Build vs. Buy: The Real Cost of DIY AI in HR Tech

Somewhere in your company, right now, someone is saying, "Couldn't we just build that ourselves?"

Fair. AI coding assistants can scaffold an app in an afternoon. LLM APIs cost less than your team's coffee budget. And every executive has seen a demo where a chatbot answers questions about a spreadsheet. So when HR asks for better performance reviews, smarter one-on-ones, or real insight into engagement data, "let's build it" sounds less like a moonshot and more like a weekend project.

It isn't. AI can generate a convincing prototype in days, but a prototype isn't a defensible performance system, and writing code isn't the same as maintaining secure, reliable software. The demo is the easy part. What comes after (integrations, security reviews, edge cases, oh my) is where internal AI projects go to die.

This article walks through what building really costs, what buying actually gets you, and how to run the numbers for your own team.

Why Building Feels Easier Than Ever

Ask any thrifty homeowner or overconfident dad: The build-it-yourself instinct isn't irrational. It's tempting.

But three things have made it louder now:

• The tools got good. AI coding assistants let a single engineer produce in days what used to take a sprint team months. Prototypes are cheap.
• The models are a commodity. Anyone with a credit card can call a frontier model. If the "AI" part is just an API, why pay a vendor for it?
• The pressure is real. Boards want an AI story. Leaders want to show progress. A homegrown tool feels like proof that your company is "AI-native."

Here's the catch: the market has already tested this theory, and it's moving the other way. According to Menlo Ventures' 2025 State of Generative AI in the Enterprise, 76% of enterprise AI use cases are now purchased rather than built internally — up from 53% the year before. Companies tried building. Many of them stopped.

HR is part of that shift. In Lattice's 2027 State of People Strategy Report, use of HR platforms with embedded AI jumped from 31% to 44% in a single year. And the push is coming from the top: 51% of HR leaders feel C-suite pressure to use AI for efficiency, up from 39% the year before.

The Hidden Costs

Internal AI builds rarely fail on day one. They fail on day 200, when the person who built them changes teams. MIT's NANDA initiative found that AI tools bought from specialized vendors succeed about two-thirds of the time — and internal builds succeed only one-third as often.

Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027, citing "escalating costs, unclear business value or inadequate risk controls."

In HR, those risks are sharper, because the data is the most sensitive your company holds. Here's what a homegrown tool has to handle:

• Engineering time that never ends. Building is a one-time cost. Maintaining is forever. Models get deprecated, APIs change, prompts drift, and someone has to own all of it — usually an engineer you'd rather have working on your actual product. A homegrown platform creates an ongoing dependency on engineering, IT, HR, security, legal, and leadership. You might build version one fast. You'll own every version after that, too.
• Data plumbing. Useful HR AI needs context: reviews, goals, one-on-one notes, feedback, engagement results, org structure. Stitching that together from your HRIS, survey tool, and a dozen spreadsheets is the real project. The chatbot is the garnish.
• Security and privacy. Performance ratings, compensation, and manager notes can't leak to the wrong person (or the wrong model). Permissions have to mirror your org chart exactly, and every change needs a security review. You can't set these controls once and forget them, either: they have to be validated and maintained as your product, org, and regulations evolve. A tool that works isn't automatically a tool employees can trust.
• Compliance and fairness. AI that touches performance or promotion decisions draws scrutiny from legal, works councils, and a growing list of regulators. Somebody has to document how it works, test it for bias, and defend it.
• Change management. A tool nobody uses is a very expensive side project. Adoption takes training, support, and iteration. That's work that HR teams already don't have time for. According to our 2027 report, only 30% of HR professionals describe themselves as fully engaged and energized (down from 41%), and 55% considered leaving HR this year.
• Opportunity cost. Every hour your people team spends being a product manager for an internal tool is an hour not spent on retention, development, or culture.

Put more bluntly: Time spent building HR software is time not spent building your business. And the goal was never to build forms and workflows. It's to improve manager effectiveness, develop talent, increase accountability, and make better workforce decisions. A homegrown tool solves for software, not for better performance.

What Buying Actually Gets You

The best argument for buying isn't the software. It's the context. An AI model is only as useful as the data it can see, and a purpose-built HR platform already holds the data that matters. It's already structured, permissioned, and connected.

That's the idea behind the Lattice AI Agent, which is grounded in your company's people data across performance, one-on-ones, goals, feedback, engagement, and growth. Out of the box, it handles the work a homegrown tool would spend a year trying to reach:

• Review drafting that pulls from one-on-ones, feedback, goals, and updates, so managers start from evidence instead of a blank page. That matters when 70% of US managers say time is their biggest barrier to one-on-ones and feedback.
• Coaching for managers preparing for tough conversations, performance discussions, and growth plans.
• Meeting notes that turn one-on-ones into summaries, action items, and agendas, synced back to the employee's record.
• Analysis that surfaces patterns, trends, and risks in people data — with charts and tables, not just paragraphs.
• Integrations with the systems you already run, including a bidirectional Workday integration that sends performance insights into Workday HCM.

You also get the things nobody demos: a security team, a compliance roadmap, permissions that already match your org, and a product team shipping improvements every month. When the underlying models improve, you get the upgrade. When they break, it's not your engineer's weekend.

HR leaders seem to agree: 69% say they're satisfied with the third-party AI tools they use. And the results show up fast. Waterford.org runs reviews 2x faster with Lattice. GoCardless lifted review participation from 62% to 100%. Weave saves 30 hours per engagement survey. More than 5,000 organizations, from Robinhood to Duolingo to NPR, run on Lattice. That's the payoff of skipping the year-one build and starting at year three.

Lattice gives us a central hub where performance, development, engagement, and planning come together. It lets us operate strategically and speak the same language as our Finance partners.

Karishma Barua

VP of People

@

Fairmarkit

Running the Numbers

Start with a back-of-the-napkin version. Two mid-level software engineers at $182K each cost $364K a year, and that's before benefits, management, QA, security, design, integrations, maintenance, or opportunity cost. Rolling out Lattice Foundations (Performance and Goals) to a 500-person company runs about $78K a year. A two-person build team can cost more than four times as much as giving your whole company Lattice.

The returns come faster, too. A Forrester Total Economic Impact study found Lattice delivered 195% ROI over three years, with a three-month payback period and $3.5M in three-year benefits, from cutting attrition by 40% to reclaiming thousands of work hours. You'd be seeing returns before an in-house system would even ship.

Six Questions to Ask Before You Build

If your team is still leaning toward building, pressure-test the plan with these:

1. Who owns it in two years? Name the person. If the answer is "the engineer who built it," ask what happens when they leave.
2. Where does the data come from? List every system the tool needs to read from, and who maintains each connection.
3. Who signs off on security and compliance? Get legal, IT, and security in the room before the prototype, not after.
4. Is this our competitive advantage? Build what makes your company different. Performance reviews and one-on-one notes probably aren't it.
5. What's the total cost over three years? Include salaries, model costs, maintenance, and the projects that didn't get done.
6. If we buy, have we vetted the vendor? Buying isn't automatically safe. 47% of HR leaders say at least one HR tech purchase in the last two years underperformed or was a clear mistake, most often because of missing functionality (50%) or integration problems (34%). Look for a platform that already holds your people data and connects to the systems you run.

If you can answer the first five with confidence, building might be right for you. If you can't, consider shopping around instead.

Build what's yours. Buy what's proven.

The build vs. buy question was never really about whether you can build. With today's tools, almost anyone can. It's about whether you should spend your engineers, your budget, and your HR team's patience on a problem someone has already solved.

For most companies, the smart move is to put internal resources toward what makes them unique, and buy the HR AI that's already grounded in people data, secured, and improving every month.

See the Lattice AI Agent in action or schedule a personalized demo to learn more.