Planning Numbers
How Much Does AI Implementation Actually Cost a Small Business?
No vague ranges, no “it depends.” Here are planning numbers for each level of AI work, the hidden costs nobody mentions, and how to know if the numbers make sense for your business.
Quick Answer
How much does AI implementation cost a small business?
A practical starting point for custom AI implementation is approximately $1,000 per consulting or development day. One focused fix — turning form submissions into CRM records, or alerting you to high-priority quote requests — runs about $1,000. A connected workflow that handles several steps of a process runs about $5,000. Department-level change starts around $50,000 and only makes sense when the underlying process has enough volume or value to justify it. Treat these as planning numbers for your budget, not a price list.
- →$1,000 per consulting or development day is a practical planning number
- →$0: free tools to learn where AI fits your business
- →$1,000: one focused fix, in operation
- →$5,000: one connected workflow with a measurable beginning and end
- →Budget from the cost of the problem, not from what you can afford to spend
The Starting Number
$1,000 a day. Now do the math on your problem.
A practical starting point for custom AI implementation is approximately $1,000 per consulting or development day. If the problem can be understood, solved, tested, and implemented in one day, it may be a $1,000 project. More complex workflows take longer.
That number makes owners nervous, so let me make it concrete. The question is never whether $1,000 is a lot of money. The question is what the broken process costs you every single week. A quoting process that eats six hours of staff time per quote, three quotes a week, is burning roughly 18 hours a week. At any reasonable loaded wage, that's a four-figure monthly problem.
A $1,000 fix that removes half of that work can pay for itself in weeks. A $50,000 system built for a problem that happens twice a year is unlikely to make economic sense. The size of the check matters far less than the math behind it.
To be plain about it: every number in this article is a planning number for your budget, not a BotLogix price list. The only fixed price we publish is the $1,000 Strategy Day. Everything after it is scoped once we've seen the work, and priced on the value we find.
The budget should not begin with, 'How much AI can I afford?' It should begin with, 'What is this problem currently costing me, and what would it be worth to solve it?'
— Steffen deGraaf, BotLogix
The Budget Ladder
Four levels of spend, from free to department-level.
Every level has a legitimate use. The mistake is jumping levels before the level below has proven anything.
$0 — Learn on free tools
Use ChatGPT, Claude, Gemini, Perplexity. Summarize documents, organize meeting notes, draft emails, analyze a repeated process. The goal isn't to learn prompts — it's to understand where AI actually fits your business before you spend a dollar.
$1,000 — One focused fix
Choose one or two specific tasks. Alert the owner to high-priority quote requests. Turn form submissions into CRM records. Convert voice notes into structured tasks. Prepare standardized first drafts. The goal is something useful in operation, not a report.
$5,000 — Connected workflow
Instead of one isolated task, connect several steps of a business process. Capture an opportunity, organize the information, create tasks, prepare a draft response, update the CRM, alert the right person. One complete workflow with a measurable beginning and end.
$50,000 — Department-level change
The project may connect multiple departments, data sources, customer touchpoints, and internal systems. Only move forward when the process has enough volume or value to justify it. In the right business, this prevents or recovers a much larger amount. In the wrong one, it becomes an expensive system nobody uses.
Agents, Build and Run
What an AI agent costs to build and to run each month.
If you're pricing an AI agent, the numbers above still apply. They just split into two bills: the one you pay once to build it, and the one that arrives every month after it goes live.
First, what we mean by an agent. It's software that reads what comes in, decides the next step, and does it inside your other systems: checks the calendar, drafts the reply, updates the CRM, hands the odd case to a person. (If you're still deciding whether you need one at all, start with ChatGPT vs custom GPTs vs AI agents.)
The build. Most agents worth building in a small business are a connected workflow from the ladder in Chapter 02: one process, several steps, a clear start and finish. So plan around the same number, roughly $5,000. An agent that has to reach across several departments and data sources moves toward the department-level end, and it has to justify that spend the same way. We don't price a build until after the Strategy Day, because the cost depends on what the agent has to connect to and what a mistake would cost you.
The monthly bill. The usage costs card in Chapter 03is the warning. Here's the arithmetic behind it for the model itself, because owners often expect the model to be the big line.
Example math, not a quote
An agent that handles 1,000 tasks a month. Each task sends the model about 3,000 tokens (your instructions, the incoming message, a few records from your system) and gets about 800 back. That's 3 million tokens in and 800,000 out each month, at standard list prices with no discounts.
| Model | List price per million tokens (in / out) | Model cost per month | Per task |
|---|---|---|---|
| Claude Haiku 4.5Anthropic's smallest model, as of September 2026 | $1 / $5 | $7.00 | $0.007 |
| Claude Sonnet 5Anthropic's mid-size model | $2 / $10 | $14.00 | $0.014 |
| GPT-6 LunaOpenAI's small, high-volume model | $0.10 / $0.50 | $0.70 | $0.0007 |
| GPT-6 SolOpenAI's mid-tier model | $2 / $10 | $14.00 | $0.014 |
All figures in US dollars, no exchange rate applied. Prices checked September 2026 on Anthropic's pricing page and OpenAI's pricing page. Vendors change these often, so check the page yourself before you budget.
Three things push the real number up. An agent that calls the model several times per task sends its context again on every call, so multiply by the number of calls. A model that thinks before it answers is billed for that thinking as output, whether or not you ever see it. And paid tools carry their own fees: Anthropic charges $10 per 1,000 web searches, so one search per task adds $10 a month, more than the whole Haiku bill above. Token counts also differ between models. Anthropic says the tokenizer behind Sonnet 5 produces roughly 30% more tokens for the same text than older Claude models like Haiku 4.5, so the same work could put that row closer to $18.
Two things bring it down. Caching the instructions that repeat on every task: at both vendors, reading cached input costs a tenth of the normal input price. And batching work that can wait until later, which both vendors discount by half. How much caching saves depends on how often your tasks arrive, so it isn't in the table.
At this volume the model bill is measured in dollars, not hundreds of dollars. The rest of the monthly cost lives somewhere else:
- Hosting. A small server, a container or serverless functions for the agent, plus whatever schedules and queues its jobs.
- A database. Somewhere to keep records and the documents the agent searches (Postgres with a vector extension, or a managed vector database), with backups. Making those documents searchable adds small token charges of its own.
- Integration fees. Some CRMs, helpdesks and accounting tools keep API access on a higher plan. Automation platforms like Zapier or Make charge per task. Email and SMS sending cost money too.
- Monitoring. Error alerts, a log of what the agent was asked and what it did, and a spending cap set at the model vendor so a stuck loop can't run up a bill overnight.
- Review time. Someone still spot-checks the output and handles the cases the agent hands back. Price it with the formula in Chapter 06: hours × loaded wage × frequency.
- Model retirements. Anthropic lists Haiku 4.5, one of the four models in the table, as active with retirement not sooner than October 15, 2026. Every model gets a date like that eventually. When a model you rely on retires, someone re-tests the agent on its replacement. That's the maintenance card from Chapter 03, with a date on it.
One more thing on picking a model. A cheaper model that needs more of its answers corrected can cost more in review time than it saves in tokens. Test the candidates on your own examples first, then look at the price.
What does an AI agent cost to build and to run each month?
Our Own Quoting System
Six hours to 30 minutes: the quoting math.
Before we built our quoting system, preparing a group-benefits quote was an extremely manual process. Review the client's existing contract, locate dozens of specific details buried in different carrier formats, build a spreadsheet, prepare submissions for three or more insurers, send individual emails, wait, review, rebuild, present.
For one opportunity, the combined work could take more than six hours. We reduced that to approximately 30 minutes of active human work — a reduction of more than 90 percent of the active processing time. The system also made the customer experience better: information captured consistently, submissions prepared sooner, insurers approached faster.
The Numbers
6+ hrs
Manual process per quote
30 min
Active human work after automation
90%+
Reduction in processing time
Examine both volume and value. A six-hour automation may not be worth building for a task that happens once a year. It can be extremely valuable when the task happens every week, affects revenue, delays customers, or requires expensive staff time. Do not automate something merely because it can be automated. Automate it because the repeated cost of leaving it manual is greater than the cost of building and maintaining the solution.
The Framework
How to run the numbers on your own problem.
Before you talk to anyone about AI, answer six questions about the process you want to fix. Write the answers down. They tell you what the solution is worth.
How often does the task occur?
Daily, weekly, monthly? Frequency is the multiplier on every other answer. A small daily problem usually beats a large annual one.
How many hours does it take?
Track the real elapsed time, including the waiting, the rework, and the interruptions — not the idealized version.
Who does it?
A task done by your most experienced person costs more than the same hours done by anyone else. If that's you, the owner, the cost is whatever your time is worth.
What does that time cost?
Hours × loaded wage × frequency. This is the floor of what the problem costs you every month.
What happens when it's delayed?
Quotes that sit for days get lost. Follow-ups that slip become competitors' customers. Delay cost is often larger than labor cost.
What revenue is attached?
If the process touches sales, the question isn't just what it costs — it's what faster, more consistent handling would recover.
When you have those six answers, the budget conversation changes completely. You're no longer asking what AI costs. You're deciding whether a specific solution is cheaper than the problem — which is the only question that matters.
Go deeper
Using AI in Your Business
The owner's guide for 5–25 person companies.
AI Strategy Day
$1,000 on-site. A written AI Opportunity Blueprint with a 30-day action plan.
What Is an AI Strategy Day?
The operational breakdown — hour by hour.
ROI Estimator
Run your own numbers in 3 minutes.
Workflow Proof Sprint
Test one AI workflow before committing to the build.
Fractional AI Officer
Ongoing AI leadership without the full-time hire.
30-minute strategy session
For real businesses with real problems.
Not for tire kickers. Not for tech tourists looking for a demo. For business owners who already know something in their operation is broken — and want to know if AI can fix it.
This isn't about replacing people.It's about killing the bad business processes your team hates as much as you do — the manual work, the missed calls, the after-hours admin.
Strategy sessions are free. Strategy Days are $1,000, capped at 4 people per session. Larger teams or multi-location workshops by special arrangement — ask when you book.
Your session is with
Steffen deGraaf
Founder, BotLogix · Burlington, ON