How our recommendations work

There's no single "best" DEWALT drill. A cabinet installer, a deck builder and a weekend DIYer each need something different, and the right answer also depends on price. So we don't publish a fixed top ten. Our model re-ranks every product around the needs you set and looks for the best trade-off between performance, price and fit for you.

Here's exactly what happens when you move a slider.

1. Only current products, from the source

Every product comes from DEWALT's own regional websites for the US, the UK and the EU. We check those listings daily. When DEWALT stops listing a product in a market, it drops out of our results for that region, so discontinued tools never get recommended. Specs come from the same listings, converted to common units: torque in newton-metres, weight in kilograms, and 18V XR treated as 20V MAX because they're the same platform.

2. Specs become comparable scores

Raw specs aren't comparable on their own. 2,000 rpm and 1.6 kg are different units, and a single outlier would dominate a simple average. So within the set of products you're comparing, we turn each spec into a percentile:

score_spec = share of the other products this one beats on that spec

For specs where less is better, such as weight or charge time, the scale is flipped. A drill at the 90th percentile for torque beats 90% of the drills you're comparing on torque.

3. Your needs weight the specs

Each slider is a need, and each need blends several specs. For drills, for example:

Your slider positions (0–5) become the weights. The fit score is the weighted average of the need scores:

fit = Σ w_need × score_need / Σ w

Set "Light & compact" to 5 and "Power" to 1, and a compact brushless driver will beat a heavy hammer drill, even though the hammer drill has better specs on paper. That's the "adapts to you" part: the same catalog yields a different optimum for each set of needs.

4. A regression finds the real value

Price matters, but "cheapest" isn't the same as "best value". A low price on a weak tool is no bargain. To separate the two we fit a multivariate hedonic regression: a model that learns how much each spec adds to the price within a category.

log(price) = β₀ + β₁·torque + β₂·weight + β₃·brushless + … + ε

From this we get a predicted fair price for every product given its specs. A tool selling below its predicted price offers more than you pay for. Its value score is how far below prediction it sits, ranked against the other products.

A few details keep the model honest:

• Log prices, so the model works in percentage terms: a $20 gap means more on a $99 tool than on a $499 one.
• Standardised features with ridge regularisation (λ = 1). Specs like power and torque move together, and with few priced products an unregularised fit would swing wildly. The ridge penalty shrinks those swings.
• Minimum sample size. The value model only runs when enough products in the set have current prices: at least 6, and at least 3 more than the number of specs used. Below that, we hide the value slider rather than show a number we can't back up.
• Visible fit quality. The results header shows how many priced products the model used and its R², so you can judge for yourself.

5. Budget is a soft limit

A hard cut-off would hide the tool that's 5% over budget but far better suited. Instead, scores fall off smoothly once a product goes past your budget: a tool 25% over loses about half its score. Over-budget picks are labelled.

6. Missing data is handled, not hidden

Not every listing states every spec. A missing spec counts as average (50th percentile) rather than best or worst. Each result shows a data coverage indicator, so you can see when a ranking rests on partial information.

What the model doesn't do

• We don't track you. Your needs live in the page while you use it; nothing is sent to us.
• Retailer payments don't affect rankings. Affiliate links, when added, sit next to results. They never feed into the score.
• We don't make up specs. If DEWALT doesn't publish a figure, we show it as missing.

Spotted a wrong spec, or want a need we don't offer? The model improves as the data does.