Custom AI vs Off-the-Shelf AI: ROI Guide for Business Workflows in 2026
Buy AI seats for work every company does the same way; build for one workflow that acts inside your systems. Prices read 5 October 2026 and a 40-person worked example.
Jun 4, 2026
Short answer: buy off-the-shelf AI for work every company does the same way (writing, summarising, searching documents, meeting notes) and build custom AI for one workflow that is specific to your business, runs at volume, and has to act inside your own systems with rules and an audit trail. For a 40-person business, off-the-shelf seats cost about $8,600 to $9,600 a year (ChatGPT Business or Microsoft 365 Copilot Business, prices read 5 October 2026). A custom AI workflow from a senior team costs $15,000 to $50,000 once for a first production version. Most businesses end up with both: seats for everyone, and one custom workflow where the money is.
If you want a range for your own workflow first, the AI development cost calculator gives one in about a minute.
Published prices, side by side
| Route | Price | What you get | Source, read 5 Oct 2026 |
|---|---|---|---|
| ChatGPT Business | Standard seat $20 per user per month billed annually ($25 monthly); Premium seat $100 billed annually ($125 monthly); for teams of 2 to 200 | Chat, files, connectors to Google Workspace, Slack, Microsoft 365 and more, workspace agents, no training on business data by default | openai.com/business/pricing |
| Microsoft 365 Copilot Business | From $18 per user per month paid yearly (shown as originally $21); bundles with Business Standard at $23.50 and Business Premium at $32 | Copilot inside Word, Excel, Outlook, Teams and the rest of Microsoft 365 | microsoft.com Copilot business pricing |
| Freelance AI integration (Upwork) | $30 to $150 per hour; single AI feature $1,000 to $4,000; chatbot $2,500 to $9,000; retrieval build $4,000 to $14,000; end-to-end workflow automation $8,000 to $20,000 | One developer connecting a model to your tools | upwork.com AI integration developers |
| Custom AI workflow, first production version (KUMO Starter Build) | $15,000 to $50,000 once, typically 2 to 3 months, code and data in your own accounts | A senior team: integrations, approvals, logging, evaluation, handover | KUMO published band |
| Custom AI system with several integrations or a customer-facing product (KUMO) | $50,000 to $100,000 once | As above, across more systems | KUMO published band |
| Ongoing engineering after launch (KUMO) | $5,000 to $10,000 per month while the workflow is still changing | Monitoring, model updates, new steps | KUMO published band |
A custom workflow also pays for model usage, billed per token by OpenAI, Anthropic or Google. Estimate it from your monthly volume before you sign; for most single workflows it is a smaller line than the build.
What 40 people cost a year
| Route | Monthly | Yearly | Over 3 years |
|---|---|---|---|
| ChatGPT Business, 40 standard seats at $20 | $800 | $9,600 | $28,800 |
| Microsoft 365 Copilot Business, 40 users at $18 | $720 | $8,640 | $25,920 |
| Microsoft 365 Copilot Business, 40 users at $21 (the shown original price) | $840 | $10,080 | $30,240 |
Seats are cheap per person, and they are the right buy for general productivity. What they do not do on their own is run a process: read a new order from your system, check it against your pricing rules, route exceptions to the right person, write the result back and keep a record of every decision. That is the job a custom workflow does, and it is why comparing the two on price alone misleads. The useful question is which one fits the work.
Which route fits the work
Four questions decide it.
1. Is the task the same in every company? Drafting emails, summarising calls, searching a shared drive: buy. Vendors improve these tools every month, and you will not out-build them.
2. Does the AI need to act inside your systems? If the output has to land in your ERP, CRM, ticketing or order system, with your field names and your approval rules, an assistant in a chat window stops short. Connectors help with reading data; acting on it, with permissions and a log, is usually where a build starts.
3. How much volume and risk is there? A workflow handling hundreds of documents, tickets or orders a week, where a wrong answer costs money or a customer, needs testing against your own past cases, a human approval step and monitoring. Off-the-shelf tools do not give you that per workflow.
4. Is the workflow part of how you compete? If the way you quote, triage, underwrite or schedule is the reason customers pick you, the logic belongs in a system you own, not in a vendor's configuration screen.
Two or more answers pointing at "build" is the signal to price a custom workflow. If one or none do, buy seats and revisit in six months.
Where off-the-shelf AI falls short
- Data stays where the vendor can reach it. Connectors cover the popular tools; an older or industry-specific system usually is not on the list.
- Rules live in prompts, not in code. Pricing rules, approval limits and exceptions written into a chat instruction are hard to test and easy to break.
- No per-workflow audit trail. When a customer disputes an outcome, you need to show what the AI saw and decided. Seat-based tools are not built around that record.
- Seat costs scale with headcount, not with value. A workflow that saves one team 20 hours a week does not need every employee on a premium seat.
For support specifically, our Intercom Fin vs custom support chatbot comparison runs the same arithmetic on per-resolution pricing, and custom AI vs ChatGPT Enterprise covers larger teams.
What a custom AI workflow includes
A Starter Build ($15,000 to $50,000) is one AI workflow in production: connections to the systems it reads from and writes to, the model steps, a human approval step where a wrong answer is costly, logging of every decision, an evaluation set built from your own past cases, deployment into your own cloud account, and a handover another engineer can continue from. Two to three months is typical. AI workflow automation describes how these projects are scoped, and what an AI agent costs to build breaks the band down by driver.
KUMO's founders built Volopay's first production version as its founding engineers, and the FLIN case study walks through another production build from scoped brief to launch.
To get a band for your own workflow, book a free 30-minute call with an engineer. Bring one workflow: what comes in, which systems it touches, who approves what, and how many a week.
FAQ
What is the difference between custom AI and off-the-shelf AI? Off-the-shelf AI is a product sold per seat or per use, such as ChatGPT Business or Microsoft 365 Copilot, that does the same job for every customer. Custom AI is a workflow built around your own systems, rules and data, which you own and which acts inside your processes.
Is custom AI worth it for a small or mid-size business? For one workflow with volume, rules and a cost of being wrong, often yes. For general writing, summarising and search, no: seats cost about $20 per person per month at prices read 5 October 2026, and a build will not beat them on those tasks.
How much does custom AI cost in 2026? A first production workflow from a senior team costs $15,000 to $50,000 and typically takes two to three months. Systems across several integrations or a customer-facing product run $50,000 to $100,000. Ongoing engineering is $5,000 to $10,000 per month while the workflow is still changing, plus model usage billed per token.
Should we build or buy AI? Buy for tasks every company does the same way. Build when the AI must act inside your systems, at volume, with rules and an audit trail, or when the workflow is part of how you compete. Most businesses do both.
Can ChatGPT Business connect to our own systems? OpenAI's business pricing page lists connectors to Google Workspace, Slack, GitHub, Microsoft 365 and more, and workspace agents for customised workflows. Check whether your core system is on that list, and whether the agent can write back to it with the permissions and records you need.
Last verified 5 October 2026. ChatGPT Business prices from openai.com/business/pricing, Microsoft 365 Copilot prices from microsoft.com Copilot business pricing, freelance rates from upwork.com AI integration developers, all read on that date. Yearly and three-year totals are KUMO's arithmetic on those published prices. KUMO bands are KUMO's published prices.
