Digital Signage ROI: How to Calculate It Honestly

Almost every digital signage ROI figure you will read was written by someone selling digital signage: a percentage with no denominator, no control group, no mention of running costs, and a footnote pointing at a study nobody can open.

This guide gives you the arithmetic and leaves the inputs blank, because the inputs are yours. There is no uplift percentage anywhere in it — we do not have a defensible one, and neither does anyone quoting you one.

What digital signage ROI actually means

Return on investment is a ratio, and it is only as good as its two halves. Most signage business cases are wrong on both. The cost half is understated because it counts hardware and software and stops. The benefit half is overstated because it treats a hoped-for sales uplift as a booked saving.

A defensible model has three properties. Every cost line is something you could produce an invoice for. Every benefit line is either measurable from data you already hold, or clearly flagged as a hypothesis you intend to test. And the model still works when you delete the hypothesis.

The cost side: what a screen really costs

Six components. Two appear on every quote you receive. Four do not.

Displays, players and mounts. The display is the largest line and the widest range, because the specification is set by where the screen goes rather than by what it shows: a panel behind glass in direct sun needs several times the brightness of one above a counter. Players are easier — an Android TV box or a Fire TV stick can be replaced locally on the day it fails, which matters more to a cost model than the unit price; the players page covers the trade-offs. Price mounts per screen, not per site.

Installation. The line that most often doubles between business case and invoice. Get quotes, not estimates, for a socket at each screen position, a wired network drop, mounting and making good, and — the largest hidden multiplier on a multi-site rollout — work that must happen outside trading hours.

Software. The line people focus on and almost never the one that decides the model. Real numbers rather than a range: Qmanja Signage is free for one screen with 200 MB of storage, free forever and with no credit card; Pro is $6.99 per screen per month, or $4.99 per screen per month billed yearly; twenty screens or more is quoted as Business pricing, so at that size ask for the number. Check that a headline rate is the whole rate — per-feature pricing, where scheduling or layouts or reporting are separate modules, makes two quotes uncomparable. The buyer's guide covers those traps.

Content production. A screen is a distribution channel; it does not produce anything to distribute. There is an up-front piece — templates, brand assets, the first layouts — that you can bound like a project, and a recurring piece that runs as long as the screens do. If an agency makes it, you have an invoice; if you make it internally, it is staff time.

The staff time nobody budgets for

The line that turns positive business cases negative, and it is missing from essentially every vendor model. Count the hours at a loaded rate — salary plus employer costs, not the wage — for each of:

• Making and publishing content, including the approval cycle, which at multi-site organisations often costs more hours than the design.
• Commissioning and pairing. Small per screen — the player shows a six-character code, you enter it in the dashboard and the screen claims itself within seconds — but not zero across forty screens, and it comes with travel.
• Noticing and fixing. Someone has to see that a screen is blank and either fix it remotely or go there. Offline alerts reduce this; they do not remove it.
• Onboarding and governance. Every new manager who changes their own promotion, plus somebody who owns who may publish what.

If nobody can name the person who will do these things, that is a finding rather than a gap in the spreadsheet. Deployments with unassigned content ownership stop being updated within months, at which point the benefit goes to zero while the cost carries on unchanged.

Three smaller recurring lines complete the picture: electricity, which on a modest estate frequently exceeds the software subscription and which you can cut by scheduling screens to sleep outside trading hours; a replacement sinking fund; and a second round of capital if your horizon outlasts the hardware.

The benefit side: three things signage can actually move

Be ruthless. A screen can influence a small number of measurable quantities. Everything else on a benefits slide restates one of these three, or is not a number at all.

Lever 1: revenue per transaction

The largest potential benefit, the least reliable, and the only one you cannot read off an existing invoice. Note the metric: per transaction, not total revenue. A screen inside your premises does not bring people through the door — it acts on people already there and already buying. So what it can move is average transaction value, attachment rate, or units of a promoted category. Modelling signage as a driver of footfall is a category error.

Two conditions must hold before the lever exists at all. A transaction has to happen within sight of the screen, at a moment when the customer can still change what they are buying. And the content has to change — a screen showing a fixed image of your old printed menu has converted a print cost into an electricity cost. The mechanisms are covered in how digital menu boards increase restaurant sales. Enter this lever as cents on the average ticket rather than as a percentage.

Lever 2: labour time on manual updates

Less exciting and more dependable, because both halves come from your own operation rather than a customer's behaviour. Time one content change end to end — designing, printing, distributing, installing at each site, confirming it happened — then multiply by frequency and site count. The result is usually larger than anyone's estimate, because the walking and the confirming are invisible to the person doing them. Price the replacement honestly too: publishing to a screen group is fast, but someone still designs the change and checks it landed. The benefit is the difference, not the whole of the first number.

One discipline separates a real saving from a fictional one. Are the freed hours cash-releasing — do they reduce a rota, avoid overtime, or let you leave a vacancy unfilled? If so, count them at the loaded rate. If they simply give a manager forty-five minutes of slack, that is a genuine operational improvement and it is not money. Keep it out of the payback calculation.

Lever 3: print and replacement cost avoided

The easiest lever to verify, because the evidence is a stack of invoices you already hold. Pull twelve months of spend on posters, menu inserts, shelf strips, window vinyls and the couriers attached to them, including the reprints caused by a price change or a supply problem — the portion people remember but do not count. Total the reprints separately from the planned runs; the split between them is the number that decides whether this lever is worth anything to you.

Then be honest about retention. Screens do not replace everything: allergen sheets, statutory notices, shelf-edge price labels and anything a customer carries away stay printed. Applying a retention percentage to historic print spend, rather than assuming it goes to zero, is the difference between a credible model and a hopeful one.

What is not a benefit, however much you want it to be

These belong in the decision, not in the arithmetic, because you cannot put a defensible number on them: looking modern; brand consistency, whose cash value is already partly captured in the labour and print lines; internal staff communication, almost never a cash saving; and compliance or wayfinding, often the actual justification and better argued on its own terms. Keep a second column headed "not counted" — reviewers trust a model more when they can see what you left out. Campus deployments are the clearest case, because there is usually no transaction to move at all, which the campus deployment guide covers.

A fully worked example

Everything below is illustrative. The inputs are constructed to demonstrate the arithmetic, they are not observed results from any customer, and every one should be replaced with your own figure before the model means anything.

The illustrative inputs

Assume a three-site food business with two screens per site, six screens in total. Assume each site takes 200 transactions a day and trades 350 days a year — 70,000 per site, 210,000 transactions a year across the estate — at an average transaction value of $12. Assume a loaded staff cost of $25 an hour and electricity at $0.18 per kWh. Substitute your own numbers, particularly the transaction count, which drives the entire revenue lever.

Cost side

Note that the software subscription is the smallest recurring line — smaller than the electricity to run the same six screens — and that the up-front total is dominated by hardware and installation. Payback is far more sensitive to your installation quote than to anything a software vendor does.

Benefit side

Assume two content changes a month per site, each taking 45 minutes end to end: 4.5 hours a month across three sites, or 54 hours a year. Assume the replacement is one central change published to all six screens, taking 20 minutes, twice a month — 8 hours a year. The saving is 46 hours, or $1,150 a year, counted only because we have assumed it is cash-releasing. If it is not, the honest figure is zero.

Assume print of $60 per site per change ($4,320 a year) plus two full menu reprints a year at $400 per site ($2,400), giving $6,720. Assume 20% of print is retained, so avoided print is $5,376 a year.

For the revenue lever, assume — as a hypothesis to be tested, not a forecast — $0.10 on the average transaction, about eight tenths of one percent of a $12 ticket. Across 210,000 transactions that is $21,000 a year. It is a placeholder, not evidence, and it exists so we can show what happens when it is removed. Do not also enter content upkeep as a cost: the 8 residual hours are already netted off the labour benefit, and double-counting them is the commonest error in these models.

Three answers from one model

One model, three answers spanning a factor of six, and the only thing that changed is which assumptions you allowed yourself. Row C is the brochure number, driven entirely by the one input nobody has evidence for. Build the case on row A or B, and treat row C as upside to be measured.

The payback and break-even formulas

Payback period, in months = total up-front cost divided by ((annual benefit minus annual running cost) divided by 12)

If the denominator is zero or negative there is no payback period, and no horizon will change that. Say so plainly rather than extending the horizon until the number goes positive, which is how these models are quietly rescued.

Return over N years = ((total benefit over N years minus total cost over N years) divided by total cost over N years), multiplied by 100

Three years is a sensible default horizon; be explicit about hardware refresh beyond it. Discounting future cash flows is technically correct and, at this scale, mostly adds another place to hide an assumption.

Break-even uplift per transaction = (annualised total cost minus annual non-revenue benefit) divided by annual transaction count

This is the one to put in front of a decision-maker. If the result is negative, the deployment pays for itself on labour and print alone and any sales effect is upside. Run it per site as well as per estate; an average conceals the location where the maths never works.

How to actually measure the lift

Write down, before the screens go up, what you expect to move, by how much, over what period, and what result would make you conclude it did not work. A metric chosen afterwards is a metric chosen because it moved. Choose something the screen can plausibly influence and that you already collect: attachment rate, units of a promoted category, average transaction value — not total store revenue, which is dominated by footfall and weather.

Control sites and matched pairs

If you have more than one site, you have the tool that makes this tractable. Fit screens at some sites, keep others on the existing process, and compare the change in each group over the same calendar period. Seasonality, weather, a national promotion and a price rise then hit both arms and cancel out.

Match on things that drive your metric: format, trading pattern, transaction volume, customer mix, staffing stability. Assign sites to arms before looking at their recent performance, and do not swap one out later because it looks awkward — that decision ruins more signage measurements than any other. Run at least eight weeks after discarding the first two, which contain a novelty effect and your own teething problems.

With one site, alternate content weekly in an A-B-A-B pattern over eight or more weeks, so trend and seasonality spread across both arms rather than confounding them. Decide the number of weeks in advance and run all of them; stopping early on a favourable reading is how people talk themselves into results that are not there. The same logic applies zone by zone in a larger store, which the retail playbook covers.

Before-and-after windows

Weaker than a control group, and sometimes all you have. Make it as strong as you can.

• Equal-length windows with the same day-of-week composition, so a four-Saturday month is not compared against a five-Saturday one.
• Compare year-on-year where the data exists, which controls for season in a way month-on-month cannot.
• Exclude public holidays, local events, and any period containing a price change or an unusual promotion.
• Log everything else that changed — a new supplier, a refit, a staffing change, a competitor opening. You will not remember in three months.

Why attribution is genuinely hard

You changed two things at once. Screens go live carrying new content, so any effect combines medium and message. Isolating the medium means running the same content in print at a control site, which almost nobody does. Be honest that what you measured was "screens plus a content refresh", because that is what you bought.

Promotion cannibalises. The promoted item rises; another may fall, so measuring only the promoted line always looks encouraging. Measure the category and the basket, and look at gross margin rather than units — pushing customers towards a heavily promoted low-margin item is a way to sell more and earn less.

Staff behave differently when measured, so a trial flatters what you get once attention moves on. Re-measure a quarter later.

You need to know what actually played. Scheduled content and displayed content are not the same thing, and a screen that was blank for a fortnight silently destroys any comparison built on top of it. [SCREENSHOT: proof-of-play report showing which items played on which screen and when] The decision it supports is simple — exclude any site-week where the screens did not run as scheduled. The feature overview covers what is recorded.

When digital signage will not pay back

A guide that only explains when something works is a brochure. These are the situations where the arithmetic does not close.

Nobody owns the content. The strongest single predictor of a negative return. If the answer to "who changes this in month seven" is a shrug, the benefit decays to nothing while the cost continues.

There is no transaction near the screen. A screen in a corridor or a break room may be worthwhile, but it cannot move revenue per transaction because there is no transaction. Its case has to rest on labour and print, or on a non-financial argument made honestly.

Print and update costs are already near zero. A single site that changes its poster twice a year and prints it in the back office is not spending money you can save. Two of the three levers are gone.

Traffic is too low. At a few dozen transactions a day, even a generous per-transaction assumption produces a number smaller than the electricity bill. Run the break-even uplift for your quietest site before including it in a rollout; the estate average will hide it.

The screen is wrong for its position, or its content never changes. A consumer panel in a sunlit window is unreadable for most of the day, and a screen showing a fixed image of what was on the wall is an expensive, power-hungry poster. Both set the benefit to zero and the cost to full price.

You need the answer faster than you can measure it. If the decision requires proof of a sales effect within a quarter, and your volume means a plausible effect would take two quarters to separate from noise, the study you want cannot be run.

Stress-test the model before you sign anything

Spend twenty minutes trying to break the spreadsheet: set the revenue lever to zero, halve the labour saving, add 25% to installation, raise print retention to 40%, and shorten the horizon by a year. Survive all five and you have a decision. Survive none of them without the revenue hypothesis and you have a pilot at best.

Start with the cheapest possible test

The strongest position to be in when you build this model is having already run a screen. Not a demo — a real screen, with your own content, in the position you propose to use, for a month. It converts the largest assumptions into measurements: how long a content change actually takes, whether anyone will keep it current, and whether the position you chose is one customers look at.

That test costs nothing on the free tier — one screen, 200 MB of storage, free forever, no credit card — and runs on hardware you probably already have, since the player works in any modern browser as well as on Android TV, Fire TV and Windows. The free versus paid comparison covers where a free tier stops being enough.