DIY vs Agency Chatbot: What You Are Really Paying For

I am an agency. You should read this page with that in mind, and I am going to try to earn your trust by telling you when not to hire one.

The truth is that a lot of small businesses should build their own chatbot. The tools got good. A capable owner with a free weekend can get something answering questions on their website, and it will be better than the contact form it replaces. I would rather you know that up front than pay me for something you could have done yourself in an afternoon.

What I will also tell you is where DIY builds reliably fall apart, because I have been called in to fix enough of them to see the pattern.

What you are actually buying

Nobody is selling you a chatbot. The software costs what it costs whether you or I click the buttons. What an agency sells is three things:

1. Knowing which trap you are about to walk into. Most of my value on a build is not construction, it is avoiding two weeks of confusion over a problem I have already hit.
2. The integration work. Booking into a real calendar, writing into a real CRM, transferring a live call to the right department, forwarding a carrier phone system correctly. This is where DIY builds stall.
3. Somebody who keeps reading the transcripts in month seven. This is the one people underestimate most and the one that determines whether the bot is still running in a year.

If you can supply all three yourself, do not hire anyone.

Side by side

	DIY	Agency
Up-front cost	Platform fee only	Setup fee in the hundreds, plus platform
Ongoing cost	Platform fee	Retainer, typically $99 to a few hundred monthly for small businesses
Time to first live version	A weekend, if you are handy	Days to a couple of weeks
Time to a good version	Months, learning as you go	Weeks
Business knowledge in the bot	Yours, which is the best possible source	Extracted from you, which is lossy
Integrations	You are on your own with APIs	Included, and this is most of the labor
Phone plumbing	Genuinely hard the first time	Routine
Testing rigor	Usually "it worked when I tried it"	Should be systematic, ask them how
Maintenance	You, when you remember	Contracted
When it breaks at 4pm Friday	You	Them
Risk of abandonment	High, this is the real DIY failure mode	Low while you are paying

What DIY does better

You know your business. This is not a small thing. The single biggest quality driver in a chatbot is the accuracy and depth of what it knows, and you have all of that in your head already. My first two weeks on any client build are spent extracting knowledge that the owner could have typed out in an evening. Some of it never gets fully extracted, and the bot is worse for it.

Iteration speed. You see a bad answer, you fix the prompt, done. No ticket, no email, no waiting. For a bot that is mostly answering questions, that loop is worth a lot.

Cost. Obviously. A DIY bot on a product platform costs the platform fee and your time.

You learn the system. Which means the next thing you want to automate is easier, and you can call your own shots on vendors.

For a straightforward website FAQ bot on documents you already have, DIY is often just the right answer. Start there. If it works, you have saved yourself money and learned something.

Where DIY builds reliably break

Here is the pattern I see when I get called in to fix one.

It never got tested properly. The owner tried it a few times, it worked, it shipped. Then real users did something unexpected. For voice this is worse, because simulator testing cannot inject system variables like caller ID, so an entire class of bug only appears on a real phone call. My hard rule is to call and text the live number from my own cell before any rollout, and to replay real production records through the deployed system rather than trusting a passing test. Green tests once hid a wiring bug that cost a client eleven leads. Testing voice agents has the checklist I use.

It confidently makes things up. Every unguarded agent invents capabilities. I had a receptionist tell callers it had blocked a phone number for them, an ability it never had. The fix was an explicit "abilities you do NOT have" list written into the prompt. A DIY builder usually does not know to write that section, because you only learn it after it happens to you in front of a customer. See stopping chatbot hallucinations.

The phone part never quite worked. Voice DIY dies in telephony, not AI. One client's carrier phone system needed group forwarding configured rather than its auto-receptionist feature before calls would reach the AI at all. There is no tutorial for your specific carrier. Phone forwarding for AI agents covers the common cases.

Silent failures nobody notices. Dead air on answer that looks like a carrier fault but is a config bug. Leftover {{placeholder}} tokens sitting in the instructions, quietly degrading the prompt with no error anywhere. A transfer tool pointing at a hardcoded value instead of the live call's control ID, so transfers fail and callers are stranded. All of these fail quietly. If nobody is reading transcripts, they run for months.

It got abandoned. This is the most common ending. The bot works, everyone moves on, the business changes, the bot keeps confidently quoting last year's prices. Six months later someone turns it off. Not because the technology failed, because nobody owned it.

What the retainer actually buys

When a client pays me a few hundred a month, this is the work, and you should hold any agency to the same list:

• Reading real transcripts every week at first, then monthly, and fixing what shows up.
• Updating the knowledge base when prices, hours, services, or staff change.
• Watching for platform changes: model deprecations, pricing shifts, connector removals.
• Escalation quality: making sure handoffs to a human actually connect to a human.
• Reporting on what people asked, what the bot could not answer, and what got booked. Chatbot analytics covers the metrics worth tracking.

If an agency cannot describe their ongoing work in concrete terms like that, you are buying a build and calling it a retainer.

Questions to ask any agency

Ask these. The answers separate operators from resellers.

• What is running in production for you right now, and how long has it been live?
• How do you test before rollout? Do you place real calls or send real texts?
• What happens when the bot says something wrong to my customer? Who notices, and how fast?
• Do I own the account and the number, or do you?
• What does month four look like? Show me a change you made to an existing client bot last month.
• If I leave, what do I get to take with me?

That last one matters more than people expect. Prompts, flows, and knowledge bases do not export cleanly between platforms, so leaving is often a rebuild. Know that going in.

Pick DIY if

• Your bot's job is answering questions from documents you already have.
• Nobody needs it to book, transfer, or write into another system.
• You are comfortable poking at settings and reading a vendor's docs.
• You will actually schedule time to review conversations, monthly, on a calendar.
• Budget is genuinely tight, in which case a DIY bot beats no bot by a wide margin.

Pick an agency if

• Voice is involved. Telephony is where DIY loses weeks.
• It must integrate: calendar, CRM, ordering system, custom API.
• There is compliance exposure, like consent rules for texting. See TCPA compliance for AI SMS.
• The cost of one embarrassing wrong answer to a customer is high.
• You know, honestly, that you will not maintain it. This is the most common real reason and the most valid one.

Verdict

Build it yourself first if it is a website FAQ bot. Seriously. The tools are good, you know your business better than anyone you could hire, and the fastest way to learn what you actually need is to run something imperfect for a month and read what people ask it.

Hire an agency when the bot has to do things rather than say things: answer the phone, book the job, write into your system, honor a consent rule. That work is integration and operations, not prompt writing, and integration is where unmanaged builds quietly die.

And whichever you choose, name an owner. A chatbot with no owner degrades on a predictable schedule. That is true whether the owner is you or someone you pay, and it is the closest thing to a rule I have in this business. For the related question of which platform layer to build on, see build vs buy an AI chatbot, and for the money, how much does an AI chatbot cost.

FAQ

How much does an agency chatbot actually cost? For small businesses, the shape I see is a setup fee in the hundreds and a monthly retainer from around $99 to a few hundred, plus platform and usage costs. Voice usage runs cents per minute on top. Be suspicious of a large setup fee with no retainer, because that structure means nobody is maintaining it after launch.

Can I build it myself and hire someone just for the hard part? Yes, and this is an underrated option. Own the prompt and knowledge base yourself, since that is your business knowledge, and pay for the integration and phone plumbing. Just agree clearly on who owns what, or you will get finger-pointing the first time something breaks.

How long until a DIY bot is actually good? Live in a weekend, good in a few months. The gap is entirely reading real conversations and fixing what you find. If you skip that loop, it never gets good, it just gets old.

What is the most common reason a DIY chatbot gets turned off? Abandonment, not failure. It works, attention moves elsewhere, the business changes, the bot keeps giving last year's answers, and eventually someone decides it is embarrassing. Almost none of the ones I have been called in on were broken. They were stale.

Do agencies lock you in? The platform does more than the agency does, since prompts and flows rarely move between tools. Reduce your exposure by insisting the accounts and phone numbers are in your name from day one. Any agency that refuses that is telling you something important.