Most operations and strategy leaders inside mid-market companies know they need an AI strategy. What they do not have is a concrete planning process — one that produces a prioritized, funded, and governable roadmap rather than a slide deck that collects dust after the quarterly offsite. The gap between “we should be doing something with AI” and “we have three funded use cases in flight” is almost always a planning and prioritization problem, not a technology problem. This post lays out a 90-day framework for closing that gap, phase by phase, with the specific outputs each phase must produce.

Why Most AI Roadmaps Fail Before They Start

The most common failure mode we see is what might be called the pilot graveyard: organizations that have run between three and eight AI pilots over the past two years, none of which scaled, none of which were formally terminated, and none of which produced a reusable lesson. Resources were consumed, vendors were cycled through, and the organization is now more skeptical of AI than it was before the pilots started.

The root causes are consistent across organizations of different sizes and sectors. First, use cases were selected based on what a vendor was selling or what a senior leader read about, not based on a structured assessment of where the organization’s data, processes, and change capacity actually support AI deployment. Second, there was no governance framework to decide who was accountable when the AI output was wrong — so when it inevitably was, the project stalled while people argued about ownership. Third, the business case was built backward: the team decided on the technology first and then constructed ROI math to justify it, which meant the baseline numbers were soft and the finance team never fully believed the model.

A 90-day planning process, properly structured, avoids all three failure modes. It produces a current-state inventory before selecting use cases, sequences use cases based on a rigorous impact-versus-feasibility assessment, and builds the governance framework in parallel with the business case so that both are ready before a single dollar is committed to implementation.

Days 1–30: The Current-State AI Inventory

The first 30 days are not about generating ideas. They are about understanding what is already in place. In our experience, mid-market organizations conducting a formal AI inventory for the first time discover that they are already running significantly more AI-adjacent capability than leadership realized — typically embedded inside CRM platforms, ERP modules, marketing automation tools, HR systems, and customer service software. These capabilities are often untouched, misconfigured, or producing outputs that no one is reviewing.

The inventory has four components:

1. Tool and capability audit. Document every software platform in use across operations, finance, sales, marketing, HR, and IT. For each platform, identify whether it includes AI or machine learning functionality, whether that functionality is currently enabled, who owns it, and whether anyone is actively reviewing the outputs. This is typically a two-week exercise involving IT, department heads, and procurement records.
2. Data readiness assessment. AI use cases are only as good as the data that feeds them. For each major operational domain, assess data completeness, data quality, data accessibility (is it locked in a system with no API?), and data governance maturity. Organizations with fragmented ERP implementations, inconsistent master data management, or significant manual data entry processes have a lower effective data readiness score and should sequence use cases accordingly.
3. Process documentation scan. Identify the 20 to 30 highest-volume, highest-variability processes in the organization. These are the candidates for AI augmentation. High-volume processes generate the data density that AI models need; high-variability processes are where intelligent automation produces the most value relative to traditional rules-based RPA.
4. Change capacity assessment. This is the component most organizations skip, and skipping it is expensive. Assess your organization’s current change load — how many other transformation initiatives are running in parallel — and your historical change absorption rate. AI deployments that land on top of an ERP migration, a merger integration, or a major regulatory compliance project almost always underperform because the people who need to adopt the new workflow are already at capacity.

The output of Days 1–30 is a single consolidated inventory document: tools, data readiness ratings by domain, candidate process list, and a change capacity score. This document becomes the factual foundation for use-case prioritization in Phase 2.

Days 31–60: Building the Use-Case Pipeline

With the inventory complete, the second 30 days focus on generating, filtering, and prioritizing a use-case pipeline. The mechanism for this is a structured workshop series — typically three half-day sessions with cross-functional representation from operations, finance, IT, and the business lines most likely to be affected.

The workshops produce a long list of potential AI applications, which is then scored on two dimensions to build the prioritization matrix.

The Impact-versus-Feasibility Matrix

Impact is assessed along three sub-dimensions: revenue effect (does this use case protect or grow revenue?), cost effect (does it reduce fully loaded operational cost, not just hours?), and risk effect (does it reduce error rates, compliance exposure, or customer churn?). Each use case receives a composite impact score from 1 to 5 based on estimated magnitude and confidence level. Estimates should be conservative and should reference the organization’s own baseline data from the inventory, not industry benchmarks from vendor white papers.

Feasibility is assessed along four sub-dimensions: data readiness (from the inventory), integration complexity (how many systems does this touch?), regulatory exposure (are there compliance, privacy, or liability considerations?), and organizational readiness (is the affected team likely to adopt the change, and do they have the capacity to do so right now?). Each use case receives a feasibility score from 1 to 5.

The mistake most organizations make in this step is letting executive preference override the matrix. A CFO who is personally excited about AI-powered financial forecasting will push it into Wave 1 even if the organization’s financial data is fragmented across three systems and the FP&A team is in the middle of a re-org. The matrix is the instrument that creates a defensible, objective basis for sequencing — and it should be shared with leadership before the sequencing conversation begins, not after.

By the end of Day 60, the output is a prioritized use-case pipeline: a ranked list of AI applications segmented into Wave 1 (the three to five use cases that will be funded and implemented first), Wave 2 (the next cohort, targeted for 6–18 months out), and the parking lot (use cases that require foundational data or process improvements before they are viable). For each Wave 1 use case, the team should have a preliminary estimate of implementation cost, timeline, and expected benefit — rough at this stage, but grounded in the inventory data.

Days 61–90: Governance Design and Business Case Construction

The final 30 days have two parallel workstreams: governance framework design and business case development for the Wave 1 use cases. These must run in parallel because the governance framework directly affects the risk section of the business case, and the business case scope directly affects which governance controls are proportionate.

Governance Framework Design

AI governance for mid-market companies does not need to be elaborate. It needs to be functional. The minimum viable governance framework covers five areas:

• Decision authority. For each AI use case, define which decisions the AI is authorized to make autonomously, which decisions require human review before action, and which decisions require human approval. This is not a technology question — it is a risk tolerance question that should be answered by operations and compliance leadership, not IT.
• Accountability assignment. Every AI system in production needs a named human owner who is accountable for its outputs. This person is responsible for monitoring performance, escalating anomalies, and initiating retirement when the model is no longer fit for purpose. In practice, this is typically a senior manager in the business function the AI supports.
• Data governance linkage. The governance framework should explicitly reference which data sources feed each AI system, who owns those data sources, and what the protocol is if source data quality degrades. This is the single most commonly missing element in mid-market AI governance.
• Incident response protocol. Define what constitutes an AI incident (a materially wrong output that affected a business decision or a customer), how incidents are logged, who is notified, and how the system is suspended pending investigation. Organizations that do not define this in advance spend weeks in post-incident confusion about process.
• Model review cadence. AI models drift. Data distributions shift. Define a minimum review cadence for each production AI system — quarterly for high-stakes applications, semi-annually for lower-stakes ones — and assign responsibility for conducting the review.

The Business Case Structure That Gets Funded

The business cases that finance teams approve share a consistent structure. They start with a precise, auditable cost baseline — not an estimate of what work costs, but a documented calculation using actual headcount, loaded labor rates, error rates, and processing volumes from the organization’s own data. Business cases that use industry benchmarks for the baseline are immediately suspect.

The benefit calculation should be conservative and decomposed. Rather than stating a single ROI number, break the benefit into components: labor hours redirected (not eliminated, unless headcount reduction is an explicit objective), error-driven rework reduced, customer-facing cycle time improved, and compliance incidents avoided. For each component, show the calculation and the assumption. Finance teams that can audit each assumption will approve the case faster than teams that are handed a single aggregate number they cannot reconstruct.

The cost section is where most AI business cases fail. Common errors include omitting the cost of integration work (which is almost always more expensive than estimated), omitting ongoing model maintenance and monitoring costs, and omitting the cost of training and change management. In our experience with mid-market deployments, change management and training account for between 20 and 35 percent of total implementation cost for AI initiatives — a figure that is almost never included in the initial business case and that frequently surfaces as a budget overrun.

The final output of Days 61–90 is a governance framework document and three fully developed Wave 1 business cases, each with a cost baseline, decomposed benefit estimate, implementation cost breakdown, risk register, and recommended vendor or build approach. These three documents, presented together, constitute a decision package that an executive team can act on.

What Comes After Day 90

The 90-day planning process is not the end — it is the foundation. Organizations that complete this process with discipline have something most of their peers do not: a shared, evidence-based understanding of where AI will actually create value in their specific operational context, sequenced in an order they can actually execute, with governance controls that were designed before the first deployment went live rather than retrofitted after something went wrong.

The implementation of Wave 1 use cases typically begins in Month 4. Organizations should plan for a formal retrospective at the end of Wave 1 (typically 9–12 months after the planning process concludes) to assess what the use cases actually delivered versus what the business cases projected, update the use-case pipeline scoring, and refresh the governance framework based on what the first production deployments revealed. This retrospective is the mechanism that converts AI from a series of one-off projects into an ongoing organizational capability.

Frequently Asked Questions

How much internal time does the 90-day planning process actually require?

For a mid-market organization of 200 to 1,000 employees, the planning process typically requires a dedicated project lead at 50 to 75 percent time allocation, two to three subject matter experts from IT and operations at 20 to 30 percent time, and three to four half-day workshops with senior leadership. If the organization is engaging an external partner to facilitate the process, the internal time commitment is lower, but the subject matter expert availability is non-negotiable — no external facilitator can complete a credible current-state inventory without deep access to internal systems, process owners, and data.

Should we start with a generative AI use case or a more traditional machine learning application?

This is a sequencing question, and the answer depends on your data readiness and organizational context rather than on what technology is currently attracting attention. Generative AI applications — document summarization, internal knowledge retrieval, customer-facing conversational interfaces — often have lower data infrastructure requirements and faster time to value than traditional ML applications, which typically require substantial labeled training data and model development effort. For organizations with low data maturity scores, starting with a well-scoped generative AI use case is often the faster path to a funded Wave 1 win. For organizations with strong structured data assets and clear predictive modeling needs, a traditional ML approach may deliver more durable value.

What is the most common mistake organizations make when presenting the AI roadmap to the executive team?

Presenting the roadmap before the business cases are ready. The roadmap without business cases is a wish list. Executive teams are not equipped to approve a prioritized use-case pipeline — they are equipped to approve specific funded initiatives with defined scope, cost, benefit, and accountability. Presenting the roadmap and the business cases together, as a single decision package, collapses two approval cycles into one and significantly increases the probability that Wave 1 moves forward with full executive commitment rather than partial endorsement.

How do we handle AI vendors who approach us during the planning process?

Accept the meetings; do not accept the demos until after the use-case pipeline is finalized. Vendor demos are designed to generate enthusiasm for specific solutions and will distort your prioritization if they happen before you have a clear view of your own use-case priorities. Use vendor meetings during the planning process to gather market intelligence — what implementation timelines are realistic, what integration complexity looks like, what the ongoing cost structure is — but defer solution selection until the Wave 1 use cases are confirmed and the business cases are drafted. Vendors who pressure for faster decisions before your planning process is complete are telling you something important about how they operate as partners.

How should we think about build versus buy for Wave 1 use cases?

For mid-market organizations entering their first wave of AI deployment, buy or configure is almost always the right default for Wave 1. Custom model development requires data science capability, MLOps infrastructure, and model maintenance capacity that most mid-market organizations do not have and should not acquire as their first AI investment. The appropriate time to consider building custom models is when you have validated a use case through a commercial solution, understand the data requirements well, and have identified a specific limitation of available commercial options that a custom build would address. In our experience, organizations that lead with custom build for Wave 1 are trading speed and risk management for perceived differentiation that the organization is not yet positioned to sustain.

The AI Roadmap: A 90-Day Planning Framework for Operations and Strategy Leaders

Most operations and strategy leaders at mid-market companies understand the urgency of AI adoption but lack a structured planning process that translates executive intent into a funded, sequenced, and governable roadmap. This framework provides the phase-by-phase methodology — from current-state inventory through business case construction — to close that gap in 90 days.