A Machine-Learning Impact Wrench Records Every Bolt It Drives and Flags the Ones You Got Wrong. It Costs $450. Your Deck Builder Has Never Heard of It.
Pick up a Milwaukee M18 FUEL ONE-KEY impact wrench with TORQUE-SENSE. Drive a structural bolt. Custom-designed sensors inside measure what happened during that fastening event, feed data through a machine-learning algorithm trained on thousands of tightening profiles, and determine whether the joint hit its target torque. Green light means good. No green light means try again. Every single trigger pull gets logged to the ONE-KEY app, downloadable for inspection documentation that proves, bolt by bolt, that work was done correctly.
This tool exists. You can buy it. Its platform has been deployed across utility-scale solar farms and commercial steel erection projects for two years. Hilti makes a competing system, the Nuron SIW 4AT-22 with its Adaptive Torque Module, that collects data on every embed t-bolt a curtain-wall installer drives and flags errors before the facade goes up. Atlas Copco sells torque-controlled electric tools that detect cross-threaded screws, missing washers, and stripped threads in real time for manufacturing and industrial assembly.
Residential construction, where 90 percent of deck collapses originate from a single connection failure, uses a cordless impact driver set to "3" and a builder's best guess about when the bolt feels tight enough.
What the Machine Learning Actually Does
Milwaukee's TORQUE-SENSE is not a torque limiter. Standard impact wrenches with "torque control" use blow-counting algorithms: they count how many impacts the mechanism delivers and stop after a preset number, assuming that a certain number of hits equals a certain torque. That assumption falls apart fast. Rust on the threads, a cold bolt, a wet washer, a slightly oversized hole in the ledger, the wrong lubricant on a Simpson Strong-Tie connector. Any variable that changes friction between fastener and material throws blow counting off.
TORQUE-SENSE uses purpose-built sensors, designed and manufactured by Milwaukee, that measure what the joint is actually doing during each impact event. Its ML model interprets sensor data against thousands of known tightening profiles and calculates achieved torque in the joint, not torque the tool delivered. A blow counter tells you what the tool did. ML tells you what the bolt did. That distinction is worth everything.
Hilti's approach is complementary. Their Adaptive Torque Module sits atop the SIW 4AT-22 impact wrench and records data points for each fastener. It targets curtain-wall bracket installation, where an ironworker might drive two hundred t-bolts in a day. Hilti's engineering team cites research showing that the human brain involuntarily switches to rest mode during repetitive tasks. Hands-on fatigue accumulates. Attention drifts. On bolt 140, the AT Module catches what the worker's body stopped caring about thirty bolts ago. It generates a log that a QA inspector can review without touching the wall.
Sixty Million Decks, Zero Verified Bolts
NADRA, the North American Deck and Railing Association, estimates that nearly half of the more than 60 million residential and commercial decks in the United States are past their useful life. Of deck collapses reported nationally, 94 percent resulted in personal injuries. Virginia Tech researchers who reconstructed various deck-to-house connections and tested them to failure found that many collapsed at loads much smaller than what design should have supported. Professor Joe Loferski put it plainly: if a deck is designed and constructed to carry intended loads, very few failures would occur.
Ledger connections are the primary failure point. According to NADRA, 90 percent of deck collapses trace to separation between the deck's ledger board and the house. Michael Morse, who studied deck construction and building codes for 25 years as founder of DeckLok, pointed to the mechanism: metal fasteners embedded in wood depend on resistance of wood fibers to hold them in place. Water follows the fastener into the wood. Seasons of freezing, thawing, and heat cause wood to dry, split, or rot. Once a fastener starts to withdraw, it takes almost nothing for it to pull the rest of the way.
IRC Section R507.9.1 specifies that ledger connections require either 1/2-inch lag screws or through-bolts at spacing determined by joist span and species group. Detailed table. Specific fastener schedule. But the code requires no verification that the fastener achieved its intended clamping force. An inspector checks that bolts are present and appear tight. Nobody measures torque. Nobody records it. Nobody comes back in five years to check if the bolts are still doing their job.
What Commercial Gets That You Don't
Structural steel erection follows AISC 360, which specifies minimum bolt pretension values for slip-critical and bearing-type connections. That standard requires one of several verification methods: calibrated wrench, turn-of-nut, twist-off-type tension control bolts, or direct-tension indicators. Each method produces a measurable, documentable result. An inspector can verify that a W10x49 beam connects to a W14x90 column with A325 bolts torqued to specification. Data exists. Building files contain it.
Solar installation has adopted smart torque tools aggressively because one undertorqued mounting bolt in a 500-acre array means a panel flies off in a storm and the contractor eats a warranty claim. Milwaukee designed TORQUE-SENSE specifically for this market, eliminating the two-step method (impact to snug, torque wrench to final value) and replacing it with a single operation that ML verifies in real time. Milwaukee claims this delivers up to three times faster installation speed while maintaining target torque better than blow-counting algorithms.
Hilti's Tracefast technology goes further. Every fastener gets a data matrix code stamped during manufacturing. Scan it with the Hilti Connect App on a smartphone, and you see fastener details, setting instructions, training documentation, inspection certificates. In ten years, a maintenance inspector can scan a bolt head and pull up when it was installed, by whom, at what torque, using which tool, on which project. A single fastener becomes a permanent record.
Meanwhile, a residential deck builder installs a 1/2-inch lag screw by driving it with an impact until the washer bottoms out and the bit starts to cam. Quality control: the bolt is there.
Why Price Is Not an Excuse
A Milwaukee M18 FUEL ONE-KEY 1/2-inch impact wrench retails for roughly $350 to $500 depending on configuration and kit. Same price range as the non-smart version of the same tool class. ONE-KEY is a free app. Data logging is included. ML torque verification is built into the firmware.
A deck builder who already owns an M18 platform is buying into the same battery ecosystem. Incremental cost of getting torque verification is essentially the price difference between the ONE-KEY model and the base model. For a tool that a professional uses every day, that gap is one afternoon of labor on a job. Fasteners don't get more expensive. Lumber doesn't get more expensive. Inspections don't take longer because measurement happens while the work is happening.
Hilti's AT System is pricier, sitting in the commercial-industrial tier for contractors managing large curtain-wall or structural steel projects. But embedding a sensor in a power tool and running its signal through an algorithm adds marginal manufacturing cost. Nobody has put it in a $200 residential-grade impact driver, and that is not an engineering problem. It is market segmentation. Manufacturers sell verification as a premium feature to markets that demand documentation. Residential doesn't demand it, so residential doesn't get it.
What a Torque Record Would Actually Prevent
Consider a standard deck failure timeline. A contractor builds a deck in 2020. Ledger bolts are driven with an impact driver. An inspector walks the deck, sees bolt heads, signs off. Six years later, the homeowner hosts a party. Twenty-two adults on a 12-by-16 deck. Live load exceeds what deteriorated connections can hold, and the deck peels away from the house.
If the contractor had used a smart torque tool, two things would be different. First, each bolt would have been verified to its target clamping force at installation, catching any that were undertorqued because of misaligned holes, insufficient thread engagement, or defective hardware. Second, a torque record would exist. When the lawsuit arrives, the contractor has documentation showing every fastener met specification on the day of installation. That shifts the question from "did you torque these correctly?" to "what happened to the wood between then and now?" Legally, those are two entirely different conversations.
For a homeowner, a torque record attached to the building permit file would provide something that doesn't currently exist: a baseline. A structural engineer hired to evaluate a ten-year-old deck could compare current bolt condition against original installation data and make an informed judgment about remaining capacity. Without that baseline, evaluation is a visual inspection and a guess.
What You Can Do Right Now
If you are building a deck, specify torque values for structural connections in your contract. IRC R507.9 gives you the fastener schedule. Lag screw manufacturers publish recommended installation torque on their technical data sheets. Write that number into the contract and require the builder to verify it with a calibrated tool. A beam-type torque wrench costs $40 and takes ten seconds per bolt. No technical reason not to do this. No economic reason either. Just habit.
If you are a deck builder, look at Milwaukee's ONE-KEY platform or DeWalt's Tool Connect ecosystem. ML-verified tools cost what you'd spend on a good impact wrench anyway, and the data log gives you a litigation shield that a handshake and a memory do not. When the callback comes in 2032 and someone's lawyer wants to know whether you torqued the ledger bolts, you pull out a spreadsheet with timestamps, torque values, and GPS coordinates for every fastener on the job. Worth more than the price of the tool.
If you are a building official, consider requiring torque verification on ledger connections the way commercial codes require verification on structural steel. Retail-available technology, negligible cost, and it addresses the failure mode responsible for nine out of ten deck collapses in your jurisdiction. Tools already exist. Code hasn't caught up.
Limitations
Milwaukee's TORQUE-SENSE ML model accuracy claims are based on company testing for solar industry applications. Independent third-party verification of the algorithm's performance on wood-to-wood or wood-to-rim-board residential connections was not publicly available at time of publication. NADRA deck collapse statistics derive from reported incidents only; actual failure rates are higher because many collapses go unreported when no emergency crews respond. NADRA's 60 million deck count and "half past useful life" estimate lack a described methodology. IRC torque requirements, or the lack thereof, vary by jurisdiction because local amendments can modify the base code. Pricing for smart torque tools reflects retail in August 2026 and varies by configuration and retailer.