An AI Implementation Checklist for Companies Without a Dev Team
Before implementing AI, a company should be able to name the goal and its baseline number, the first department, the system of record, an internal owner, and what the AI may never do alone. During the build, knowledge comes before integration, and integration before autonomy. After launch, someone has to watch output and keep the company’s record current. The full checklist is below.
This is the list we run on our own projects. It’s written so an owner or an operations lead can use it on any vendor, us included.
Before anything gets built
1. One goal, in a sentence the owner would say. “Every lead gets a reply before they call someone else.” If the goal needs a slide to explain, it isn’t ready.
2. One number, with today’s value. Minutes to first reply. Days from request to proposal. Write down where it stands now. Without a baseline you’ll never know whether any of this worked.
3. One department. Pick the team with the most repetitive, high-consequence work. The rest of the company goes second.
4. One system of record. Decide which system holds the truth for that department. Usually it’s the CRM. If two systems disagree, settle that before an AI starts reading both.
5. One owner on your side. A named person who’ll learn the system first and answer for it. Usually an operations lead or office manager. It doesn’t need to be anyone technical.
6. A clean enough system. If the CRM has 2,000 tags, the agent will believe all of them. Clean it first.
7. A written “never” list. What the AI may not do without a person. Ours starts with: never send a message on its own, never overwrite a record.
During the build
8. The company gets written down before any tool is connected. What’s true, how you talk, who you serve, what you won’t say. Scoped to the first department. In files you own.
9. Integration reads before it writes. The first connection to your CRM should be read-only, then append-only. Nothing should be able to overwrite.
10. Drafts before sends. Every outgoing message starts as a draft a person approves. Extra autonomy is granted later, one permission at a time, by a person with a name.
11. Your team tests with real cases. Give the owner from item 5 last month’s actual leads or requests and have them judge what the system would have done.
After launch
12. The system reports its own number. The metric from item 2 should arrive weekly without anyone building a report.
13. Someone watches for drift. Models update. Prices change. Output slides. Whether that’s your vendor or your own staff, it has to be someone’s job.
14. Training by role, then a check at 30 and 90 days. The estimator needs different training than the front desk. Then come back and see who’s using it and who’s gone back to the old way.
How to use this on a vendor
Read them the list and ask which items they handle, which are yours, and which they skip. The answers will tell you a lot. At MKTNG, items 1 through 5 happen in the first three of our six steps, 6 through 11 are Foundation, and 12 through 14 are what we call Managed Operation.
AI implementation services · How the six steps work
Questions we get
What should be on an AI implementation checklist?
A stated goal with a baseline metric, a first department, a designated system of record, an internal owner, data cleanup, a list of actions the AI can’t take alone, a knowledge foundation, staged integration and autonomy, real-case testing, automatic reporting, drift monitoring, and role-based training.
Can a company implement AI without developers?
Yes, with an implementation partner who builds and operates the system. The company still needs an internal owner, but that person doesn’t need technical skills.
What’s the first step in AI implementation?
Naming one goal and recording its current number. Everything else is sized from that.
What is the most common AI implementation mistake?
Connecting a tool to messy systems before the company’s knowledge has been written down. The tool then repeats outdated information with confidence.
What is Groundwork?
Groundwork is the structured record of your company that every system we build reads from. It’s item 8 on this list.