The mid-market AI adoption playbook: first 90 days

A $30M ARR SaaS company cannot run a six-month AI transformation. The cost of capital is wrong, the team bandwidth is wrong, and the budget is wrong. What can run is a 90-day playbook that ships working agents inside specific workflows and produces measurable outcomes by day 60.

Here is the version we run. It assumes a mid-market company between $5M and $200M revenue with an existing growth function and an AI line item of roughly $50k to $250k annual.

Days 1-30: diagnose and pick one workflow

The first 30 days are not implementation. They are diagnosis. The team that wants to "deploy AI" usually has not picked which workflow yet. The first task is picking one.

Run an audit of every repetitive task in the growth function. Sales emails. SDR research. Content production. Lead enrichment. Reporting. Pipeline analysis. List them. Rank by hours spent per week. Pick the single workflow that has the highest hours spent AND the lowest judgment requirement.

That is the first workflow. Resist the temptation to pick three. Three workflows in parallel is a Big Four engagement, not a mid-market playbook. One workflow, one agent, one measurement.

Days 30-60: ship the agent

Days 30 to 60 is the build. By day 35 the agent has a prompt. By day 45 it has tool access to whatever data sources the workflow needs. By day 55 it is running alongside the human operator with a human-in-the-loop checkpoint.

The two things that kill agents in this window are scope creep and data plumbing. Both are managed by the operator running the engagement, not by the technology choice. Pick a small language model, give it specific tool access, and ship the v0.

By day 60 the agent has shipped at least one full week of work. Measure it. Compare it to the human baseline. Look for three things: speed (is the work happening faster), quality (is the work as good or better than the human baseline), and trust (does the team running the workflow believe the agent's output without verifying every step).

Days 60-90: scale or kill

By day 60 you know whether the agent works. If it does, days 60-90 is scaling: removing the human-in-the-loop checkpoint, expanding the agent's data access, integrating it with adjacent workflows. If it does not, days 60-90 is killing the agent cleanly and starting over with a different workflow.

The mistake mid-market teams make in this window is the third option: keep the agent alive and add humans around it. That is the worst of both worlds. The agent is not delivering and the team is paying for both the AI cost and the human cost. Kill the agent or scale the agent. Do not maintain it.

What to avoid

Three traps consume mid-market AI budgets.

First: agent sprawl. A team that deploys six different agents in 90 days creates a coordination problem larger than the one it solved. One agent at a time.

Second: vendor stacking. Six AI tools each doing a slice of one workflow is more expensive and less effective than one in-house agent built on a small language model. The "AI tool" market is the new MarTech sprawl. Resist.

Third: F500 templates. A McKinsey AI transformation deck is built for a different buyer. Most of the framework does not apply to mid-market. Borrow specific tactics, ignore the strategic frame.

What 90 days looks like when it works

One workflow agent shipped. One measurable outcome. Either scaling for the next 90 days, or shut down cleanly and starting on workflow two. A team that understands what AI deployment actually feels like, with one engagement under their belt. No transformation budget, no consultancy roadmap, no six-month implementation hangover.

That is the playbook. We run it inside IG engagements as the AI Adoption Sprint. The deliverable at day 90 is the agent, plus the operating manual the team needs to run the next two without us.