Contents
- Path 1: SaaS chatbot platform
- What it costs
- What you get
- What you don’t get
- Who this works for
- Path 2: Hire a studio to build it
- What it costs
- What you get
- What you don’t get
- Who this works for
- Path 3: Build in-house
- What it costs
- What you get
- What you don’t get
- Who this works for
- The decision matrix
- The hidden costs that change the math
- Hidden cost 1: Integration complexity
- Hidden cost 2: Training data and knowledge management
- Hidden cost 3: The accuracy gap
- Hidden cost 4: What “good enough” costs when it isn’t
- Hidden cost 5: Switching costs
- The question that actually matters
You just got budget approval. The exec team wants an AI chatbot for business operations, customer support, lead qualification, or some combination of all three. Congratulations. Now you have three options on the table, and each one will cost you a different amount of money, time, and organizational patience.
Option A: Buy a SaaS chatbot platform. Intercom, Tidio, ChatBot, Zendesk, or one of the fifty others. Plug it in, configure it, go live.
Option B: Hire a studio or agency to build a custom chatbot. They scope it, build it, hand it off. You own the code.
Option C: Build it in-house. Your engineers, your architecture, your timeline.
The internet will tell you that option A is the cheapest, option C gives you the most control, and option B is the middle ground. That’s roughly true and completely useless for making an actual decision. The real comparison requires looking at what you’re actually paying for, what you’re giving up, and what happens eighteen months from now when the chatbot needs to do something it wasn’t built to do.
Path 1: SaaS chatbot platform
What it costs
SaaS chatbot pricing has fractured into three models in 2026: per-seat, per-resolution, and hybrid. These are the current rates from the major players.
Per-seat pricing:
- Intercom Essential: $29/seat/month (annual). Advanced: $85/seat/month. Expert: $132/seat/month. Their AI agent (Fin) adds $0.99 per resolution on top.
- ChatBot by Text: $19/user/month (annual) for 10 AI resolutions/month. Growth tier: $79/user/month for 200+ resolutions. Extra resolutions cost $0.99 each.
- Tidio Starter: $24/month for 100 conversations and 50 AI conversations (one-time). Growth: from $49/month. Enterprise (Plus): from $749/month.
Per-resolution pricing:
- Intercom’s standalone Fin AI Agent: $0.99 per resolved conversation, $50/month minimum. No seat fees.
- Most platforms are moving toward this model because it aligns cost with value delivered.
What a realistic monthly bill looks like:
For a mid-market company running 2,000 support conversations per month with a 5-person support team:
| Platform tier | Monthly cost | Annual cost |
|---|---|---|
| Basic SaaS (Tidio Starter level) | $200 to $400 | $2,400 to $4,800 |
| Mid-tier SaaS (Intercom Advanced, 5 seats + Fin) | $1,400 to $2,400 | $17K to $29K |
| Enterprise SaaS (custom pricing) | $3,000 to $8,000 | $36K to $96K |
What you get
Speed. A SaaS chatbot can go live in days. The AI-powered ones (Intercom Fin, Tidio Lyro, Zendesk AI) will crawl your help center, learn your knowledge base, and start answering questions with minimal configuration.
You also get maintenance handled for you. The vendor manages uptime, model updates, and security patches. Your team configures flows and monitors performance. That’s a real advantage if you don’t have ML engineers on staff.
What you don’t get
Customization beyond what the platform allows. You’re renting someone else’s architecture. The chatbot will do what the platform was designed to do, and nothing more. If your use case requires pulling data from a proprietary system, applying custom business logic, or triggering actions in an internal tool, you’re hitting the platform’s walls fast.
You also don’t get data portability. Your conversation history, training data, and flow configurations live on someone else’s servers. Switching platforms means starting from scratch.
And the pricing scales with usage, not value. If your chatbot successfully deflects 3,000 tickets instead of 2,000, your bill goes up. You’re paying more for getting better results. That math works in the vendor’s favor, not yours.
Who this works for
Companies where the chatbot’s job is clearly defined, well-covered by existing platform capabilities, and doesn’t require deep integration with proprietary systems. Customer support deflection for a product with good documentation is the sweet spot. If 80% of your support tickets are some variation of the same twenty questions, a SaaS platform handles that well.
Through the 10X Framework lens. Three lenses: impact (which number it moves, and how much of the work it takes on, 1 to 5), trust to earn (what a wrong answer costs, who decides, and how sensitive the data is, written in words from Little through Some, Real, A lot, to The most, because a number there invites averaging and strong impact should never cancel out a trust problem), and readiness (whether the data and a named owner already exist, 1 to 5). The combination lands in one of four tiers: build now, proof first, build with guardrails, or not yet.
SaaS chatbots read high on readiness (4-5) and have little trust to earn, since a wrong answer costs a correction and nothing is decided about a person. Impact is moderate (2-3), because the ceiling on customization caps the upside. Moderate impact with little trust to earn and strong readiness lands at Tier 2, proof first: run it on one queue, count the deflections, then widen it. For anything requiring custom logic or deep integrations, the impact reading drops and you’re looking at Tier 4, not yet, on this path. (For a walk-through of the framework applied to three real client scorecards, see our 10X Framework deep-dive.)
Path 2: Hire a studio to build it
What it costs
Custom chatbot development from a specialized studio typically falls into three tiers:
| Complexity | Cost range | Timeline | What’s included |
|---|---|---|---|
| Basic NLP chatbot (single data source, FAQ-style) | $15K to $40K | 4 to 8 weeks | LLM integration, one knowledge base, basic web interface, deployment |
| Mid-complexity chatbot (2-3 integrations, workflows) | $40K to $100K | 8 to 16 weeks | Multi-source RAG, API integrations, conversation flows, testing, 60-90 day support |
| Advanced chatbot or hybrid (agent-like capabilities) | $100K to $200K | 12 to 24 weeks | Multi-system integration, conditional logic, human handoff, analytics, compliance docs |
The biggest variable isn’t the AI. It’s integrations. Every API connection, every data transformation, every edge case in your legacy systems adds weeks and dollars.
What you get
A system built for your specific workflow, not adapted from a generic template. The chatbot understands your data model, connects to your systems, and handles the exact scenarios your users encounter.
You also own the code. No per-seat fees, no per-resolution charges, no vendor lock-in. Your ongoing costs are hosting (typically $500 to $2,000/month for a mid-complexity chatbot) and maintenance.
Good studios also bring something your internal team might not have: pattern recognition from building dozens of chatbots across industries. They’ve seen the failure modes. They know which architectures hold up at scale and which ones create technical debt that surfaces six months later.
What you don’t get
Instant gratification. Even a basic custom build takes 4 to 8 weeks. A mid-complexity one takes 3 to 4 months. If you need something live next week, this isn’t the path.
You also don’t get a self-maintaining system. After the studio hands off the project, your team is responsible for keeping it running. Model performance degrades over time. Your knowledge base changes. APIs release breaking updates. Budget for ongoing maintenance at 15 to 25% of the initial build cost per year.
And there’s vendor risk of a different kind. You’re dependent on the studio’s quality, communication, and follow-through during the build. A bad studio engagement doesn’t just waste money. It wastes 3 to 6 months of your roadmap.
Who this works for
Companies where the chatbot needs to do something a SaaS platform can’t: connect to proprietary systems, apply custom business logic, handle domain-specific conversations that require specialized training data. Also companies that want to own the asset long-term but don’t have ML engineers or AI architects on staff.
10X Framework: Custom studio builds read high on impact (4-5) when the use case needs integrations a platform can’t provide. There’s some trust to earn, since the bot answers in your name, though a wrong answer costs a correction and nothing is decided about a person. Readiness is where this one turns: whether the systems the bot has to reach are actually reachable, and whether someone on your side owns the outcome once the studio hands over. Call it 3 for most companies, even with a clean scope and a proven vendor. Impact 4, some trust to earn, readiness 3 lands at Tier 2, proof first. With vague requirements and nobody named to own it, readiness drops to 2 and it’s Tier 4, not yet.
Path 3: Build in-house
What it costs
In-house chatbot development cost depends on your team’s existing capabilities. The honest math looks like this.
The team you need (minimum for a mid-complexity chatbot):
- 1 ML/AI engineer: $160K to $220K/year fully loaded
- 1 full-stack developer: $140K to $190K/year fully loaded
- 0.5 DevOps engineer (shared): $75K to $110K/year equivalent
- 1 product manager (partial allocation): $50K to $80K/year equivalent
That’s $425K to $600K/year in team cost. For a chatbot project that takes 3 to 6 months of their time, you’re looking at $110K to $300K in labor cost alone. On top of that:
- LLM API costs during development and testing: $2K to $10K
- Cloud infrastructure setup: $5K to $15K
- Third-party tools and libraries: $1K to $5K/month
Total in-house build cost: $120K to $330K for a mid-complexity chatbot. That’s before ongoing operating costs.
What you get
Maximum control. You choose the architecture, the models, the integrations, the data handling. No platform limitations. No vendor dependency. When your CEO walks in and says “can the chatbot also do X?” your team can evaluate and build it without negotiating a scope change with an outside party.
You also build institutional knowledge. Your team understands the system deeply because they built it. When something breaks at 2 AM, they know where to look.
What you don’t get
Speed. Your engineers have other priorities. The chatbot project competes with your product roadmap, infrastructure work, and every other initiative. What a studio ships in 10 weeks takes an internal team 4 to 6 months because they’re not working on it full-time.
You also don’t get the benefit of someone else’s mistakes. A studio that’s built forty chatbots knows, for example, that conversation context windows need to be managed carefully or the LLM starts hallucinating after fifteen turns. Your team discovers this at month three. That lesson costs time and money.
And there’s an opportunity cost that’s easy to ignore: every month your ML engineer spends on a chatbot is a month they’re not spending on your core product. If AI is your product, building in-house makes sense. If the chatbot is a support tool, you’re pulling your most expensive engineers off revenue-generating work.
Who this works for
Companies where the chatbot is core to the product, not a support function. If you’re building a conversational AI product, an AI-native customer experience, or a chatbot that needs to evolve weekly based on user behavior, you need it in-house. Also companies that already have ML engineers and AI infrastructure. If you’re hiring a team from scratch just for this project, you’re paying $300K+ before a single line of chatbot code gets written.
10X Framework: In-house builds read the highest possible impact (5) when the chatbot is a core product differentiator. There’s real trust to earn, since the bot speaks in your name at scale and there’s no vendor to escalate to when it gets something wrong. Readiness is the constraint. It turns on whether you already have the team (4-5) or need to hire (2-3), and hiring mid-build is how projects join the 42% of enterprise AI initiatives abandoned before they ship (S&P Global). A company with an existing AI team and a core product feature reads impact 5, real trust to earn, readiness 4, which is Tier 3, build with guardrails. For everyone else readiness sits at 2, and the math points back to Path 2.
The decision matrix
Every dimension that matters, in one table.
| Factor | SaaS platform | Studio build | In-house build |
|---|---|---|---|
| Upfront cost | $0 to $5K setup | $15K to $200K | $120K to $330K |
| Monthly ongoing | $200 to $8,000 | $500 to $4,000 (hosting + maintenance) | $3K to $13K (hosting + team allocation) |
| Year 1 total | $2,400 to $96K | $25K to $250K | $150K to $490K |
| Year 3 total | $7K to $290K | $40K to $400K | $250K to $750K+ |
| Time to live | Days to weeks | 4 to 24 weeks | 3 to 6+ months |
| Customization | Low to moderate | High | Maximum |
| Integration depth | Platform-limited | 2 to 5+ custom integrations | Unlimited |
| Code ownership | No | Yes | Yes |
| Maintenance burden | Vendor handles it | Your team (or retainer) | Your team |
| Scaling cost | Linear (usage-based) | Sub-linear (fixed infrastructure) | Sub-linear (fixed team) |
| Vendor lock-in | High | Low (you own code) | None |
| AI expertise needed | None | During build only | Permanent hire |
Notice how the Year 3 total changes the story. SaaS looks cheap in month one. By year three, a mid-tier SaaS platform costs $50K to $90K cumulative. A studio-built chatbot running on your own infrastructure costs roughly the same, but you own the asset and your costs flatten over time while SaaS costs keep climbing.
The hidden costs that change the math
Every path above has expenses that don’t show up in the initial quote. These are the numbers that separate the “what we budgeted” from the “what we actually spent.”
Hidden cost 1: Integration complexity
SaaS platforms advertise “500+ integrations.” What they mean is 500+ pre-built connectors that handle simple data flows. The moment you need a custom integration (and you will), you’re either hiring a developer anyway or paying the platform’s professional services team $200 to $400/hour.
For studio and in-house builds, integration work typically consumes 40 to 70% of the total project effort, based on what we’ve seen across dozens of engagements. A “simple” chatbot that connects to three legacy systems isn’t simple at all.
Hidden cost 2: Training data and knowledge management
Your chatbot is only as good as the knowledge it has access to. Most teams underestimate the effort required to prepare, structure, and maintain their knowledge base.
Expect to spend 40 to 80 hours up front organizing your documentation, FAQs, and product information into a format the chatbot can use. Then budget 5 to 10 hours per month to keep it current. That’s $2K to $4K/month in someone’s time, regardless of which path you chose.
Hidden cost 3: The accuracy gap
No chatbot gets everything right out of the box. You’ll spend the first 2 to 4 weeks after launch reviewing conversations, catching errors, and refining responses. For SaaS platforms, this means adjusting configurations and knowledge base entries. For custom builds, it means prompt engineering iterations and possibly retraining.
Budget for 2 to 3 months of active tuning after launch. The chatbot that goes live on day one and the chatbot that’s actually good on day ninety are meaningfully different systems.
Hidden cost 4: What “good enough” costs when it isn’t
IBM research puts the cost of a human-handled support interaction at around $6 and an AI-handled one at roughly $0.50. That 12x difference looks incredible until the chatbot gives a wrong answer that escalates into a 45-minute call with your most senior support rep, a negative review, or a lost customer.
If your chatbot resolves 80% of conversations correctly but handles the other 20% poorly, you’re not saving 80% on support costs. You’re saving 60% and spending 15% on damage control. The accuracy threshold where a chatbot actually saves you money net of error recovery is around 85 to 90%, depending on your industry and the cost of a bad answer.
Hidden cost 5: Switching costs
SaaS platform: if you outgrow the platform in 18 months, you’re starting from zero on a custom build. Every conversation flow, every integration, every piece of training data needs to be rebuilt.
Studio build: if you chose the wrong studio, the code they delivered might be so poorly structured that rebuilding is cheaper than maintaining. We’ve taken over projects where the “custom chatbot” was a tangle of hardcoded prompts and undocumented API calls. The rebuild cost more than the original build.
In-house: if the engineer who built it leaves, you’re dependent on their documentation quality. In our experience, about 30% of in-house AI projects have documentation sufficient for a new engineer to pick up the codebase without extensive archaeology.
The question that actually matters
The cheapest path isn’t the right path. Neither is the most expensive one.
The question isn’t “how do I get a chatbot?” It’s “what am I optimizing for?”
If you’re optimizing for speed: Buy SaaS. Go live this week. Validate that a chatbot actually helps before you invest in building one.
If you’re optimizing for fit: Hire a studio. Get a system designed for your exact workflow, own the code, and avoid pulling your engineering team off core product work. Scope it properly before you sign anything.
If you’re optimizing for long-term control: Build in-house, but only if the chatbot is central to your product and you already have (or are committed to hiring) the AI engineering talent.
If you’re not sure: Start with SaaS. Use it for 3 to 6 months. Learn what your users actually need from the chatbot. Then make the build-vs-buy decision with real usage data instead of assumptions.
Most companies we talk to think they need Path 3 when they actually need Path 2. They overestimate the amount of control they need and underestimate the cost of building and maintaining AI systems with a team that has other priorities. The chatbot isn’t your product. It’s a tool that supports your product. Treat the build decision accordingly.
Deciding between these paths right now? We can tell you which one fits your specific situation in a 30-minute call. No pitch deck, just an honest assessment. Talk to us.