What Does It Cost to Integrate AI Into Your Existing Software?

It depends on the route. Switching on AI your software already has ranges from free to a per-user add-on (Microsoft 365 Copilot Business is $21 a user a month, billed annually). Connecting AI through Zapier or Make usually costs a few hundred dollars to build plus $10 to $75 a month. Custom API work often runs past $1,000.

The build price is rarely the part that surprises people. The surprises are the plan upgrade needed to reach an API or an AI feature, the data clean-up before anything works, and the upkeep once it's live. Price all three before comparing quotes, and compare routes over a year of use, not on the day of the invoice.

Follow me on Instagram@sagnikteaches

Four ways to add AI to the software you run, with typical costs

"Integrating AI" covers four quite different jobs. The route decides most of the cost, so work out which one a quote is actually for.

Connect on LinkedInSagnik Bhattacharya

Built-in prices are the easiest to pin down. Microsoft 365 Copilot Business costs $21 per user per month on annual billing ($25.20 if billed monthly), with a promotional $18 through 31 December 2026 on annual billing; Microsoft now calls the app Microsoft Copilot, though product pages still use both names. HubSpot's Breeze AI runs on credits at $10 per 1,000, and its Customer Agent uses 50 credits, about $0.50, per resolved conversation. If the software you rely on most is less well known, its pricing page and release notes are the place to look. Whether AI can work with the tools you already use covers how to tell which route your software supports.

Subscribe on YouTube@codingliquids

Check what your current plans already include

Before paying for any connection, spend half an hour on the AI your subscriptions already carry. Examples, as the vendors describe them in September 2026:

• Microsoft 365 business plans include Copilot Chat at no extra cost, and it now works on Outlook mail and open files. The paid Copilot licence adds reasoning across emails, meetings and files together.
• Google Workspace plans include Gemini: Business Starter gets Gemini in Gmail and the Gemini app; Business Standard and above get it across Gmail, Docs, Sheets, Drive and more.
• Xero offers JAX auto-reconciliation on the Growing plan ($55) and above.
• HubSpot Free and Starter get Breeze Assistant for writing and summaries; its agents, including Customer Agent, need Professional or Enterprise.

A campsite on Google Workspace Business Standard, for instance, might find that Gemini in Gmail already drafts decent replies to pitch enquiries, which settles the "should we integrate AI with our email" question at no extra cost. What it can't do is check the booking system's availability, so the paid connection, if any, should be scoped to that gap alone. The AI features already in your software lists what to switch on first.

Marketplace add-ons: quick to install, worth vetting first

The second route sits between built-in features and a custom connection. Most booking systems, CRMs and accounting packages have an app store where third-party developers sell AI add-ons: review-reply assistants, document readers, chat widgets. They install in minutes and charge a monthly fee, so the cost is easy to see. The risk is less visible.

Consider a tour operator that installs an AI review-reply app from its booking platform's app store. Set-up takes about 40 minutes and, in this illustration, the app costs $29 a month. Before paying, the owner checks four things. What permissions the app asks for at install: read-only access to reviews is fine, write access to the whole customer list is not. Whether replies post automatically or wait in a queue for approval. Where the app stores the review text and for how long. And what happens if the developer closes. That last question isn't theoretical: the AI calendar tool Clockwise shut down on 27 March 2026 and deleted user data rather than transferring it. For a small add-on, assume you may need to export your settings and prompts one day, and keep a copy.

At $29 a month the sum is simple. If the app saves the owner 3 minutes on each of 60 reviews a month, that's 3 hours; at $25 an hour of the owner's time, $75 of value for $29 of cost. If it saves 1 minute each, the add-on loses money, which is why a two-week trial with timed replies beats the app store's star rating.

A letting agency prices the same job three ways

Consider an illustrative letting agency managing 220 tenancies with six staff, Microsoft 365 Business Standard and a cloud property management system. Tenants send about 90 maintenance requests a month by email and web form. Each takes a staff member about 8 minutes: read it, judge urgency, reply to the tenant, create a job in the property system and notify a contractor. That's 12 hours a month.

The goal: AI classifies urgency, drafts the tenant reply for approval and creates a draft job with the right category, cutting handling to about 3 minutes. That saves 5 minutes a request, 7.5 hours a month, or 90 hours a year. At an illustrative $28 an hour, the job is worth about $2,520 a year. For the triage logic itself, see how property managers use AI to triage maintenance requests.

The agency collected three quotes (all figures illustrative):

• Quote A, the property system's own AI add-on: $1 per property per month, so $220 a month or $2,640 a year. No build cost. Triage and reply drafts appear inside the system, and the add-on also writes listing descriptions, which the agency doesn't need.
• Quote B, a freelancer's no-code build on Make: $650 fixed, running on Make's entry plan (from about $9 a month) with Claude Haiku 4.5 for the AI step. It needed the property system's API, which is only on the vendor's higher tier, an extra $60 a month.
• Quote C, a developer's custom integration: 45 hours at $50 an hour, $2,250, plus a small cloud function the developer estimated at about $15 a month, and the same $60-a-month API tier.

AI usage is the same small number for B and C. Each request is roughly 3,000 input tokens (the request plus the triage rules) and 500 output tokens. At $1 and $5 per million tokens, 90 requests cost about $0.50 a month, or $6 a year.

Then the freelancer on quote B spotted something: the property system can create jobs from an email sent in a fixed format to a special address, and that works on the agency's current tier. Using it instead of the API removed the $60-a-month upgrade. That one question was worth $720 a year.

Cost per request: making different quotes comparable

Quotes arrive in different shapes: a monthly fee, a fixed build, an hourly estimate. Turning each into a first-year and second-year total, then dividing by the requests handled, makes them comparable. The agency's figures, with 12 hours a year of internal upkeep at $28 and one outside fix a year:

Each request handled saves 5 minutes, worth about $2.33 at $28 an hour. So B pays for itself in year one, C only from year two, and A costs slightly more than the time it saves unless the agency starts using the listing-description feature. At 90 requests a month, custom code is the wrong tool. Run the same sum at 900 requests a month and the order flips. C's build and hosting barely change, AI usage rises to about $60 a year, and its cost per request falls to about 35 cents in year one and 14 cents after. A's price follows the number of properties, so it stays near $2.50 a request. B would need a bigger Make plan, because 900 requests at eight steps each plus the mailbox checks come to about 10,000 credits a month, twice the entry plan. That's the point at which developer quotes become worth getting.

The agency chose B. Its staff already had Copilot Chat in Outlook for polishing replies at no extra cost, so the paid work was confined to the one thing no included tool could do: turning a tenant's email into a categorised job in the property system.

Plan upgrades, data clean-up and other costs behind "it integrates"

"It integrates with your CRM" on a sales page tells you nothing about cost. The lines that most often appear later:

1. The tier the feature sits on. API access, webhooks or AI agents are often reserved for higher plans. A boutique hotel on HubSpot Starter that buys credits expecting to use Customer Agent will find the agent needs a seat on any Professional or Enterprise hub (usually Service Hub), and Starter's included credits (500 a month, no rollover) only cover Breeze Assistant.
2. Seats nobody uses. Copilot Business is sold on annual commitment only. A twelve-person office that buys twelve seats at $21 and then sees, in the Microsoft 365 admin centre's Copilot usage report, that four people use it regularly is paying $168 a month for eight idle seats until renewal. Start with the people who will use it.
3. Data clean-up. AI triage that relies on property addresses and contractor categories fails when half the records are inconsistent. Fiverr's cost guide notes many businesses spend $40 to $300 on data preparation before automations work, and that's before your own staff time.
4. Testing and documentation. The same guide says comprehensive testing, documentation and training materials typically add 20% to 35% to base project costs. A quote without them isn't cheaper; it has moved the cost to you.
5. Running costs. Platform plans, AI usage and fixes continue after the invoice. What you pay after an automation goes live itemises them.
6. Controls you can't switch off. Intuit says QuickBooks Online's AI features can't currently be switched off individually. If a feature isn't acceptable for your data, find out before you build around the software.

Seven questions that turn an integration quote into a number

Send the same questions to every vendor or builder, in writing, so the answers line up:

1. Which connector, API or import route will you use, and does our current plan include it?
2. Which data fields do you need, and who cleans them if they're messy?
3. What's included: testing on our real examples, documentation, a handover session, and how many rounds of changes?
4. What will it cost each month at our volume (platform, AI usage, hosting), and at double our volume?
5. Whose accounts will it run in, and who owns the workflows or code?
6. When our software updates its API or fields, who fixes the integration, how fast, and at what rate?
7. How is AI output checked before a customer or tenant sees it?

Answers show their quality quickly. On question 4, "running costs are minimal" is a non-answer. "Make's entry plan covers about 3,600 credits a month at your volume, most of them the mailbox check every 15 minutes; AI usage is about 50 cents a month" is a builder who has done the sum. The difference between the two is usually the difference between a quote you can budget and one you can't. If you're not sure what an API is or why question 1 matters, what an API is and why it matters when you buy software explains it in plain terms.

Getting a rough estimate from AI before anyone quotes

A chat assistant can turn your description into a checklist of integration points and a rough effort range, which helps you judge quotes. Keep customer data out of it.

I run a letting agency with 220 tenancies. We use a cloud property
management system, Microsoft 365 and a website form for maintenance
requests. I want AI to classify each request's urgency, draft a
reply to the tenant for staff approval, and create a draft job in
the property system.
List:
1. every integration point needed (what connects to what)
2. the data fields each step needs
3. three ways to build it: built-in feature, no-code, custom code
4. a range of build hours for each, with your assumptions
Mark anything you are not sure our software supports as UNVERIFIED.

An illustrative answer was useful as a map and wrong as a price. It listed the right integration points and fields, but it recommended "the property system's native AI triage module" without marking it UNVERIFIED (no such module existed), assumed a Zapier app for the property system that hadn't been checked, and estimated the no-code build at 4 to 6 hours with no time for error handling, testing or documentation. Fixes: check each integration point in the vendor's app directory or API documentation, add 20% to 35% for testing and documentation, and use the output as the checklist you send with your quote request rather than as a budget.

When a bigger integration budget is justified

Custom work and larger add-ons earn their cost in a few situations. Volume is the first: once the per-request sum shows the build spread across thousands of items a year, a higher up-front cost buys a lower running cost. Breadth is the second: one integration that serves three jobs (triage, tenant updates and contractor chasing) spreads its cost further than three separate connections. Control is the third: where data must stay in systems you manage, or a regulator or insurer expects specific logging, a custom route may be the only acceptable one.

Outside those cases, start with what's included, then the cheapest connection that does the one job the included tools can't. If the question is really whether to build your own AI tool at all, building versus buying AI for a small business takes that decision on directly.

AI integration costs: follow-up questions

Is it cheaper to switch software than to integrate AI into what we have?

Rarely, once you count migration. Moving data, retraining staff and running two systems in parallel usually costs more than a connection or an add-on. Switching makes sense when your current software has no API, no connector support and no AI plans, or when you were planning to leave it anyway. Price the switch as a project of its own before deciding.

Do I need a developer to integrate AI with our software?

Not for most small-business jobs. If your software has a Zapier or Make connector, or can send and receive email in a set format, a no-code build is usually enough. You need a developer when the software only offers a raw API, when volumes are high enough that per-task pricing gets expensive, or when the logic is too complex to maintain in a visual builder.

Who pays when our software vendor changes its API?

You do, unless your contract says otherwise. Ask any builder to state what a fix after a vendor change costs, and whether small fixes are included for a period after handover. Keep documentation and a test set so the fix is quick, whoever does it.

Can AI features in our software be switched off if we don't want them?

Usually, but not always individually. Intuit, for example, says QuickBooks Online's AI features can't currently be switched off one by one. Check the admin settings and the vendor's help pages before you rely on being able to disable a feature for data or quality reasons.

Further reads

• What Connecting Two Business Apps Really Costs: Four Options — The four ways to connect two apps, priced in more detail.
• Native Integrations vs Zapier: Which Should Connect Your Apps? — When a built-in integration beats a connector platform.
• What Does the ChatGPT API Cost for a Business Automation? — Work out the AI usage line for your own volumes.
• Hidden Costs of AI Implementation Most Small Businesses Miss — The costs that appear before anything goes live.
• Your Software Raised Its Price for AI: Keep, Downgrade or Switch? — What to do when a vendor raises prices for its AI.
• Is Microsoft 365 Copilot Worth It for a Small Business? — Whether the per-seat Copilot licence earns its cost.
• AI for Letting Agents: Handle Tenant Queries Without Extra Staff — The tenant-query side of AI for letting agents.
• No-Code AI Tools vs a Custom Build: When Templates Stop Working — Six signs a Zapier or Make workflow has outgrown no-code, the middle options to try first, and a worked cost comparison for a care agency's shift offers.
• Is It Worth Switching CRM Just for Better AI Features? — Where CRM AI features sit on each price list, the migration costs switching quotes leave out, and a three-year worked sum for a small sales team.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Microsoft 365 Copilot Business pricing and Microsoft Learn on Copilot Chat; HubSpot pricing and knowledge base on credits and Customer Agent; Zapier pricing page; Make pricing; Anthropic API pricing; Intuit on QuickBooks Online AI settings; Upwork hourly rates guide (2026); Fiverr cost guide for AI automation experts (April 2026).