The CFO slid the printout back across the table and tapped cell D14 with one finger. "Where did this number come from?" The ops director who built the deck did what ops directors do: she said "industry benchmark" and watched the engagement die a little. If you have ever tried to get an embedded engineering engagement approved, you have been in some version of that meeting.
Here is the formula that survives it. Forward deployed engineer ROI equals annualized value created, minus annualized engagement cost, divided by engagement cost. The value comes from three buckets: hours saved, errors prevented, and revenue unblocked. Every number in the model must trace to something you can point at: a timed workflow, an error sample, a named customer. That traceability is what separates a business case from a wish.
The rest of this post builds the three buckets, prices the cost side honestly, and ends with a worked example you can steal cell by cell.
The Forward Deployed Engineer ROI Formula, No Hand-Waving
Write it on one line, because one line is all a skeptical reader gives you: ROI = (annualized value - annualized cost) / annualized cost. A result of 1.0 means the engagement returned its cost in year one and broke even at month twelve. A result of 2.0 means it returned its cost twice.
Two rules before any numbers. First, annualize everything, because engagements are lumpy and CFOs think in years. Second, every input gets a source column in your spreadsheet: "timed on March 12," "sampled 50 invoices," "owner estimate, marked as such." The source column is the difference between a model and a mood.
Bucket 1: Hours Saved (The Easy Math)
Hours saved is hours per week, times loaded hourly cost, times 52. The catch hides inside the word "saved." Count only time that is eliminated or redeployed into work that already exists. If your automation frees ten hours a week and those hours evaporate into longer lunches, you saved nothing the business can feel.
A mini example. A regional distributor has two coordinators spending a combined 12 hours a week copying order confirmations into the TMS. Time it, do not guess: three stopwatch samples across a normal week. At a loaded cost of $65 an hour, that is 12 x 65 x 52, or about $40,600 a year. Copy-paste is expensive. It is just expensive quietly, which is why the hidden cost of manual processes deserves its own line in the deck.
The loaded-cost mistake everyone makes
Loaded cost is salary plus benefits, taxes, equipment, and overhead, typically 1.25 to 1.4 times base salary. Using bare salary understates your case by a third, which sounds admirably conservative until the CFO's analyst "corrects" your spreadsheet with the real multiplier and starts wondering what else you eyeballed. Use loaded cost. Cite HR for the multiplier. Move on.
Bucket 2: Errors and Rework Prevented
The formula: error rate, times cost per error, times volume. A 3% error rate on 200 invoices a week, at $45 per error in rework, credits, and apology emails, comes to roughly $14,000 a year. Errors have a price list. Somebody just has to write it down.
One obvious problem: nobody tracks errors, because tracking errors feels like documenting your own crimes. So sample instead. Pull 50 recent cases by hand, mark the ones that needed rework, and extrapolate with the word "sampled" attached. A hand-counted 6% beats a guessed 2% every time, because "we checked 50 invoices" is a sentence you can say with a straight face in the meeting.
Bucket 3: Revenue Unblocked (Handle With Care)
This bucket is real and radioactive. Faster customer onboarding, bids you can now make because turnaround dropped, churn you prevented because the dashboard caught the problem first. Real money, all of it, and also the bucket where business cases go to get laughed at.
The honesty rule: claim only revenue a skeptic would accept, and park the rest in a row labeled "upside, not in ROI." Onboarding that drops from two weeks to three days has defensible value if sales can name the deals that stalled on it. "Improved morale" does not. One hard revenue number with a name attached outperforms five soft ones, and it makes the soft ones believable as a bonus.
The Cost Side, All-In
Costs get the same rigor, because understating cost is how year two becomes an awkward conversation. Include the engagement fees, obviously. Then add your internal time: the champion's hours, the subject-matter experts tied up in discovery, the people who will own the system afterward. Add tooling and infrastructure, usually the small line. Finally add the maintenance tail, the retainer or internal hours that keep the system alive after the engagement ends. For realistic ranges on the fee side, the FDE cost breakdown covers pilots, projects, and retainers with real numbers.
A Worked Example You Can Steal
One 40-person distributor, one quarter-long engagement. Hours: 18 hours a week redeployed across dispatch and billing, times $60 loaded, times 52, call it $56k a year. Errors: the sampled invoice math from above, $14k. Revenue: onboarding turnaround cut from two weeks to three days, and sales names two stalled deals worth $85k a year in margin that now close. Total annualized value: $155k.
Cost, all-in for year one: $52k in fees, $5k of internal time, $3k of tooling and maintenance, so $60k. ROI = (155 - 60) / 60, about 1.6. The value line runs roughly $13k a month against a $60k cost that lands mostly in the first quarter, which puts the breakeven point near month five. The one-liner for the deck: "Year one returns the investment about 1.6 times, with payback around May."
Then ask the question that wins the room: what if we are half right? Halve the value, keep the cost, and the engagement still returns about $78k against $60k, breakeven inside the year. A case that survives being half wrong does not need defending. It needs a signature, and selling the engagement to your CFO is the art of delivering exactly this math without the theatrical spreadsheet.