Back to Insights Educational article

The 90-Day AI Implementation Roadmap: From Scope to Production

Ninety days only works if you have already picked the use case. Most plans burn the first thirty days on scoping theatre and then act surprised at week ten. Here is the schedule I actually run.

Last updated: May 24, 2026 Implementation

~10 min read Reading time 7 Sections Guide Long-form guide

Ninety days is the wrong unit of analysis if you have not picked the use case, and most projects I see burn the first thirty days on scoping theatre.

$3 / $15

Claude Sonnet 4.5 input / output per million tokens

~€60 /mo

n8n self-hosted on a Hetzner CCX33

10 weeks

If you can't show production traffic by week 10, re-scope

01 The wrong unit

Week 1 is the only week where scope can still move cheaply

The pattern I keep seeing is this. Someone reads a McKinsey deck, books a kickoff, then spends three weeks running "AI opportunity workshops" while the actual workflow stays unwritten on a whiteboard. By the time anyone touches an API, half the runway is gone and the team still cannot name a single ticket type they want the system to handle. Most of the public 90-day templates make this worse by treating discovery as its own deliverable. It is not. Discovery is the cost you pay to write one paragraph: which user, which workflow, which metric and what number that metric is at today. If you can't write that paragraph in five working days, you do not have a project, you have a wish.

MIT's Project NANDA reported in July 2025 that 95 percent of enterprise GenAI pilots showed no measurable P&L return. The common thread in the failures was scope: pilots that explored instead of shipping one workflow to production.

I write the scope artifact in a Linear doc or a Notion page, never in Slack. The medium matters because Slack lets scope drift in silence. A page has a URL, a last-edited timestamp and a comment thread. That audit trail is what stops the founder from quietly adding "and also voice" in week 6. If you want a deeper map of the broader category landscape before scoping, the piece on agents vs chatbots vs copilots is the lens I use to talk operators out of building the wrong thing.

“ Discovery is the cost of writing one paragraph. Not a deliverable.

02 The stack you actually need

Pick boring infrastructure on day 8

The 2026 stack for a first production AI workflow is small. One frontier model API (I default to Claude Sonnet 4.5 via the Anthropic Agent SDK because the tool-use loop is the cleanest in the market right now, with GPT-5 as fallback for tasks where the cost of an extra round trip is fine). One eval tool, usually LangSmith, because it writes traces you can actually read back at 11pm three months in. One vector store if the use case needs retrieval – pgvector when you already run Postgres, Pinecone serverless when you want to skip ops, Weaviate when you genuinely need hybrid search and are willing to pay for it in operator time. One workflow runner, often n8n self-hosted on a Hetzner CCX33 box at roughly €60 a month, for the glue around the model. That is the whole list.

What I would not do in the first 90 days. I would not stand up Modal or Replicate for inference hosting if a hosted API solves the problem, because you are paying with your team's attention. I would not write a custom RAG framework when the Vercel AI SDK plus pgvector covers 80% of cases. I would not pick a vector store on the strength of a benchmark blog post written by the vendor of that vector store. Boring beats clever for the first ship. The a16z emerging architectures piece is still the most honest single map of the space and worth a read before you sign any procurement contract, even where it has aged. If budget realism is what you want, the breakdown in what AI implementation actually costs matches the prices I see in invoices.

03 Timeline

Ninety days, but only six of them matter

The schedule below is the one I run when the use case is already chosen. If the use case is not chosen, the schedule does not start. I have watched too many teams launch a 90-day clock against an undefined target and then act surprised at day 70.

FIG. 01 – THE 90-DAY SCHEDULE Six checkpoints. The rest is execution.

Days 1–5 – Scope artifact One Linear or Notion page. One paragraph. One number. Workflow, user, metric, baseline. If you can't write it in five days, the use case is wrong and no model will save it.
Days 6–14 – Eval set Fifty real examples with expected outputs Pulled from your own data, not synthesized. Eval first, model second. Teams that ship without an eval set ship a vibe, not a system.
Days 15–35 – Build Smallest possible loop that touches production data Hit the API, run the eval, look at the failures. The first three weeks of build are an exercise in resisting the urge to handle edge cases.
Days 36–55 – Integrations Glue to the existing systems via n8n or code Integration is roughly 60% of the build effort. Underbudgeting this is the single most predictable failure mode I see.
Days 56–70 – Shadow mode System runs alongside humans on 100% of traffic Output goes to a queue, not to the customer. You're comparing the system's answer to the human answer on real volume.
Days 71–90 – Live with a kill switch Real traffic, ramped, with a documented rollback 10% on day 71, 50% by day 80, full by day 90 if the eval holds. The kill switch is a feature flag a non-engineer can flip.

04 Build vs buy

The SaaS option is almost always cheaper at first and usually wrong by month nine

I see operators staring at a $499/month vertical SaaS and a $40K custom build and concluding the SaaS wins. On a 90-day horizon, it does. On an 18-month horizon, the SaaS lock-in tax and the inability to change the prompt or the retrieval logic compounds into a six-figure regret. The honest comparison is below. Read it as a starting point, not a verdict. Your specific workflow may genuinely live inside a Zendesk macro forever, and that is fine.

FIG. 02 – BUILD VS BUY AT 18 MONTHS Honest cost comparison once the novelty wears off

Dimension – Vertical SaaS – Custom build on Claude + n8n
01 – Time to first production traffic – 2–3 weeks – 8–10 weeks
02 – 12-month total cost (mid-market) – $15–40K – $50–120K with one engineer half-time
03 – Prompt and retrieval ownership – Vendor's, opaque – Yours, in git
04 – Switching cost at month 18 – High. Data and prompts are theirs. – Low. The repo is portable.

05 Definition of done

Day 90 is not a date, it is a checklist

Shipped, not theatre

Real traffic running through the system, not a demo URL
Eval set in CI. Every prompt change runs against it.
Per-call cost tracked. You can name the unit economics.
Kill switch tested at least once
Baseline metric from week 1 has moved by a number
The next workflow is already named on a page

Theatre, not shipped

Pilot has been at 10% traffic for over four weeks
Nobody can quote the per-token cost without a spreadsheet
Prompt changes happen in a vibes review, not against an eval
The rollback plan is "call the vendor"
Original scope has grown by more than 30%
The team can't say which week they slipped on

If you reach day 90 and three or more of the right column apply, the most useful thing you can do is stop, freeze the project for a week and re-scope. I know that is the advice nobody wants to hear in month three. It is still the cheaper move. The full audit framework I use on stalled programs is closer to what is in measuring AI ROI – the gap between "we are doing AI" and "the AI is doing the work" is usually a measurement gap, not a model gap.

Deloitte's June 2022 automation survey put the average payback period for scaled programs at about 22 months, up from 16 a year earlier. A 90-day scope exists precisely to beat that curve.

06 Contrarian

Most of what you read about AI implementation is written by people who have never shipped one

The published 90-day playbooks tend to come from three places. Consultancies selling six-month engagements. Vendors selling the platform that conveniently sits at the centre of the diagram. And LinkedIn personalities who have never owned a production on-call rotation for an LLM system. None of these sources have to live with the consequences of a bad week-2 decision, which is why their advice rounds to "do everything carefully" rather than "skip discovery if you already know the workflow and start the eval set on day one."

The other thing the playbooks get wrong is the assumption that the model is the hard part. It is not. The model is a paid API. The hard part is the integration into a CRM that nobody has touched since 2019, the consent flow your legal team will rewrite three times and the question of who carries the pager when the model decides to refuse a legitimate refund request at 2am. Talking to operators running this in production, the time split looks roughly 20% model, 60% integration and operations, 20% policy and rollback. The model is the easy part because Anthropic and OpenAI have done that work for you.

If you are catching up in late 2026

The temptation is to skip the eval set because "we are behind." Don't. Skipping evals does not save time, it shifts the cost from week 2 to week 14, when you cannot tell whether a prompt change made things better or worse. Before signing a partner, the questions in how to choose an AI implementation partner filter out the agencies that will quietly skip this step.

07 Working together

How we'd actually approach this

When operators come to us with "we want to do AI in 90 days," the first thing we do is push back on the timer. If the use case is named, the metric is baselined and the data is reachable, the timer makes sense and we run the schedule above with weekly checkpoints. If any of those three are missing, the first conversation is about getting them in writing, and we will say so before quoting anything. Our services page has the shape of the engagements we take, and the contact form is the right place to start if you want a thirty-minute call where we mostly listen and then tell you which of those three things is missing.

Frequently Asked Questions

How much should we budget for a first 90-day AI implementation?+ Realistic budgets for mid-market businesses run $80K to $250K for the build plus $5K to $25K per month in ongoing costs. Smaller scope at the low end, complex integrations at the high end. Budgets below $50K usually produce systems that don't survive contact with production. The breakdown by use case is in the 2026 AI implementation cost guide. Can we do this in less than 90 days?+ For a very narrow scope with no integration, 45 to 60 days is achievable. For anything that touches an existing CRM or billing system, 90 days is the realistic minimum. Implementations promised in 30 days usually skip the eval set and pay for it in month four. Should we hire internally or use an implementation partner?+ Most businesses are better served by an external partner for their first one or two implementations. Hiring AI talent in 2026 is expensive ($180K to $350K base for senior engineers in the US), takes months and you're learning what to look for at the same time you are interviewing. What if the workflow we picked turns out to be wrong?+ Discover this in weeks 1 to 4, not week 8. The scope artifact and the eval set exist specifically to surface "this isn't the right workflow" early. If you reach the build phase and realize the workflow is unsuitable, stop, return to scope selection and pick again. The cost of stopping at week 4 is two weeks. The cost at week 8 is the whole quarter. How do we know if our success metrics are realistic?+ Anchor against industry benchmarks and your own historical baseline. If you're claiming a 90% reduction in support time when published implementations top out at 60% to 80%, you are either smarter than the industry or wrong about your scope. The second is more common. When does the next 90-day cycle start?+ Immediately after day 90, while the team and the momentum are fresh. Skip the gap and the team loses focus, the system feels like maintenance overhead and the next implementation gets stuck in re-planning.

ANM SOLUTIONS / CONTACT US

Need help applying this to your business?

We turn AI insights into measurable business outcomes. Tell us about your workflow and we'll show you the highest-impact place to start.

Vendor Selection

How to Choose an AI Implementation Partner Without Overpaying

An honest look at the five categories of AI implementation partner, where each one fits and how to tell a useful vendor from an expensive deck factory in a single technical conversation.

Read article

Customer Support

AI for Customer Support: A Practitioner's Playbook (2026)

An operator's guide to shipping AI customer support in 2026. Model choice, vendor reality, tool design and the eval set everyone skips until it bites.

ROI

Measuring AI ROI: How to Get a Number You Can Defend (2026)

An opinionated guide to measuring AI ROI for CFO-leaning operators. Three return layers, a defensible formula, honest evidence standards and when to wind a project down.

Published: 2026-05-05 · Author: A&M Flow