A custom AI agent built around one clear outcome, such as texting back a lead or answering the phone, realistically launches in a few weeks to about three months. The range that stretches to six months or longer almost always belongs to a different kind of project: a multi-system enterprise rollout, not a single practice connecting a phone line and a calendar.
This isn't a guess. Forrester's own 2026 analysis of conversational AI deployments reports that in 2024, customer time to value ran six to nine months, and that in 2026 "multiple customers reported going live in under three weeks," with six-month deployments still happening, but only when "internal security approvals and complex integrations" were the limiting factor, not the AI itself.
The harder truth is a different number entirely: most AI pilots never launch at all. This post is the honest timeline, what actually stretches it, and how a small practice avoids becoming one of the projects that quietly dies before it ever answers a single call.
So what's the real number: weeks, or months?
Both are real, and the difference isn't mysterious: it's scope. A single agent built to do one job, say, answer the phone after hours and text a lead back in seconds, is a fundamentally smaller build than a multi-agent system wired into a dozen enterprise tools. Forrester analyst William McKeon-White, writing in July 2026, put the shift plainly: "In 2024, customer reported time to value ranged from six to nine months. Now, in 2026, with agentic systems, multiple customers reported going live in under three weeks."
That is not a small change. It is roughly a ten-times improvement in the fast case, in two years, on the same category of software. The slow case did not disappear, it just moved to a smaller, more predictable share of projects, and Forrester names exactly what causes it.
Why do some AI agent projects take six to nine months?
Forrester's own reporting names the two real culprits, and neither one is the AI: "Six-month deployments were reported, but the aforementioned internal security approvals and complex integrations were cited as the limiting factors." A security review at a hospital system or a bank moves at the pace of a security review, on purpose, and an agent that has to read and write to a dozen legacy systems has a dozen places to get stuck.
Neither of those describes most single-location practices. A dental office or a med-spa typically runs one phone system, one scheduling tool, maybe one practice-management platform, and one person who can actually say yes to the build. Remove the multi-system integration and the multi-layer approval chain, and most of what stretches an enterprise timeline into months simply isn't there to stretch.
Is a single-location practice actually faster to launch than an enterprise?
Usually, yes, for the same reason a small boat turns faster than a container ship. Fewer systems to connect, fewer people who have to sign off, one decision-maker instead of a procurement committee. Forrester's own framing supports this directly: with the right vendor doing the heavy lifting, fast go-lives are "increasingly accessible for teams of all sizes," not just large IT departments with dedicated integration engineers.
| Single scoped agent | Multi-system enterprise rollout | |
|---|---|---|
| Systems it connects to | Usually 1-2 (phone, calendar) | Often a dozen or more legacy systems |
| Who approves the build | Usually one owner-operator | A security + procurement chain |
| Security review depth | Standard, not formal audit | Often a multi-week formal review |
| Reported 2026 range | Under 3 weeks to ~3 months | Up to 6 months or more |
Why do most AI pilots never actually go live?
This is the number that actually matters more than the week count, because it's the fear underneath the question. MIT's Project NANDA, in its "GenAI Divide: State of AI in Business 2025" report, found that roughly 95% of generative AI pilots fail to reach measurable business impact, with only about 5% of organizations seeing real, extractable value, despite an estimated $30 to 40 billion in enterprise GenAI spending behind the other 95%.
The reason isn't that the underlying AI is too weak. MIT NANDA's own conclusion is that most GenAI tools "do not retain feedback, adapt to context, or improve over time," meaning the systems that stall are usually missing a feedback loop and real workflow integration, a design and ownership problem, not a capability problem. That distinction matters, because it means the fix isn't waiting for a smarter model. It's building the agent correctly the first time.
Does buying a done-for-you build actually launch faster than building it yourself?
The data on this is one of the more useful, and more surprising, numbers in the MIT NANDA report: buying from a specialized vendor succeeded roughly 67% of the time, against roughly 33% for internal DIY builds, about half as often. An internal build usually means someone's already-full job description just grew a second job: maintaining an AI system with no dedicated owner and no outside accountability for whether it ever actually goes live.
Gartner's own prediction lines up with the same pattern from a different angle: over 40% of agentic AI projects will be canceled by the end of 2027, driven by "escalating costs, unclear business value and inadequate risk controls." That is close to a textbook description of an ambitious internal build that never had a single person accountable for shipping it.
What does a realistic week-by-week timeline look like?
There's no single verified public timeline for a one-location practice specifically, so treat the breakdown below as an honest, illustrative pattern built from the ranges above, not a measured study. It's the shape a scoped build for a single outcome typically follows when the scope stays narrow.
- Week 1-2, scope: pick the one outcome (answer the phone, text back a lead, fill a cancellation), connect the phone line, calendar, or inbox it needs to touch.
- Week 2-4, build and test: the agent is built and run against real call and message patterns from the practice, not generic test cases.
- Week 4-6, supervised soft launch: the agent goes live with a human reviewing its actions before or shortly after they happen, catching anything that needs correcting.
- Week 6-10, light-review live: review tapers as trust builds, with an escalation path still in place for anything ambiguous or high-stakes.
That pattern comfortably lands inside Forrester's "under three weeks to a few months" range for a scoped build. It stretches toward the six-month enterprise end only when a practice adds more locations, more connected systems, or a slower internal approval chain, exactly the two levers Forrester names, not company size on its own.
What can a practice owner actually do to avoid becoming part of the stall?
Four things, each one mapped directly to a documented failure mode above, not a guess.
- Scope one outcome, not a general assistant. "Handle whatever comes up" is how a project drifts for months without ever going live. "Text back every missed call in under a minute" ships.
- Buy from a specialized builder instead of an internal DIY build. MIT NANDA's 67% vs. 33% gap is the single largest lever in this whole post, and it costs nothing to take advantage of.
- Ask to see a working version early, not only a slide deck. A vendor who can show the agent handling a real call pattern in week two is not the vendor whose project quietly dies in month five.
- Build the escalation path in from day one. An agent that hands off anything ambiguous to a person, by design, is the difference between a supervised soft launch and a liability near real patients or clients.
Where does Neuron HQ fit?
Neuron HQ builds one clearly scoped AI agent at a time for local service businesses, done for you, human-in-the-loop by design, and priced to the specific business rather than off a fixed rate card. We're upfront that Neuron doesn't yet have published case studies to point to. What we offer honestly is a plainly scoped build and a straight answer, before you commit to anything, about how long yours would realistically take. For the buying guide and the cost comparison against a human hire, see What Is an AI Employee?. For what makes an agent genuinely custom versus off-the-shelf, see Custom AI Agents: When Off-the-Shelf Isn't Enough.
Tell us the one job. We'll give you a straight timeline.
Describe the single outcome you'd hand off first: the phone, a missed-lead text-back, a cancellation waitlist. We'll tell you honestly how long a scoped build like that realistically takes for a practice your size.
See the full approach on the AI Agents page, browse more guides on the Neuron blog, or start from the Neuron HQ homepage. A real reply, usually within one business day.
Frequently asked questions
How long does it actually take to launch a custom AI agent?
For a single, clearly scoped outcome, such as answering calls or texting back a lead, the honest range is a few weeks to about three months. Forrester's own 2026 analysis reports customers going live in under three weeks on the fast end, with six-month deployments on the slow end, the difference driven almost entirely by security review and how many existing systems the agent has to connect to, not by company size.
Why do some AI agent projects take six to nine months?
Forrester found that in 2024, customer-reported time to value ranged from six to nine months, and even in 2026 six-month deployments still happen when internal security approvals and complex back-end integrations are the limiting factors, not the AI itself. A large enterprise wiring an agent into a dozen legacy systems with a formal security review is a fundamentally different project than one clinic connecting a phone line and a calendar.
Is a single-location practice actually faster to launch than an enterprise?
Usually, yes, for the same reason a small boat turns faster than a container ship: fewer systems, fewer approvals, one decision-maker. Forrester notes that with vendor assistance, fast go-lives are increasingly accessible for teams of all sizes, not just large IT departments. A one-location practice typically has one phone system, one scheduling tool, and one owner who can approve the build, which removes most of what stretches an enterprise timeline into months.
Why do most AI pilots never actually go live?
MIT's Project NANDA found that roughly 95% of generative AI pilots fail to reach measurable business impact, with only about 5% seeing real results, despite an estimated $30 to 40 billion in enterprise GenAI spending. The report traces most failures to tools that cannot retain feedback or adapt to context, not to weak underlying AI models, meaning the stall is usually a design and integration problem, not a capability problem.
Does buying a done-for-you build launch faster than building it yourself?
The data says yes by a wide margin. MIT NANDA found that buying from a specialized vendor succeeded about 67% of the time, versus roughly 33% for internal builds, about half as often. Separately, Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027, citing escalating cost, unclear value, and inadequate risk controls, the exact failure pattern of an ambitious internal build with no dedicated owner.
What does a realistic week-by-week timeline look like for a small practice?
An illustrative, honest pattern for one clearly scoped agent: week 1 to 2 is scoping the single outcome and connecting the phone, calendar, or inbox; week 2 to 4 is building and testing against real call and message patterns; week 4 to 6 is a supervised soft launch where a human reviews the agent's actions; and by week 6 to 10 it is running with light human review as trust builds. This stretches toward Forrester's six-month range only when a practice adds multiple locations, multiple systems, or a slow internal approval chain.
What can a practice owner do to avoid becoming part of the stall?
Scope one outcome instead of a general assistant, buy from a specialized builder instead of attempting an internal build, pick a partner who shows a working version early instead of only a slide deck, and insist on a human-in-the-loop escalation path from day one instead of bolting one on after something goes wrong. Each of these maps directly to a documented failure mode in the Forrester, MIT NANDA, and Gartner findings above.
Where does Neuron HQ fit?
Neuron HQ builds one clearly scoped AI agent at a time for local service businesses, done for you, human-in-the-loop by design, and priced to the specific business rather than off a fixed rate card. Neuron has no published case studies yet; what we offer honestly is a plainly scoped build and a straight answer about how long yours would realistically take before you commit to anything.
Sources & methodology
Every figure on this page traces to a primary source we opened and read directly. We drop any stat we can't independently corroborate rather than round it or guess. Note: the week-by-week breakdown above is our own illustrative pattern, labeled as such, not a cited measured study.
- Forrester: William McKeon-White, "Five Lessons From The Forrester Wave: Conversational AI Platforms For Employee Services, Q3 2026" (Forrester blog, published July 23, 2026). Source of the 2024 six-to-nine-month baseline, the 2026 under-three-weeks figure, and the security-approvals/integration-complexity explanation for six-month outliers.
- MIT Project NANDA: "The GenAI Divide: State of AI in Business 2025" (MIT Media Lab, published August 2025). Source of the roughly 95%-pilot-failure / 5%-success figures and the 67% vs. 33% buy-vs-build success rates.
- Gartner: "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027" (press release, June 25, 2025). Source of the 40%+ cancellation figure and its stated causes.
Last reviewed: August 30, 2026. Found a figure that's drifted? Email support@neuron-hq.com and we'll review it.
Agent vs. employee vs. chatbot vs. RPA, the honest, sourced difference.
What makes an agent truly custom, and how it's built.
What a scoped build actually runs, and what drives the range.
The truth about "it learns", no fine-tuning on your data, just captured judgment and tuned playbooks.