Free calculator · WhatsApp and Automation
AI Chatbot and Agent Cost Estimator.
Model usage, platform fees and human escalation priced separately, because escalation is the line that decides the project and the one every quote leaves out. Includes what the same volume costs with no bot at all.
Changes the build and the cost of the people handling escalations. Model prices are global.
This sets the build, how much context each conversation needs, and how much the bot can finish on its own. An agent that writes to your systems costs several times one that answers questions, in build and in running cost.
Count conversations, not messages. A conversation is one person's session from first question to done.
Prices per million tokens move constantly and are cut without notice. These are anchors for sizing, not quotes: check the provider before committing a budget.
Every price is editable. Guardrails and evaluation is the one people cut first and regret: it is what stops the bot confidently saying something wrong.
Zero if you build against the API directly. Set it if you are buying an off-the-shelf bot platform.
The handovers the bot cannot finish.
Fully loaded: salary, employer costs, the software they need. This is the number that decides whether the whole project pays for itself.
Add your own line items. A discount goes in as a negative number.
Build, then every month
AED 13,200 + AED 840/mo
For 2,000 conversations a month, of which the bot finishes about 68% on its own. The build is a one-off; the monthly figure is the one that decides whether this is worth doing.
Against doing it with people only
At this volume the bot costs more to run than answering by hand. That is normal below a few hundred conversations a month: build it for the response speed and the out-of-hours cover, not for the saving.
The line that decides this project is human escalation, not model usage. A bot never deflects everything, and the share that still reaches a person costs far more per contact than the model does. Any quote that leaves escalation out is describing a system that does not exist yet. Model prices, reviewed August 2026, are anchors for sizing rather than vendor quotes.
Estimate done
Want the real number for your project?
An estimate works from typical scope. A quotation works from yours. Send this breakdown over and you get a fixed price back, from the person who would do the work, within one business day.
What this tool measures
The build is the part everyone quotes and the part nobody is surprised by. The running cost is where AI projects go wrong, because it has three separate components that almost every quote conflates into one number: what the model charges, what the platform charges, and what the humans still cost.
That third one is the whole game. A bot that resolves seventy percent of contacts leaves thirty percent for a person, and a person costs far more per contact than the model does. Any estimate that does not have an escalation line is describing a system that does not exist yet, which is why this tool puts it front and centre and shows you the same volume handled entirely by people for comparison.
How the running cost actually works
Model usage is billed per million tokens in and out. Tokens in are your prompt plus whatever context the bot retrieves before answering, which is why a bot that reads your entire knowledge base every turn costs several times one that retrieves three relevant paragraphs. Tokens out are the reply, and they are priced several times higher than tokens in on every provider.
The tier matters more than the provider. Between a small model and a frontier one there is roughly an order of magnitude in price, and for answering questions from your own documents the small one is frequently indistinguishable. Paying frontier prices to read a FAQ is the most common way an AI budget disappears.
Escalation is the largest line once volume is real. It scales with the share the bot cannot finish, and that share never reaches zero. Improving it from thirty percent to twenty is worth more than any model saving available to you.
How to read the result
Look at cost per conversation before you look at the total. It is the figure you can compare against what a contact costs you today, and it is the one that tells you whether the project is worth starting.
Below a few hundred conversations a month the bot will cost more to run than answering by hand, and the tool says so plainly rather than hiding it. That is not a reason to abandon the idea: at low volume you build for the instant reply and the out-of-hours cover, and you should be honest that the saving is not the reason.
The payback figure treats the build as an investment against the monthly difference. If it comes out beyond about eighteen months, the volume is not there yet and the sensible move is a narrower bot on a smaller model.
What to do next
Write down the ten questions your customers actually ask most. If a bot answers those ten well it will deflect most of your volume, and the scope you need is far smaller than the one you were about to buy.
Then decide what it must never do. Guardrails and an evaluation set are the line people cut first and regret hardest, because a bot that confidently says something wrong costs more than the whole project saved.
Questions people ask
Why is the model price editable rather than fixed?
Because it moves constantly and gets cut without notice, and the tiers differ by an order of magnitude. Hardcoding a model name would make this tool wrong within a quarter. The anchors are there for sizing; check the provider before committing a budget.
What deflection rate is realistic?
Fifty to sixty percent for a bot answering from your own content, higher for one that can act in your systems, and lower than any vendor demo suggests in the first month. Treat the defaults here as a target you reach after tuning, not a launch figure.
Should I use a platform or build on the API?
A platform if you want it live in weeks and are happy with what it can do; the API if the bot needs to act inside systems you already run. The platform fee is visible and the API is cheaper per conversation, so the crossover depends almost entirely on volume.
What is not in this number?
Per-resolution charges where a platform has them, the content work to make a knowledge base worth retrieving from, and your team's time answering everything the bot gets wrong in the first month. That last one is real and it is always underestimated.
Where to go from here
- WhatsApp Business API Cost Estimator — if the bot answers on WhatsApp, this prices the messages it sends.
- Lead Response Time Grader — the case for a bot is often speed rather than saving. This grades what you have now.
- CRM and Automation Cost Estimator — prices the build around it: the workflows and integrations the bot triggers.
The guide behind this tool
Built by GoDesign FZE, who build AI assistants and automation.
Get in touch
Want someone to just do this part?
The tool is free and stays free. If you would rather hand the work over, tell us what you are dealing with and you get a straight answer on scope, cost and timeline.
- You own everythingRepository, hosting and domain credentials sit in your name from day one, not handed over at the end.
- A written timelineMilestone dates are agreed in writing before work starts, so you always know what ships next.
- One business dayEvery enquiry gets a reply from the person who would scope the work, not a sales sequence.
Media City, Dubai, UAE · DHA Phase 2, Islamabad, Pakistan
Would rather type than fill in a form?
Message on WhatsApp