Skip to main content
OT/ICS Cybersecurity

When Is a Plant Recovered? Four Numbers Your OT Incident Plan Has to Write Down

Eleven days separate two Coca-Cola statements about Fairlife: one disclosed the attack, the other said production was mostly back and systems were still being restored. Which day was recovered? MTD, RTO, WRT and RPO, defined for a plant, with a worked timeline you can copy into your plan.

Documentary photograph of an engineer in a hard hat and safety vest checking a tablet beside a dairy bottling line, milk bottles moving past an open control cabinet and a stack light showing green

Eleven days. That is the distance between two statements Coca-Cola made about Fairlife, its US dairy business, in July 2026. The first disclosed a ransomware attack. The second said most of the production was back, and that the systems were still being restored. So on which day was the plant recovered?

The facts first, because they are short. On 16 July, Coca-Cola's Form 8-K disclosed the attack and the suspension of Fairlife's US production. On 27 July, a company release looked back on "a temporary suspension of production operations" and reported most output running again at Fairlife's four US plants. The same release said work to "restore impacted systems and operations" was still going on. Stores stayed largely supplied from stock already made, and the company called the incident not material.

Read those lines as an analyst and you find three clocks: production, systems, and what reached the shelf, which was protected by product already in warehouses. Eleven days is the gap between two statements, not the length of the outage, and nothing public says what Fairlife's own plan counted as "recovered". The question is about your plan, not theirs.

Maybe you inherited the recovery chapter of a plant's incident plan and found one recovery time, lifted from the IT disaster recovery plan. Nobody was careless: IT recovery usually does end when the service is back. A plant needs four numbers, and they fit in one table three people can fill in an afternoon.

The four numbers, and what each one measures on a plant floor

The definitions come from two NIST documents. The contingency planning guide, NIST SP 800-34 Rev. 1, defines them in section 3.2.1. The OT security guide, NIST SP 800-82 Rev. 3, applies them to control systems in section 6.2.4.5. Where the two disagree, this article says so.

A worked timeline, on a plant invented for this article

Take an illustrative dairy packaging line, made up to show the method; none of it describes Fairlife or any real site. The plant manager has accepted an MTD of 72 hours for filling and packing, because finished-goods stock covers three days of orders and then the customer contracts bite. That reason goes into the plan beside the number, because the next reviewer will ask.

Engineering then sets a target for each system the function depends on, measured from the moment the line stopped:

SystemRTO end point an operator can seeRTOWRTTotalMargin against 72 h
Plant firewallApproved rule set loaded, plant traffic flowing4 h2 h6 h66 h
HMI serversLive values on every line screen12 h6 h18 h54 h
Line PLCsLogic verified against the approved baseline24 h20 h44 h28 h
Engineering workstationProject files open and the compare tool runs40 h26 h66 h6 h, red

Read it as a timeline. At hour 6 the network is trustworthy again. At hour 18 the operators can see the line. At hour 44 the controllers are verified. At hour 66 engineering has its tools back, and there are six hours left before the deadline the owner signed. The rule this table uses is a rule of thumb, not a standard: a margin smaller than one shift is red, because the first surprise of the night will eat it.

Now the surprise. The investigation finds that the attacker first logged in through a remote access account 12 days before the encryption, also invented. The latest PLC logic backup was taken after that login, so it may carry the attacker's changes. The rule for the plan is short: the recovery point must predate the earliest known compromise, not the moment the ransomware went off. That rule has a cost in retention. Mandiant's M-Trends 2026 puts the global median dwell time at 14 days, across all intrusion types, so keeping only a week of backups means the clean one may already be gone.

Going back further has a second cost. Three approved logic changes were made after the clean backup, and each has to be redone and checked. The PLC row's WRT grows from 20 to 28 hours, its margin drops from 28 to 20, and it is still green. A choice about the RPO, which sits outside the MTD on paper, just moved a number inside it. Does your current plan let you see that happen before the night, or only during it?

"We already have an RTO. IT set one for the SCADA years ago."

Our plan has a recovery time. Four hours for SCADA, signed off by IT. Why would I rip that up and start again with four acronyms?

Keep the number; stop letting it stand alone. Ask three people what has happened at hour four. The IT lead says the servers are back. The integrator says the screens are up. The shift supervisor says product is running in spec. Only one of those readings is about the plant. The fix is one column: write the end point as something an operator can observe, add the WRT, and see whether the total still fits the MTD.

A four-hour RTO with a twenty-hour WRT behind it is not a four-hour recovery. It is a twenty-four hour recovery that has been reported as four.

"I build PLC logic, not business cases."

Nobody has ever told me what an hour of this line is worth. I know PLCs, not finance. How am I meant to set an MTD?

Setting it is not your job. The owner of the function accepts the MTD, and NIST's own guidance puts it with the system owner working alongside the process owners. What the owner cannot do without you is see the physical limits, and those come from engineering and operations. Bring three things to the meeting: how long the process can run without the control system before harm starts, how long operators can hold it by hand and with how many people, and what forces a shutdown. Then ask the owner one question: "After how many hours does this stop being an inconvenience and start being a loss you cannot accept?" Record both the figure and the owner's reasoning.

If that gets a shrug, ask the inventory version the Fairlife release points at: how many days of product or storage stand between a stopped line and a missed order, a permit breach or a public notice? That is often the first honest draft of an MTD.

"Our integrator does the restores. My numbers depend on them."

The integrator holds our project files and does every restore. If they turn up on Monday morning, my 40-hour RTO is fiction, and nothing I write changes that.

Often true, and the table is how you prove it. The engineering workstation's 40-hour RTO only holds if the integrator is on site within a few hours. Now read the contract. If it says "best efforts, next business day", a Friday-night incident puts them on site after the deadline, and the row goes from red to impossible. NIST SP 800-34 Rev. 1 has a name for what happens next: when the RTO cannot be met and the MTD cannot move, open a Plan of Action and Milestones. The gap becomes a tracked risk with an owner and a funding decision, whether that is a spare image on site, a contract term at renewal, or a technician trained on the engineering software.

Back to the eleven days

Coca-Cola's two statements show three clocks running at once: production mostly back, systems still being restored, shelves protected by stock that was already made. A plan that says only "recover the systems" measures one of those clocks and calls it the whole story. The four numbers make the plan say which clock counts, who accepted it, and how much room is left.

Recovered is not the moment the servers answer. It is the moment every system a function depends on has met its written end point, finished its work recovery, and done both inside a deadline the owner signed.

Sources: Coca-Cola 8-K filing (16 Jul 2026) | Coca-Cola press release on restoring Fairlife (27 Jul 2026) | NIST SP 800-34 Rev. 1, Contingency Planning Guide, section 3.2.1 | NIST SP 800-82 Rev. 3, Guide to OT Security, section 6.2.4.5 | Mandiant, M-Trends 2026

ICS/OT Incident Response Plan: Write One That Works

Most OT incident response plans fail before anyone writes a word: no sponsor, no charter, no agreement on who decides. This workshop course takes you from a blank page to a plan your site approves and your operators can follow.

Each of the eight sections writes one chapter of your plan: the authority and scope, the inputs, severity in the language of the plant, command and decision rights, response steps, recovery and the readiness library.

You get the plan template, more than 40 templates and worksheets, 21 worked examples and a capstone with a full model answer.

50 videos, 7.7 hours. US$147 one-time through HKSM Courses. 30-day money-back guarantee when 50% or less consumed.

Explore the Course


ICS/OT Incident Response Plan course preview

Write the OT incident plan your site will actually use

Go from a blank page to an approved, usable plan for your industrial site, one chapter at a time. ICS/OT Incident Response Plan: Write One That Works includes 50 videos, the plan template, worksheets and a full model answer, delivered through HKSM Courses for US$147 one-time. 30-day money-back guarantee when 50% or less of the course has been consumed.

Explore the Course