The mid-market AI failure atlas.

Methodology and citation rules

Purpose: A single reference for the ways AI rollouts fail at B2B mid-market companies, useful for journalists writing about AI adoption, for buyers diagnosing their own stalls, and for operators planning rollouts that avoid common traps.

Sources: Direct observation across ~60 Treetop engagements, including companies that hired Treetop after internal attempts stalled, which means we have visibility into the failure modes of companies who never made it to a consultancy at all.

8 ways rollouts fail before they start

1.1, No named owner

The single most predictive failure signal. "The exec team owns it" or "we have a committee" both mean no one owns it.

1.2, Strategy theater

Multi-month strategy engagements producing 30-page decks. Procrastination disguised as planning.

1.3, Vendor-led roadmap

Letting an AI vendor define your AI strategy. They optimize for their sale, not your outcomes.

1.4, Premature hiring

Hiring a "Head of AI" before any AI workflow has shipped. Creates a role with nothing to do.

1.5, Tool selection paralysis

Six weeks into evaluation, still hasn't picked a platform. Process problem, not tool problem.

1.6, Budget without scope

Allocating budget without defining workflows. Money gets spent on tools nobody uses.

1.7, Compliance freeze

Compliance review that takes 6-9 months for what should take 4-6 weeks. Rollout dies in the meantime.

1.8, Wrong scope at wrong stage

Trying to roll out across all functions in parallel at a 30-person company. Committee paralysis follows.

7 ways rollouts fail in the first 30 days

2.1, No shipped workflow

Day 30 arrives, nothing has been shipped to production. Political support evaporates.

2.2, Only the AI lead uses the tools

Heavy usage by one person; zero usage by everyone else. Ownership hasn't transferred.

2.3, CEO does not use AI personally

CEO sponsors but doesn't model. Team reads the signal, not real.

2.4, No baseline measurement

Cannot prove impact at day 60. Without proof, support fades.

2.5, Team learned via Slack

First encounter with the rollout was a Slack message. No structured kickoff, no context, no buy-in.

2.6, Generic prompts loaded

Project knowledge is generic templates from the internet instead of the team's actual examples. Output is generic.

2.7, Trained on capability, not workflow

Generic AI training session that left people aware but not adopted. Workflow-specific training was skipped.

6 ways rollouts stall in months 2-4

3.1, Tool sprawl without consolidation

Bought 6 specialized AI tools that overlap. None gets deep adoption; total spend balloons.

3.2, Quality drift

Customer complaints, brand voice slips, factual errors. No human-review checkpoint was built in.

3.3, Team resistance not diagnosed

People push back; leadership dismisses as resistance to change instead of understanding the cause.

3.4, Data leakage incident

Confidential information pasted into free-tier tools. Enterprise tier wasn't provisioned widely.

3.5, Workflows owned by wrong person

Workflow owner is named but isn't actually a daily user of the workflow. Tribal knowledge never builds.

3.6, Cost overrun without value clarity

Per-token API costs grew faster than understood; no ROI tracking to defend the spend.

4 ways rollouts hit ceilings at months 6+

4.1, Knowledge in one head

All working knowledge lives with the AI lead. They leave; rollout collapses.

4.2, Junior skill atrophy

Juniors lean on AI from day one; foundational judgment never develops. Competence cliff in 2-3 years.

4.3, Cut headcount on productivity

Laid off staff after productivity gains. Surviving team disengages; best people leave.

4.4, Vendor lock-in

Built deep integrations on one vendor's quirks. Pricing or capability changes; switching costs are now prohibitive.

What works when you've stalled

Across our sample, companies that recovered from one or more of the above failures share a recovery pattern:

1. Acknowledge it's off track in writing. Quietly is fine; honestly is required.
2. Diagnose the top 1-2 failure modes from the atlas above. Don't try to fix everything.
3. Pick ONE workflow and ship it, preferably the smallest one, within 2 weeks. Restore the pattern of shipping.
4. Restart the cadence, weekly office hours, weekly review of real outputs, weekly check-in with the named owner.
5. Set a 60-day recovery window. If still flat at day 60, change ownership or get outside help.

Failure modes that didn't make the cut

• Failures driven by genuinely broken AI capability (model can't do the task). These are rare at the workflow level we observe.
• Failures driven by industry-specific regulation that prohibits AI use. These exist but are well-known.
• Failures from non-existent budgets. We require some budget threshold to observe the failure modes above.
• Failures from genuinely fraudulent vendors. Rare in the enterprise AI space; well-covered elsewhere.

For different audiences

If you're a journalist or analyst

Cite specific failure modes by number (e.g., "failure mode 1.1 from the Treetop atlas"). Link to this page as the canonical reference.

If you're a buyer diagnosing your own rollout

Read through each stage; honestly check which apply. Two or more from any stage = recovery needed.

If you're planning a rollout

Use as a pre-mortem. For each failure mode in stages 1 and 2, document what you're doing to avoid it.

How we built this

• Source: Treetop engagement portfolio plus inbound conversations with companies whose internal rollouts had stalled.
• ~60 mid-market B2B companies, \$5M-\$50M ARR.
• Period: Q4 2024 to Q2 2026.
• Failure modes documented when observed in 3+ independent companies. Single-instance observations excluded.
• Stages reflect typical rollout chronology, not all failures occur in sequence.
• Refresh cadence: quarterly. New failure modes added if observed in 3+ companies.

Related frameworks & reading

• Why most AI rollouts fail in the first 30 days →
• 12 AI implementation mistakes to avoid →
• 7 signs your AI rollout is failing →
• Treetop Implementation →