AI Chatbot Consulting Pricing: What Setting One Up Costs You

There is no standard rate: chatbot consultants quote a fixed project fee or an hourly rate, and small-business set-ups vary widely. As an illustration, a simple FAQ bot of 10 to 20 hours at $100 to $150 an hour is $1,000 to $3,000, plus the platform's own monthly fee, which on the plans below runs from $32 to $150 or more.

Those figures are assumptions, not market data, and I could not find a trustworthy published price list for this work. What you can do is size the job yourself: count the hours each stage needs, multiply by whatever rate you are quoted, and add twelve months of platform fees. That sum, not the consultant's headline price, is what the bot costs you in year one. The rest of this tutorial gives the stages, three worked scenarios, and the questions that stop a quote from growing.

Follow me on Instagram@sagnikteaches

What the hours are actually spent on

People assume a chatbot project is mostly technical. It is mostly editorial. Connecting a widget to your site takes an hour; deciding what the bot may say takes days. A typical small-business project divides like this, with hours that are illustrative for a bot answering opening hours, prices, bookings and policies:

Connect on LinkedInSagnik Bhattacharya

The total, 14 to 32 hours, explains the wide spread in quotes. A consultant who skips content clean-up quotes low and delivers a bot that repeats your out-of-date prices. The tutorial on building the FAQ your chatbot needs is the clean-up stage in detail, and it is also the stage you can do yourself to cut the bill.

Subscribe on YouTube@sagnikteaches

Three set-ups priced for a year

Say an optician's practice wants a website bot to answer questions about appointments, frame brands, contact lens orders and opening hours, with about 400 conversations a month. The year-one cost depends on who does what. Each total below uses the vendor prices listed later and a made-up consultant rate of $120 an hour.

Two points hide in this table. First, the platform is a bigger yearly cost than the setup in scenario B and C, once you add up twelve months. Second, scenario A is cheap only if your time is: twelve hours at your own earning rate may exceed $384 by far. Whether that is a good trade depends on what else you would do with the twelve hours.

Scenario A's message credits matter too. Chatbase counts message credits, and Hobby gives 700 a month. If a conversation of four exchanges used four credits, 400 conversations could use 1,600 credits and the Hobby plan would not hold; Basic's 4,000 would. Check how your platform counts, then price the real volume. The tutorial on website chatbot cost and payback works through what a visitor-to-enquiry bot earns back.

Platform fees that sit on top of any consultant

Here are the prices on the vendors' own pages in November 2026. All are USD list prices; check each page before you rely on them.

• Chatbase: Hobby $40 a month with 700 message credits and one agent, Basic $150 with 4,000 credits, Business $500 with 15,000 credits. Yearly billing takes 20% off, which makes them $32, $120 and $400 a month, the figures the scenarios use. Auto-recharge tops up credits that do not expire when you fall below a threshold you set; the pricing page does not show its rate, so check it in the dashboard.
• Botpress: Free with 100 conversations and no top-ups; Plus $150 a month billed annually, with 250 conversations a month and $25 of AI usage; Team $750 billed annually, with 1,500 conversations. Extra conversations come in packs of 100 at $65 on Plus and $50 on Team, added automatically at 95% of your allowance, and the auto-recharge cannot be turned off.
• Tidio: Free includes 50 lifetime AI conversations; Starter $24.17 a month and Growth from $49.17 (annual-billing prices); the Lyro AI add-on starts at $32.50 a month for 50 conversations. The page does not give a per-conversation rate for extras.
• Intercom Fin: $0.99 for each outcome. Used with another help desk it has a $49 monthly minimum that includes 50 outcomes, and no seat cost. An outcome is a conversation the customer confirms is resolved, one where they ask for nothing more after Fin replies, or a completed workflow, including one that ends in a hand-off; if the customer asks for a person, nothing is charged.

A per-outcome tool can look cheap or expensive depending on the count. For our 400 conversations, say 60% end as outcomes: 240 x $0.99 is $237.60 a month, or about $2,851 over a year, with no plan fee. Botpress in scenario C would be $3,360 and Chatbase Basic $1,440. The comparison shifts again if your bot hands off more of its conversations, since some vendors count hand-offs as outcomes. The tutorial on what platforms charge per resolved chat compares the billing units in more depth.

DIY or paid: a plain way to decide

Do it yourself when the bot answers a closed list of questions, your content already exists in writing, and a wrong answer costs little. A pet shop bot that says "we open at nine, we stock these brands, delivery takes two days" fits. Pay for help when the bot books, takes payments or sees customer records, because mistakes there cost real money; when the content is a mess of contradictory policies that someone must resolve; or when your own time is worth more than the consultant's.

A middle route is common and cheap: pay for the scoping and testing stages only (5 to 10 hours) and do the content writing yourself. The consultant's value is in the questions you would not think to ask and the test cases you would not think to try. For choosing between a solo freelancer, a small agency and a consultant, see the comparison of the three.

Reading a quote line by line

A vague quote reads "Chatbot build and training: $2,500." A useful one lists what is inside. Here is an illustrative quote for scenario B, with the lines that most often go missing.

Fixed fee: $1,800 (15 hours at $120)

Included
1. Workshop to list the 40 most common customer questions (2 hours)
2. Rewrite of up to 25 source pages into bot-ready answers (6 hours)
3. Bot set-up on a platform in the CLIENT'S account (2 hours)
4. Written bot instructions: tone, boundaries, what it must never answer (1 hour)
5. Hand-off rules to a person, including out-of-hours (1 hour)
6. Testing with 50 questions and 2 rounds of fixes (2 hours)
7. Handover guide and 30 minutes of staff briefing (1 hour)

Not included
- Platform fees, extra integrations, new content after launch
- Changes requested after round 2 (billed at $120 an hour)

Payment: 50% on start, 50% on handover

Look for what is not there. Did it say whose account the bot lives in? Is the testing counted in hours or just promised? Does "up to 25 source pages" match how many pages you actually have? A quote without a "not included" list is a quote that has not finished thinking.

Eight questions that stop a quote growing

1. Which platform will you use, and why that one for my volume?
2. What do I pay the platform each month once it is running, including what happens above the allowance?
3. Whose account is the bot built in, and will I get admin access on day one?
4. How many test questions will you run, and will I see the results?
5. What happens when the bot cannot answer, and how is that tested out of hours?
6. What is not included, and what does each extra round cost?
7. What do I receive at the end, in editable form?
8. Who updates the answers in six months, and at what rate?

The tutorial on what to ask an AI chatbot vendor covers the data side of this, such as where conversations are stored and whether they train anyone's model.

A scope that crept: how it happened

An illustrative case. A gift shop agrees a $1,500 fixed fee for a returns-and-delivery bot. Midway, the owner asks if it can also check order status. That means an integration with the shop system, which the quote never mentioned. The consultant, fairly, bills 12 more hours. The owner feels cheated, the consultant feels unpaid, and the bot launches a month late.

What would have prevented it is one line in the original quote: "Order status lookups are not included; quoted separately if wanted." A good consultant writes it without being asked, and you can ask for it. When a quote cannot draw a line around the work, ask for a phased plan instead: phase one answers questions only, phase two connects the shop system once phase one has run for a month.

Prepare the questions before the consultant arrives

Every hour you save on discovery is an hour you do not pay for. Collect your real customer questions first, using a prompt like this on an export of your recent emails or chat messages, with names and contact details removed.

Here are 200 customer messages (names and contact details removed). Group them by topic. For each topic give: the count, three example wordings, and whether a written answer already exists in the text I pasted. Then list the 40 most frequent questions as short standalone questions. Do not invent answers; mark any question I need to write an answer for.

A reply might show "Opening hours and bank holidays: 31 messages", "Contact lens re-orders: 24 messages", "Frame repairs: 12 messages, no written answer found". That last tag is gold: it shows where content must be written before any bot is built. Walking a consultant through this list can shorten scoping from four hours to one.

Testing before you pay the final instalment

Hold the last payment until the bot passes a test you wrote. Fifty questions is enough: 30 normal ones, 10 that should make the bot hand off, and 10 designed to catch it making things up, such as asking for a discount you do not offer or a product you do not stock. Score each as right, wrong or unsafe, and agree in advance the pass mark, for instance 90% right and none unsafe.

In an illustrative case, a garden centre that tested this way finds the bot would confirm a "10% loyalty discount" that appeared in an old promotional page it had been fed. The answer was correct for last year's leaflet. The fix was to remove the page from the knowledge base, not to adjust the prompt, which is the kind of finding a content-first test turns up. Also tell visitors they are talking to an AI; the tutorial on writing an AI disclosure statement gives wording.

One answer, before and after clean-up

Content clean-up is the stage owners skip and consultants charge for, so here is what it looks like on a single answer. A pet shop's website says three different things about delivery: the delivery page says "free over $50", an old news post says "free over $40", and the returns page says "delivery is $4.95 on all orders". A bot fed all three will give whichever its search finds first.

BEFORE (three sources, contradictory)
- Delivery page: free over $50
- News post, 2023: free over $40
- Returns page: delivery is $4.95 on all orders

AFTER (one source of truth, written for the bot)
Q: How much is delivery?
A: Delivery is free on orders over $50 and $4.95 below that. Orders placed before 2 p.m. leave the same working day. If you need it sooner, tell me what you are ordering and I will pass you to the team.
Source of truth: delivery page, checked by [owner] on [date]. Delete the news post and fix the returns page.

That one fix took about ten minutes, and it is multiplied across 25 to 40 answers. It explains why content clean-up is 4 to 10 hours in the table above, and why a quote that does not mention it is suspicious. You can do this work yourself before a consultant starts, which is the cheapest saving on the list.

Three businesses, three different scopes

An optician with a website of 12 pages. Questions cluster around appointments, eye test prices, contact lens orders and frame repairs. Scope: scoping 2 hours, content 5 hours, build 3 hours, testing 3 hours, handover 2 hours, so 15 hours. The risk is clinical: the bot must never advise on symptoms, only offer to book or pass to a person. The consultant's contribution is writing that boundary and testing it.

A pet shop with an online store. The questions are about stock, delivery and ingredients, and they change weekly. A bot that answers from a static document will age within a month. Scope should therefore include how the answers are refreshed, whether by pointing the bot at the live product pages or by a weekly update routine you can actually keep. Hours are similar, but the retainer matters more than the build.

A garden centre with seasonal opening hours and click-and-collect. Hours change by season, and plant availability changes daily. Keep stock questions out of the bot entirely ("I cannot see live stock; ask the team on this number") and let it answer hours, delivery windows, advice pages and click-and-collect rules. A narrower scope is a cheaper and safer bot, and it is a perfectly good v1.

Ways to lower the bill without lowering the quality

1. Bring the 40 questions with you. Pulling them from your inbox saves two to three hours of discovery.
2. Give the consultant clean source documents, in text, with the contradictions marked.
3. Start with answers only. Add booking or order lookups in a second phase after a month of real conversations.
4. Hold the platform choice open until the test. A free plan can run the whole test before you pay anyone for a year.
5. Agree a fixed number of revision rounds. Two is normal; unlimited rounds are how projects overrun.
6. Do the monthly review yourself for the first three months, with the consultant checking only the first.

Take scenario B and apply three of those: you bring the questions (saves 2 hours), the sources are clean (saves 3 hours), and you do the first reviews yourself (saves 4 hours of quarterly review). That removes 5 hours of one-off work, $600, and $480 of review, so the year-one cost falls from $4,200 to $3,120. None of it reduces how well the bot answers.

Warning signs in a proposal

• A price with no hours, deliverables or exclusions. You cannot tell what is included, so you cannot hold anyone to it.
• A promise of a percentage of enquiries handled or sales lifted. Nobody can know that before the bot meets your customers. Ask what they would measure instead, such as share of conversations resolved without a person and the number handed off.
• A bot built in the consultant's account. You may lose it, and your conversation records, if the relationship ends.
• No testing described. If the plan jumps from build to launch, testing is on you.
• A one-year platform contract signed before the test. Ask for a monthly term until you have seen real conversations.
• A bot that says it is a person. Visitors should be told it is an AI, and in some places the law expects it.

Running costs after launch

Allow time each month. A few hours a quarter is typical for reading conversations the bot got wrong, adding answers for new questions, and updating prices and hours. If you skip it, the bot ages: it will quote last season's opening hours with complete confidence. Treat the review as a fixed line in the budget, either your own time or a retainer with the consultant, and note its cost beside the platform fee.

The honest total for a modest, well-built website bot is therefore a one-off project cost, a platform fee you can predict from your volume, and a small recurring review. If you would like help scoping which of those applies to you before you ask anyone for quotes, that is what my AI implementation consultation covers.

Questions to settle before you hire anyone

Should a fixed fee or an hourly rate be my default?

Fixed fee for a clearly written scope, hourly for anything exploratory. A fixed fee protects you from overruns, but only if the scope lists what is included: pages of content prepared, number of test questions, integrations, number of revision rounds. Without that list, a fixed fee just moves the argument to the end.

Who should own the platform account?

You. Create the account with your own email and payment card, then add the consultant as a user. If the consultant creates it under their account, moving to another provider later can mean rebuilding from scratch and asking for your own data back.

Is the monthly platform fee usually the bigger cost?

Over a year, for a small business, often yes. A modest plan at $120 a month is $1,440 a year, which can exceed a simple one-off set-up. Overage rules matter more than the headline fee, so price your real conversation volume, not the plan name.

Can I start with a free plan and upgrade?

Yes, and it is a sensible test. Free plans have small caps, such as 100 conversations on Botpress's free plan or 50 lifetime AI conversations on Tidio, and some limit features. Use one to test your content, then price the paid plan with real numbers.

What should a consultant hand over at the end?

The cleaned knowledge base in an editable file, the written bot instructions, the list of test questions with results, the escalation rules, logins in your name, and a one-page guide for updating answers. If any of those is missing, you will be paying again the first time something changes.

Further reads

• AI Chatbot for WooCommerce: Costs, Order Lookup and a Test Script — Costs and a test script for an online shop bot.
• How to Build a Company Knowledge Base AI Can Answer From — Build the content the bot answers from.
• How to Add an AI Chatbot That Captures Leads on Your Website — Set a bot to collect enquiries, not only answer them.
• Can an AI Chatbot Handle Order Tracking and Returns? — Whether a chatbot can handle order tracking and returns.
• AI Consultant or DIY: Which Suits a Small Professional Firm? — Another view of the hire-or-do-it-yourself decision.
• AI Consultant vs In-House Hire: Which Costs a Small Team Less? — When a part-time consultant beats a permanent hire.
• Personal Injury Lawyer Chatbot: Intake Questions, Limits and Cost — An intake script, disclaimer wording, handoff triggers and monthly cost bands from named vendors for a personal injury firm's website chatbot.
• AI Chatbot Development Company, Freelancer or Platform? A Small-Firm Choice — Compare a chatbot development company, a freelancer and a chatbot platform on scope, ownership of the bot, ongoing fees and contract terms.
• AI Chatbot Setup for Small Businesses: A Two-Week Pilot Plan — A day-by-day two-week pilot: build a knowledge sheet and one channel, go live on a share of chats, then review transcripts before you scale.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the pricing pages of Chatbase, Botpress, Tidio and Intercom, read in November 2026. Consultant hours, rates and conversation volumes are illustrative assumptions, not published rates.