Ting Guides · For the CEO
How do you plan a 30-day AI pilot that ends in a decision?
The short answer
Write the decision rule before day one. Measure four numbers on the current process: volume, cycle time, working time per request and error rate. Set a target for each, and agree in advance how many targets must be met to expand, which result means stop, and that anything in between means change one thing and measure again.
Reviewed by Badr Al-RashedUpdated
What it depends on
A problem small enough for thirty days
Pick one process with a clear start and end, enough requests to measure in a month, and an owner who feels the problem. "Try AI in sales" is not a pilot; "cut the time from quote request to quote sent" is.
A baseline you measured, not remembered
Measure the current process before anything changes. Without numbers from before, day thirty has nothing to be compared with, and the result becomes a matter of opinion.
A decision rule written before the results
Agree in advance how many targets must be met to expand, and what result means stop. A rule written after the results tends to bend to fit them.
A purpose the business agreed
RAND's interviews with experienced practitioners found that misunderstanding or miscommunicating a project's purpose is the most common reason AI projects fail. A pilot charter signed by the process owner is the cheapest guard against it.[1]
An owner for day thirty-one
Name now who runs the system if it expands, and who pays for running it. A pilot that succeeds with no owner stalls anyway.
What goes on the one-page pilot charter?
Pilot charter, fill in before day oneProblem: what is slow, costly or error-prone, in one sentence Process: where it starts and where it ends Owner: name and role in the business Start date: · End date: start date + 30 days
Baseline, measured before day one Volume: requests handled per week Cycle time: hours from receiving a request to finishing it Working time: minutes of staff work per request Error rate: share of requests that needed a correction
Target at day 30 for each measure, and how it is measured:
Decision rule Expand if at least of the four targets are met. Stop if or fewer are met, or if an unacceptable risk to customers or data appears. Adjust in any other case: change one thing and measure again.
Signed before day one by: process owner · sponsor
Write the targets and the rule before the pilot starts, not after the first results.
What do you measure, and how?
Four measures cover most operational pilots. Take each from the system of record or a timed sample, never from an estimate.
| Measure | What it means | How to measure it |
|---|---|---|
| Volume | Requests handled per week | Count closed requests in the system of record |
| Cycle time | From receiving a request to finishing it | Average of finish time minus receive time |
| Working time | Minutes of staff work per request | Time a sample of real requests |
| Error rate | Requests that needed a correction | Corrected requests divided by all requests |
What is each of the four weeks for?
Hold a short review at the end of every week: what we measured, what changed, the next action and who owns it.
- Week one
run it beside the old way
The AI drafts while people still do the work as before. Compare the two and record every case it got wrong, with the reason.
- Week two
hand over the cases it handles well
Let the AI take the kinds of request it got right in week one, with a person checking each output. Keep the difficult cases manual.
- Week three
measure against the baseline
Enter the four numbers, compare them with the baseline and the targets, and change at most one thing so you still know what caused the result.
- Week four
apply the rule
Take the last week's numbers to the rule you signed. The outcome is expand, adjust or stop; an open-ended extension is not one of the options.
Day 30: apply the signed rule
- Stop0 or 1
- Adjust2
- Expand3 or 4
If the rule says expand, how do you expand?
What changes for companies in Saudi Arabia
General information, not legal advice. Each regulatory point summarises the official source cited beside it, as checked on the date shown under Sources; confirm how it applies to your company with your legal adviser.
SDAIA counts pilots that reach production
SDAIA's AI Adoption Framework (version 2, May 2025) lists tracking the transition of projects from pilot to production among its indicators, and says the move from experimental stages to implementation "may serve as an indicator of enterprise maturity". In our view, a pilot that ends in a written decision is how that transition gets counted.[2]
Real records in a pilot
If the pilot uses real customer or employee records, the Personal Data Protection Law (Royal Decree M/19 of 9/2/1443H, amended by Royal Decree M/148 of 5/9/1444H) applies from the first day: Article 2 applies it to any processing of personal data in the Kingdom.[3]
Ramadan and the Eid holidays
If the thirty days include Ramadan or an Eid holiday, volume and cycle time will move for reasons that have nothing to do with AI. Move the window, or measure the baseline over the same kind of period.
A quote-drafting pilot at day thirty
ExhibitWorked example
Built on the example in our 30-day pilot plan file, here at the quotes desk of Rasikh Trading, part of the fictional Rasikh Holding: the AI drafts each quote and the salesperson reviews it. The rule was signed before day one: expand at three or more targets met, stop at one or fewer.
Targets met at day 30, out of four
| Measure | Baseline | Target at day 30 | Week 4 | Target met? |
|---|---|---|---|---|
| Requests handled per week | 40 | 40 or more | 42 | Yes |
| Cycle time (hours) | 26 | 18 or less | 17 | Yes |
| Working time per request (minutes) | 35 | 25 or less | 26 | No |
| Error rate | 8% | 5% or less | 6% | No |
The decisionTwo of the four targets were met, so the rule says adjust: change one thing, such as the review step, and measure again for a fixed, short period. Nobody had to argue about whether the pilot was a success.
Questions people ask
Why thirty days?
Can we extend the pilot if the results are unclear?
Who should sign the pilot charter?
What Ting does and doesn't do here
What we do
- When a pilot says expand, we specify the production system in an AI Solution Blueprint, and the targets from your charter become acceptance criteria the built system has to pass.
- If the pilot needs one decision made first, such as what to measure or which process to try, Targeted AI Advisory answers it in writing.
Where Ting helps: AI Solution Blueprint
What we don't do
- We don't run open-ended pilots without a written decision rule.
- We don't take a pilot into production as it stands; the Blueprint comes first.
Sources
Primary sources only, each read in full on the date shown.
- The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed (RR-A2680-1, 2024)RAND Corporation · checked
- AI Adoption Framework, version 2 (May 2025)SDAIA · checked
- Personal Data Protection Law (Royal Decree M/19 of 9/2/1443H, amended by Royal Decree M/148 of 5/9/1444H)SDAIA · checked
