it.

Bring one workflow, a build-and-operate budget, and the person responsible for the outcome. Whether you are planning a focused project or a broader transformation, you are welcome to book a discovery call.