Every few months a business owner asks whether it would be cheaper to have a developer build a custom chatbot than to pay for a platform. On paper the answer looks obvious: the model APIs are cheap, and a prototype takes a weekend. In practice, the model is a small part of a chatbot that customers can rely on. This guide breaks down what a custom build really involves, what an off-the-shelf platform gives you, how the costs compare over a year and when each choice makes sense.
What is inside a business chatbot
A demo shows a chat window and a model. A working business assistant has many more parts, and each is real engineering.
| Component | What it does | Why it is harder than it looks |
|---|---|---|
| Content ingestion | Reads PDFs, Word files, web pages and text | Messy formats, scanned files, tables and updates |
| Chunking and indexing | Splits content and stores it for search | Poor chunking causes wrong answers; see chunking strategies |
| Retrieval | Finds the right passages | Needs hybrid search and tuning; see hybrid retrieval |
| Grounding and guardrails | Makes the model answer only from your content | Testing and constant refinement; see preventing hallucinations |
| Chat widget | The window on your site | Style isolation, mobile, accessibility, performance |
| Booking engine | Real availability, slot locking, confirmations | Concurrency, time zones, buffers; see double booking prevention |
| Lead capture and delivery | Stores leads and alerts you | Email, webhooks, CRM |
| Human handover | Lets your team join | Notifications, inbox, history |
| Security and privacy | Keys, access, isolation, data deletion | Ongoing responsibility |
| Analytics and review | Shows conversations and gaps | Dashboards and tooling |
| Channels | WhatsApp, Telegram and others | Each has its own API and rules |
A weekend prototype covers two or three rows. A production assistant covers all of them and keeps covering them as models, APIs and browsers change.
What buying gives you
- Speed: live in hours or days.
- A complete feature set that already handles the components above.
- Maintenance done for you: model changes, browser changes, security patches.
- Predictable cost: a subscription instead of development invoices.
- Support and documentation.
- Continuing improvement as the vendor ships features.
The trade-offs are less control over unusual features, dependence on the vendor's roadmap and the need to accept their data practices. Review a vendor's privacy policy, sub-processors and export options before you commit.
What building gives you
- Full control over behaviour, design and data location.
- Custom integrations with internal systems that a platform does not support.
- Ownership of the code and no per-seat fees.
- Differentiation if the chatbot is the product you sell.
These are real advantages for the right business. The cost is the engineering and maintenance burden.
A cost comparison over one year
These are illustrative ranges to show the shape of the comparison. Get real quotes.
| Item | Buy a platform | Build a custom assistant |
|---|---|---|
| Initial development | None | Weeks to months of developer time |
| Setup | Hours to days | Included in development |
| Subscription or hosting | Monthly plan | Servers, databases, monitoring |
| AI usage | Your own key or bundled | Model API costs |
| Maintenance | Included | Ongoing developer time for fixes and updates |
| Security and compliance work | Shared with the vendor | Entirely yours |
| Feature additions | Vendor roadmap | Each one is a project |
A typical small business platform costs from tens to a few hundred dollars a month. A custom build that reaches the same level of function can run to many months of developer effort, plus maintenance that never ends. Unless the chatbot is core to what you sell or you have unusual requirements, the platform is almost always cheaper for the first few years.
Hidden costs of building
- Maintenance: models, libraries and APIs change; something breaks every few months.
- Quality tuning: getting answers right is an ongoing process, not a one-off task.
- Security: you are responsible for key storage, isolation between customers and incident response.
- Bus factor: if the developer leaves, who understands it?
- Opportunity cost: time spent on the chatbot is not spent on your business.
Hybrid approach: buy the platform, customise the edges
Many businesses land here. Use a platform for the core, and extend it with your own logic where needed: webhooks to your systems, automation tools such as Zapier or Make, custom widget styling, your own AI provider. See webhook automations with Zapier and Make and customising widget branding with CSS. This gives you speed and maintenance from the vendor and flexibility where it matters.
When building makes sense
- The chatbot is your product, or a core differentiator.
- You have strict requirements on data location or on-premises hosting that no platform meets.
- You need deep integration with proprietary systems.
- You have a development team with capacity and a plan for maintenance.
- Your volume is so large that per-conversation platform economics no longer work.
When buying makes sense
- You are a service business and the chatbot supports your work rather than being it.
- You want to be live this week.
- You do not have developers, or their time is better used elsewhere.
- You want maintenance and security handled.
- You want to try before you commit.
A decision framework
- Write down the three jobs the chatbot must do.
- Check whether platforms you can trial already do them.
- Test two platforms on your own documents for an hour each.
- If a platform covers eighty percent of the need, buy and extend.
- If none does and the gap is essential, price a custom build honestly, including maintenance for three years.
- Decide with numbers, not with the excitement of a demo.
Our platform comparison guide provides a scorecard, and you can try Crescelo on your content in the playground.
Security when you choose a platform
Whichever route you take, ask about key storage, workspace isolation, two-factor authentication, data deletion and sub-processors. Crescelo stores AI provider keys encrypted, isolates workspaces and lists its legal documents, including a data processing addendum and sub-processors; it does not claim compliance certifications. Read how a bring-your-own-key vault works to see what good looks like.
A realistic timeline for a custom build
| Phase | What happens | Typical effort |
|---|---|---|
| Discovery | Requirements, content audit, choice of models and architecture | Days to weeks |
| Prototype | Basic chat with retrieval over a few documents | Days |
| Hardening | Guardrails, evaluation set, fallbacks, error handling | Weeks |
| Product features | Widget, booking, leads, handover, admin | Weeks to months |
| Security and privacy | Key storage, isolation, deletion, logging, review | Weeks |
| Launch and tuning | Real traffic, fixes, content improvements | Ongoing |
The prototype phase is the one that looks cheap, and it misleads: most of the effort and cost sits after it. That is why estimates for custom builds so often grow, and why platforms that have already done this work are attractive.
Questions to ask a developer or agency
- Who will maintain it in a year, and what will that cost per month?
- How will we measure answer accuracy, and how often?
- How are keys, data and customer workspaces protected?
- What happens if the model provider changes prices or deprecates a model?
- How will bookings avoid double booking under load?
- What do we own at the end, and how do we move it if we change supplier?
If the answers are vague, treat the estimate as a lower bound.
Example: two businesses, two decisions
A three-location dental group
The group wants patients to ask about treatments, insurance and hours, and to book with the right dentist. It has no developers. It chooses a platform, loads its documents, connects calendars and limits what the chat collects. Time to launch: two weeks including approvals. Total cost is a monthly plan plus AI usage.
A software company selling a chatbot product
The company's product is a chatbot for a niche industry, with proprietary data sources and customers who need on-premises hosting. Here the chatbot is the product, so it builds. It has an engineering team, an evaluation process and a support function. Buying a platform would make its product a reseller of someone else's.
The same question gives opposite answers because the businesses are different. Be honest about which one you are.
Frequently asked questions
Is it cheaper to build my own AI chatbot?
Rarely for a service business. The model API is cheap, but ingestion, retrieval, guardrails, booking, handover, security and maintenance add up to a lot of engineering.
How long does it take to build a custom chatbot?
A prototype takes days. A reliable business assistant with booking and handover takes months, and needs ongoing maintenance.
Can I customise an off-the-shelf chatbot?
Yes, through settings, styling, webhooks, integrations and your choice of AI provider.
What are the risks of building my own?
Wrong answers, security gaps, maintenance burden and dependence on a few people.
When should a business build a custom chatbot?
When the chatbot is a core product, when strict data requirements rule out platforms, or when integration needs are unique and you have the team to maintain it.
How can I test a platform quickly?
Load your real documents, ask thirty real questions including some it cannot answer, and try a booking. The playground lets you do this without an account.
See how the AI assistant feature works and start a free trial to compare it with your build option.
Further reading
- Google Search Central: creating helpful, people-first content: What Google looks for in useful pages.
SVARA ALLA BESÖKARE. BOKA DEM SOM ÄR REDO.
Lägg in Crescelo på din webbplats på 15 minuter
Träna din assistent på dina prislistor, dokument och tjänstevillkor. Aktivera välgrundade svar dygnet runt och bokning av tider direkt i chatten utan krockar.