Scan, Pilot, Scale: A Three-Beat Loop
To bring AI into government, the state chose a three-beat method: scan, pilot, scale. Türkiye’s AI Action Plan puts this loop at the centre of both public and private sector applications. Every ministry will scan its use cases systematically, launch fast pilots in high-potential areas, and scale projects that clear the success threshold within 6 to 12 months.
The method is not specific to states. The same three beats work in a ten-person business. They work faster there, in fact.
Why Is the Question Being Asked Now?
BU BÖLÜMÜN ÖZETİ
- Plenty of experiments, few results
- A state declared a method
- Budgets are small and error margins thin
- Tools change fast
Four developments put this method on the agenda.
Plenty of experiments, few results
Most businesses have tried AI. Few have got a result. The difference is not talent but method. An experiment run loosely leaves no trace. Who tried it, what came out, why it stopped — none of it is written down. Six months later the same idea returns and gets tried again.
A state declared a method
The plan wrote its preferred approach openly. That gives the private sector a measure too. If public projects will be judged through this loop, anyone selling into them speaks the same language. And the method itself is well tested: starting with a pilot and growing by measurement has worked in corporate life for decades.
Budgets are small and error margins thin
Projects that start big and stop halfway are expensive. Pilots that start small and grow when they work are cheap. Scarcity makes the small step compulsory. The plan’s own SME pilot targets point the same way: an average productivity or quality gain of at least 10 percent, and at most four months from experiment to pilot.
Tools change fast
A project running a year can be obsolete when it finishes. A short pilot shrinks that risk. If a three-week trial turns out wrong, the loss is three weeks. The same mistake inside a year-long project costs money and credibility.
What Is Wrong?
BU BÖLÜMÜN ÖZETİ
- “Big plan first, then implementation”
- “Let’s try it once; if it fails, forget it”
- “A pilot is a small permanent system”
- “Scaling means making the same thing bigger”
Four common assumptions.
“Big plan first, then implementation”
It sounds orderly. In practice, months pass in correspondence. On the AI side the big plan clarifies after implementation, not before. The first pilot teaches more than the plan. Nobody can predict in advance how these tools will behave on your data. An idea that looks perfect on paper can stumble on day one.
“Let’s try it once; if it fails, forget it”
The opposite pole. An unplanned trial runs, no criterion is set, the outcome stays vague. Then someone says “we tried it, it didn’t work.” But what was tried was never written down. How long did it take, who used it, what was it compared against? Without answers there is no result, only an impression.
“A pilot is a small permanent system”
A pilot’s purpose is not to run but to teach. Built like a permanent system, it costs too much and becomes hard to unwind when it turns out wrong. A pilot must be cancellable. Deep integration, data migration and long contracts belong to the scaling stage.
“Scaling means making the same thing bigger”
It does not. Scaling means turning what the pilot taught into a permanent arrangement. That usually requires rebuilding. A pilot is a prototype; scale is a product. The rough edges tolerated in a pilot come back as faults at scale.
The Real Mechanism
BU BÖLÜMÜN ÖZETİ
- Scan: produce a candidate list
- Pilot: small, time-bound, measured
- Scale: decide, then rebuild
- The fourth condition: close the loop
The loop has three beats. I would add a fourth condition.
Scan: produce a candidate list
Review every process in the business. Filter with three questions. Does it repeat? Does it eat time? Is its output measurable? Processes answering yes to all three form the candidate list. Do not scan alone; the people doing the work must be at the table, because they know best where the hours go. The manager’s list and the employee’s list usually differ.
Pilot: small, time-bound, measured
Pick one candidate. A pilot has three conditions. Small: one process, one team. Time-bound: start and end dates fixed. Measured: before and after compared with numbers. The plan’s own pilot programmes make before-and-after measurement mandatory. Measurement need not be complex: elapsed time, items processed, error count.
Scale: decide, then rebuild
When the pilot ends there are three options: scale, retry, drop. All three are legitimate. Dropping is not failure but saving. If you scale, the work gets rebuilt: thresholds, approval points, a named owner, a maintenance plan. The human-approval flow is designed at this stage.
The fourth condition: close the loop
At the end of every round, write what was learned. One page is enough. Without it, the same mistakes get retried. On the state’s side, progress reports do this job. On yours, a notebook does. Three lines suffice: what we tried, what happened, what we decided.
Who Is Affected, and How?
BU BÖLÜMÜN ÖZETİ
- The micro business
- The growing SME
- The corporate organisation
- The solution provider
The same loop behaves differently in four settings.
The micro business
The advantage is speed. Scanning takes an hour, a pilot two weeks. With decisions in one person’s hands there is no approval chain. The risk: the pilot blends into daily work and nothing gets measured. The fix is to bind the pilot to dates.
The growing SME
The most productive setting. Processes are written down, data has accumulated, there is a team. Scanning produces a real list. The risk is launching too many pilots at once. The rule: at most two pilots running at a time. A third slows the first two.
The corporate organisation
The difficulty is speed. An approval chain can hold a pilot for months. The fix is a separate decision path for pilots: small budget, short duration, fast approval. Most corporates lose by slowing the pilot, not by choosing the wrong one.
The solution provider
This loop will be your sales language. The public sector will buy this way: pilot, measurement, badge, scale. Selling a customer “a two-week pilot with a measurable result” is easier and more credible than selling “a big transformation”. The public procurement chain is built on the same logic.
Decision Order
BU BÖLÜMÜN ÖZETİ
- One: hold the scanning meeting
- Two: pick the easiest candidate first
- Three: write the criterion before you start
- Four: put the decision date in the calendar
Four steps.
One: hold the scanning meeting
One hour with the team. List processes, filter with three questions, produce a candidate list. Keep it to five lines. A long list makes choosing hard and turns the meeting into brainstorming.
Two: pick the easiest candidate first
Do not start with the biggest problem. Start with the one that gives the clearest result. The first pilot’s job is to give the business the habit of piloting. Confidence won on an easy candidate makes the hard one possible.
Three: write the criterion before you start
Two numbers get written before the pilot begins: where we are now, where we want to be. Without them it is not a pilot but a trial. Criteria set afterwards always produce “looks good”. People read the results of their own effort generously; a number written upfront is the simplest protection against that.
Four: put the decision date in the calendar
The pilot’s end date and the decision meeting go into the calendar at the start. A pilot without a date does not end; it drifts. A drifting pilot is neither scaled nor closed.
Where to Start?
BU BÖLÜMÜN ÖZETİ
- Week one: the scan
- Week two: criteria and pilot definition
- Weeks three and four: the pilot
- Week five: the decision
Four jobs in the first month.
Week one: the scan
One hour, a five-line candidate list.
Week two: criteria and pilot definition
One page: which process, who runs it, which tool, which two numbers, which dates.
Weeks three and four: the pilot
Run it. Keep a log. Note where it stalls. Do not change the criterion mid-pilot.
Week five: the decision
Pick one of three options and write the one-page learning note. Then start the second round.
What Not to Do?
BU BÖLÜMÜN ÖZETİ
- Running the pilot in secret
- Changing the criterion to fit the result
- Leaving a successful pilot unscaled
- Opening too many pilots at once
Four traps.
Running the pilot in secret
A pilot hidden from the team distorts its own results. People who do not know what is happening either resist it or ignore it. State the purpose and the duration openly.
Changing the criterion to fit the result
Loosening the measure when the result comes in weak is the most common error. It also cancels the learning. The criterion is written upfront and stays fixed.
Leaving a successful pilot unscaled
Surprising but frequent. The pilot works, everyone is pleased, then daily work takes over and the job stays half-finished. The scaling decision also needs a date and an owner.
Opening too many pilots at once
Attention splits and nothing gets measured properly. Two is the limit; for small teams, one.
A Solid Digital Foundation
BU BÖLÜMÜN ÖZETİ
- The candidate list
- The pilot definition page
- The learning notes
- The scaling checklist
Four stones hold the loop up.
The candidate list
Processes filtered by three questions. Refreshed yearly, because new tools change the list.
The pilot definition page
Process, person, tool, two numbers, dates. One page. Its brevity makes starting easy.
The learning notes
One page at the end of each round. Over time this becomes the business’s most valuable internal resource.
The scaling checklist
Threshold, approval, owner, maintenance, rollback path. This is what turns a pilot into a product, completing the move from demo to production.
Frequently Asked Questions
Sık Sorulan Sorular
For most business processes, two to four weeks. Shorter produces too little data to measure; longer blurs into daily work and loses its boundary. One test decides the length: will the process repeat often enough in that window to compare two numbers? For a process that runs monthly, a two-week pilot is meaningless; there you need three months.
No, you gain two things. Knowledge: you now know that road is closed and will not spend there again. And habit: the team has learned to run a pilot, so the second one moves faster. That is the whole logic of piloting; it is kept small precisely because it might turn out wrong.
Mostly into the second and third beats. SME vouchers support the trial and pilot stage; GPU credits and sector pilot programmes support scaling. Scanning is your own job; nobody else can tell you which process eats your hours. So do the scan without waiting for support. When a call opens, whoever holds a candidate list applies.
No. Small, reversible changes get made directly; a pilot only adds weight. A pilot is needed in three cases: when the workflow will change, when money or time gets committed, or when an error would reach the customer. If none of those apply, try it and move on.