How to Decide if APC Is Worth It: A 15-Minute Self-Assessment
The question I get most often from plant engineers and operations managers is: “Does APC make sense for us?”
There is a standard consulting answer (“it depends, let’s do a study”) and there is a useful answer. This is the useful one.
Run through the following checklist. It takes 15 minutes if you have access to your historian. It tells you whether you have the prerequisites for APC — and, if not, which gaps need to be closed first.
Part 1: Process fundamentals (5 questions)
1. Is your process continuous or at least semi-continuous?
APC delivers its highest value in continuous or semi-continuous processes where conditions change on timescales of minutes to hours. Batch processes can benefit, but the control problem is structured differently — the feasibility assessment differs.
Score: Yes = proceed; Batch only = discuss separately
2. Are there at least two manipulated variables (valves, drives, setpoints) that interact with two or more controlled variables?
APC’s core advantage over standard PID is handling interactions. If you have a truly single-input single-output process, a well-tuned PID is already optimal — APC adds complexity without benefit.
Score: 2+ MVs interacting with 2+ CVs = yes
3. Does your process have constraints that limit performance — equipment limits, quality specs, environmental limits?
Constraint handling is where MPC earns its cost. If your process runs well below all physical limits and quality specs are comfortably met, there is little to squeeze. If operators are constantly navigating the edges of multiple constraints simultaneously, MPC was designed for this.
Score: Regularly navigating 2+ active constraints simultaneously = strong yes
4. Is there at least one key performance indicator (throughput, energy, quality, yield) that you know you’re not optimizing?
This is the economic driver question. APC is most easily justified when you can name a specific metric and explain why it is currently suboptimal. “We run below capacity because of quality variance” or “energy cost per tonne is higher than our competitors” are specific and actionable. “We’d like to improve generally” is not.
Score: Named KPI with quantified gap = yes
5. Does your process respond predictably to control inputs (at least qualitatively)?
APC requires a process model. Before building the model, you need to believe that the process behaves consistently enough to model. If the process response is completely different every time due to highly variable raw materials, uncontrolled disturbances, or frequent mechanical faults, the model will not generalize and APC will struggle.
Score: Generally consistent response to inputs = yes
Part 2: Data and instrumentation (5 questions)
6. Do you have a historian with at least 6 months of continuous data at a scan rate of ≤ 5 minutes?
Model identification requires sufficient historical data with enough input variation. Six months covers seasonal effects and multiple operating conditions. Scan rates faster than 5 minutes capture process dynamics adequately for most applications.
Score: Yes = proceed; No historian or < 1 month = data infrastructure gap
7. Are your key process variables (temperatures, pressures, flows, quality) actually in the historian, not just on the DCS display?
This is more common than it sounds. Many plants have sophisticated DCS displays but minimal historical logging. If key variables are not logged, you cannot identify a model — and you cannot validate that APC is working after deployment.
Score: Key variables logged continuously = yes
8. Is your primary control layer (PID layer) functioning and mostly in automatic mode?
APC sits on top of the base layer. If the base layer is poorly tuned or mostly in manual, APC cannot function — it sends setpoints to controllers that don’t follow them. A common mistake is trying to implement APC on top of a broken base layer. The result is a failed APC project, not a fixed base layer.
Score: Base layer > 80% in automatic = yes; < 80% = fix base layer first
9. Do you have a measurement (hard sensor or existing soft sensor) of the key quality variable at a frequency useful for control?
If your quality measurement has a 4-hour lab delay and there is no way to get a faster proxy, APC cannot close the quality control loop — it can only predict based on process conditions. This is still useful, but it changes the scope and expected benefit.
Score: Quality measurement < 30 min lag = yes; Lab only with > 2 hr lag = soft sensor may be needed first
10. Are your key actuators (final control elements) in good mechanical condition?
APC moves actuators more actively than conservative manual operation. Sticky valves, worn drives, and failing actuators become immediately visible — and problematic — when APC is active. A mechanical audit is part of any serious APC project.
Score: Actuators recently serviced / known good = yes
Part 3: Organizational readiness (5 questions)
11. Does operations management actively support improving automation?
This is not a soft question. The majority of APC failures I have seen are not technical — they are organizational. An APC system installed over the objections of operations management will be switched off within months of commissioning. The project sponsor needs real authority.
Score: Active management support = yes
12. Do your control engineers understand the existing base layer?
Someone in your organization needs to own the APC system after deployment. If there is no internal engineer who understands control at a sufficient level to maintain and troubleshoot the system, you are dependent on the implementer indefinitely. This is not fatal, but it is a risk and a cost.
Score: At least one internal engineer with solid controls background = yes
13. Are operators willing to trust automatic control if it demonstrably performs better?
Operator acceptance is not guaranteed. Operators who have been burned by a previous failed automation project, or who take professional pride in manual operation, may resist APC regardless of its performance. An operator co-design process — where operators provide input on the control objectives and test the system in simulation — dramatically improves acceptance.
Score: Operators willing to evaluate based on performance = yes
14. Can you run a controlled pilot on part of the process without affecting production?
The lowest-risk APC deployment starts with a pilot on a limited scope — one loop, one section, one product grade — before full deployment. If your process cannot support a pilot (fully continuous, fully integrated, no sectioning possible), the deployment risk is higher and the project structure needs to account for it.
Score: Pilot possible on partial scope = yes; All-or-nothing deployment required = higher risk, plan accordingly
15. Do you have a quantified business case, even a rough one?
“We’d like to improve” does not justify the investment. A business case does not need to be precise — a rough calculation is sufficient to decide whether a project makes sense. “If we recover 5% throughput on our €30M/year line, that’s €1.5M. APC typically costs €50k–€200k to implement and €20k/year to maintain. If the payback is 2 months, it’s worth investigating.” That calculation takes 10 minutes.
Score: Rough business case with positive payback < 18 months = yes
Interpreting your score
13–15 Yes: You have a strong foundation. A 4-week diagnostic phase to confirm the model structure and quantify the opportunity is the right next step.
9–12 Yes: Promising, but there are specific gaps to close first. The questions you answered No will tell you exactly what to fix — base layer, data infrastructure, or organizational readiness.
6–8 Yes: Potential exists but the prerequisites are not in place. A realistic roadmap starts with infrastructure: historian setup, base layer tuning, key measurements.
< 6 Yes: APC is likely premature. The first priority is process stability and data infrastructure.
What to do with this result
If you scored 9+, the fastest way to convert this into a real number is a data-based feasibility study:
- Export one week of historian data for your key loops
- We analyze the data and return a 1-page report: loop health, identifiable interactions, and a rough improvement estimate
- This takes 3–5 business days and costs you nothing except the data export
The 1-page report tells you whether there is a project worth pursuing — before any consulting fees.
Dr. Rafał Noga is an industrial AI and APC engineer with experience at CERN, IAV GmbH, SkySails Power, and evosoft/Siemens. He offers free process diagnostics for qualified industrial processes.
Have a project or a question?
Contact Dr. Noga →