Agentic AI for Mid-Market Companies: A First-90-Days Operating Plan
Run the first 90 days as three 30-day gates, not one project plan: name a governance owner and pick one process (Days 1-30), run a controlled pilot at a stated autonomy tier against a baseline (Days 31-60), then make a go/no-go call before any second process (Days 61-90).
Book a solution review See the governed agent platformBy Huy Nguyen (David), Founder & CEO · Updated · 10 min read
This guide is for the CEO, COO or CTO of a mid-market company that has already decided to try agentic AI and now needs to sequence the work, not another explainer of what an agent is. Our AI automation guide for business operations covers process selection, governance and ROI in general; this page is the agent-specific, 90-day execution plan that sits underneath it and routes into the agentic AI automation service once a company is ready to engage a delivery partner.
In this guide
- Why does a first 90 days need gates, not a project plan?
- The three-gate operating plan
- Building the first 90 days
- Where AI fits
- Alternatives: who runs each gate
- What delivery record exists, and what does not
- Limits of this guide
- Frequently asked questions
- Next step
Why does a first 90 days need gates, not a project plan?
A generic project plan has milestones a team tries to hit. A gate has a named owner, a pass condition written down in advance, and the authority to stop the work if the condition is not met. Gartner's June 2025 analysis found that most agentic AI initiatives are still early-stage experiments driven by hype rather than a scoped business case, that vendor "agent washing" (rebranding assistants, RPA or chatbots as agents) is common, and it predicts over 40% of agentic AI projects will be cancelled by the end of 2027 because of unclear value or weak risk controls, not model limitations. A 90-day plan without a stop rule at each gate is exactly the setup Gartner describes: a proof of concept that drifts because nobody was ever required to decide whether it earned the next 30 days.
The fix is structural, not motivational: write the pass condition for each gate before the clock starts, name one person or role desk who owns the decision, and make "stop" a legitimate, planned outcome rather than a failure to hide.
The three-gate operating plan
| Gate (days) | Goal | Who signs off | Pass condition before the next gate |
|---|---|---|---|
| 1. Foundation (1-30) | Name a governance owner, pick one process, confirm system access | AI/delivery governance role desk | Owner named, one process selected against the pillar's selection criteria, access and data-handling rules confirmed |
| 2. Controlled pilot (31-60) | Run the agent at a stated autonomy tier against a measured baseline | Delivery/QA role desk, with the governance owner | Agent clears the evaluation harness gate for its tier; baseline comparison recorded |
| 3. Expansion decision (61-90) | Decide: expand autonomy, add a second process, extend the pilot, or stop | Governance owner, with executive sign-off | A written go/no-go decision exists, with its reasoning, whichever way it goes |
A plan that skips straight to "roll it out company-wide" after a good demo has no gate 3; this is the step competitor playbooks most often compress into a single closing slide instead of a decision a named person actually makes.
Building the first 90 days
-
Name the governance owner in week one, not week four
One role desk (an AI/delivery governance lead) owns the decision at every gate. A company with nobody able to fill that role should consider fractional AI or technology leadership rather than leaving the role empty; an empty owner seat is the most common reason a pilot drifts past day 90 without a decision. Beyond this one gate, our AI agent operating model guide splits that single owner into a four-role matrix and an escalation ladder once a second agent exists.
-
Pick the first process with the pillar's selection criteria, not enthusiasm
Score candidate processes with the AI automation guide's knock-out filter and scoring sheet; prefer a process with a reversible action and a clear baseline metric over a flashier but irreversible one. Finance teams scoring order-to-cash as a candidate should read the order-to-cash AI agents guide first: it names which steps of that process fit this gate and which do not.
-
Confirm access and data handling before any pilot code runs
Write down which systems the agent may read or write, under which account, and which governance baseline and controls already apply; this is the gate where those controls become the pilot's written pass condition, not an afterthought once the pilot is already running.
-
Set the autonomy tier and the baseline before day 31
Decide upfront whether the pilot runs read-only, reversible-write or (rarely, this early) an external action, matching the tiers in the evaluation harness; record the pre-agent baseline (time, error rate or volume on the same process) so day 60 has something to compare against.
-
Run the pilot inside the harness, with a human in the loop
The agent runs at its stated tier only after it clears the harness for that tier, with the human-in-the-loop checkpoint placed where the architecture guide recommends, not wherever is convenient to build.
-
Write the go/no-go decision at day 90, both ways
Compare the pilot against the baseline and the pass condition agreed in week one; a documented "extend the pilot" or "stop" is as valid an outcome as "expand", and each stays on record for the next quarter's planning.
Where AI fits
The agent itself is the capability under test in gates 2 and 3; the plan around it (naming an owner, setting a baseline, writing a stop rule) is deliberately low-tech and does not depend on which model or vendor runs the agent. Netbase works with the major commercial and open-source AI tools and models, chosen per project, so the plan's gates do not assume one vendor's roadmap.
-
In delivered work
WhatsApp AI chatbot with CRM integration
Milestones ran from design and prototype through integration, CRM sync, NLP tuning, and testing and optimisation to a documented hand-over, over 4-8 weeks, for a client that is not named; the milestone shape is the same gated structure this plan generalises.
-
In delivered work
AI content moderation on a classifieds marketplace
A testing and optimisation phase ran before the flagging model reached the admin review dashboard, delivered for a client that is not named.
-
Available capability
A named 90-day agent operating plan as a delivery engagement
Sequencing governance, a gated pilot and an expansion decision, as part of agent delivery; not yet tied to a published case that names this exact 90-day structure.
Alternatives: who runs each gate
| Option | Strength | Risk | Choose it when |
|---|---|---|---|
| Fully self-run (internal team owns all three gates) | No external cost; full control of pace | No internal AI governance experience; gates get skipped under deadline pressure | The company already has a named internal owner and some AI delivery experience |
| Consulting partner co-runs the plan | Partner supplies the evaluation harness, governance baseline and gate discipline; internal team learns by doing | Still needs an internal executive who can make the gate 3 call | No internal AI governance owner yet, but leadership wants to stay close to the decision |
| Fractional AI/technology leadership fills the owner seat | A monthly-retainer role desk sits with the executive team specifically to run the gates, with a handover to a permanent hire planned from the start | Costs more than leaving the seat empty (which is not really an alternative, only a hidden risk) | No internal candidate exists for the owner role and the company wants the seat filled properly, not improvised |
Most Netbase projects are delivered on fixed-price contracts agreed after discovery, with milestone-based terms also offered; that fits the gated structure directly, since each gate above can be scoped and paid as a milestone, with the contract itself enforcing the stop option because a gate that fails is a gate Netbase is not paid to force through.
What delivery record exists, and what does not
- What exists. Netbase has delivered AI projects for clients that are not named where the recorded milestones include a design/prototype phase, an integration and tuning phase, and a testing/optimisation phase before go-live or hand-over, including the WhatsApp chatbot and classifieds moderation records above and anonymised RAG, document AI and MLOps pipelines; Netbase's commercial models include fixed-price and milestone-based contracts, which a gated 90-day plan fits directly. Netbase follows its published security practices: secure code review, TLS in transit and AES at rest, role-based access control, MFA for admin dashboards, vulnerability scanning and penetration testing. Netbase holds ISO 27001 certification and a SOC 2 Type II attestation for its own operations.
- What does not. No published Netbase record names a "first-90-days agentic AI operating plan" product, a pilot count, a cancellation rate, a time-to-ROI figure or a client name tied to this plan. No record describes an agent that skipped the evaluation-harness gate or ran at an irreversible tier inside its first 90 days. Netbase's certifications cover Netbase's own operations; they do not certify a client's governance programme or its gate decisions.
Limits of this guide
- This is a sequencing and governance structure, not a certification, an audit or a guarantee that a specific pilot will pass its gates; the day counts are a starting structure to adapt to how fast a given company can actually name an owner and get system access.
- A 90-day plan assumes the company has already chosen to try agentic AI; readers still deciding whether to adopt AI at all should start with the AI readiness assessment, and readers still choosing a delivery partner should start with choosing an AI consulting partner.
Plan the next step with a Netbase consultant
Frequently asked questions
No. It is a planning window with three named gates. A company can extend gate 1 if no owner is named yet, or extend gate 2 if the baseline comparison is inconclusive; what the plan rules out is skipping straight from a good demo to company-wide rollout with no written gate 3 decision.
That is the one condition this plan does not improvise around: fill the role, even temporarily, before gate 1 closes. Fractional AI or technology leadership is one way to fill it with a role desk on a monthly retainer and a planned handover, rather than leaving the decision to whoever is free.
No. The roadmap is the multi-year, multi-stream plan for the whole company; this page is the agent-specific first 90 days inside whichever stream of that roadmap is agentic AI. A company without a wider roadmap can still run this 90-day plan for one process.
Only after it clears the evaluation harness for that impact tier, including its adversarial test set. Most first pilots should stay at read-only or reversible-write for the full 90 days.
Next step
Tell us which process you are considering for the first 90 days and what systems it touches, and we will book a solution review to help you name the gate owner, set the pilot's baseline and plan the go/no-go decision before day 90. You can also see AI automation and agents or more Netbase insights.
Work behind this guide
A business that Netbase does not name, in a country and sector the record does not state, needed customers answered on WhatsApp Business with every lead captured in its CRM.
Keep ReadingRelated services and solutions
AI automation and agents that keep people in charge
Netbase provides AI automation and agent development for operations teams that want repetitive, multi-step work done by software while people keep approval over decisions that matter. We combine rule-based automation with AI where it adds value, build in human approval, and measure return against a baseline.
Learn More
Enterprise AI agent platform: run AI agents with approvals, audit logs and cost limits
An enterprise AI agent platform is a governed runtime where AI agents plan and carry out multi-step work through approved tools, while approvals, audit logs, access rules and cost limits keep every action accountable. Netbase offers it as a forward-looking solution, built inside your own environment rather than licensed as a product.
Learn More
Work with Netbase remotely
Every Netbase engagement is delivered remotely, so clients in any country and any industry can work with us; Netbase has worked with clients in 80 countries. Contracts and invoices are issued by Netbase JSC, and payment is made by bank transfer to Netbase JSC's company account or PayPal Business. Talk to the CEO, Huy Nguyen (David), on WhatsApp or by email.
Discuss a project
Netbase JSC helps organizations design, build, modernize, and operate digital products and AI-enabled business systems.project@netbasejsc.com
+84 937 869 689
91 Nguyen Chi Thanh, Dong Da, Hanoi, Vietnam
Get in touch
Tell us what you want to build, modernize, or operate.