One in every four solar energy systems in Brazil is in a city where no company with a solar energy name is headquartered. That is roughly 1.12 million systems, spread across 4,016 municipalities. Somebody installed all of that, and a good share of it probably came from out of town.
I got to that number by crossing two public datasets that almost nobody looks at together: the distributed generation registry from ANEEL (Brazil’s electricity regulator) and the CNPJ business registry from the Receita Federal (Brazil’s federal tax authority; CNPJ is the national company ID). The result became a free dashboard, Mercado Solar, which shows, municipality by municipality, how many systems exist and how many companies in the sector are based there.
Before the numbers, one warning that holds for the whole article: open data answers a lot, but it does not answer everything. I will be clear about where each limit sits.
What is solar market intelligence and why does an installer need it?
Solar market intelligence is knowing, from data and not from impressions, how much demand exists in your region and how many companies are fighting over that demand. For a small or regional installer, that decides where it is worth opening a sales front, where to advertise and where the fight is already about price alone.
The reason to look at this now is that the market has changed phase. The survey Greener (a Brazilian solar market research firm) runs with installers shows that the average number of quotes per month fell from 42 in the first half of 2022 to 25 in the first half of 2026. In the first half of 2025, 65% of installers pointed to the price war as a problem and 58% pointed to interest rates. In a market like that, running a campaign “across all of Brazil” burns budget. Knowing where the customer is starts to be worth money.
Oak Energia, my renewable energy startup, started in 2018 with that question. I paused Oak in 2023 to focus on Evolutiva (I told that story here). Now it comes back as an Evolutiva product, and the first step was to redraw the market map with what public data allows you to state.
How many solar energy companies are there in Brazil?
There is no official number, because the Receita Federal has no business activity code (CNAE, Brazil’s economic activity classification) that covers solar energy alone. Every count is an approximation, and public sources range from 12,000 to 30,000 companies, depending on what each one counts.
In our cut, we filtered the Receita Federal database (as of 09/12/2026) for active headquarters with a solar-specific name (“fotovoltaica”, “energia solar”, “sistemas solares” and variations) and an activity of electrical installation, electrical material trade or engineering. We removed power plants, SPEs (special purpose entities), consortia and equipment rental firms. That left 6,307 headquarters.
That number is a ceiling, not an exact count. We validated a sample of 40 companies through web searches: among the electrical installation ones (CNAE 4321500), only 7 out of 14 had verifiable online presence as installers. At the same time, a solar company with a neutral name (the owner’s name, for example) falls outside the filter. Use the cut to compare cities against each other, not to add up the market.
Source
What it counts
Number
ANEEL, open distributed generation data
registered solar systems (through 08/31/2026)
4,566,639 systems, 52.7 GW
Receita Federal, CNPJ (our cut)
active headquarters with a solar name and an installation activity
6,307 (ceiling)
Market sources (Greener, Sebrae, ABSOLAR and commercial lists)
installers or companies in the supply chain
from 12,000 to 30,000, with no shared method
Where are the solar systems with no local company?
Of the 5,567 municipalities with solar distributed generation, 4,016 have no company from our cut headquartered there, and together they add up to roughly 1.12 million systems (24.6% of the total). In 2025, 26.7% of all new distributed generation registrations happened in those cities.
This is not a spreadsheet curiosity. Among the 903 municipalities with a thousand registered systems or more, 154 have no company from the cut. There, the customer is served by a company from a neighboring city, a company with a neutral name or a network salesperson. For a regional installer, each of those cities is a market that has already bought solar energy and where nobody local is competing on Google.
The honest caveat: “no company from the cut” does not mean “no company”. It means no headquarters with a solar name and an installation activity. Before deciding on expansion, open the city in the dashboard and check on Google Maps who shows up.
The states with the most installed distributed generation capacity are São Paulo (7.0 GW), Paraná (6.9 GW) and Minas Gerais (6.4 GW). But capacity by state says little to someone serving a 100 km radius. The commercial decision happens at the municipal level.
Is the solar energy market still opening companies?
It is, but far fewer: companies in the cut opened per year fell from 796 in 2021 to 477 in 2025, while the country kept registering hundreds of thousands of systems a year. The company-opening timeline is the one data point no public distributed generation dashboard shows next to demand.
Year opened
Companies from the cut active today
2019
628
2020
625
2021
796
2022
757
2023
510
2024
554
2025
477
Two caveats change the reading. First, the Receita Federal only shows who is still active. A company opened in 2021 had more time to shut down than one opened in 2025. So the real drop in openings is probably larger than the table shows, not smaller. Second, the sector’s mortality is high: Greener reports that only 10% of installers founded before 2016 are still active.
The practical reading: after 2022, with Lei 14.300 (Brazil’s distributed generation law) and the end of part of the incentive, systems connected per year fell from more than 796,000 (2022) to more than 625,000 (2023), according to ANEEL. Fewer people opened companies, many shut down, and those who stayed are competing for a customer who is more sensitive to price and to credit. Financed sales fell from 57% (2021) to 33% (first half of 2026), according to Greener.
What can ANEEL’s data not tell you?
ANEEL reports the holder of each system (the owner of the electricity bill), not the company that installed it. So no installer ranking comes out of the open data, and be suspicious of anyone promising one.
Other limits the dashboard makes explicit:
Registration year is not connection year. We use the project’s registry update date as an approximation. It works to see a trend, not to date a specific job.
Systems per company is a rough reference. Dividing a city’s systems by its local companies mixes jobs done by out-of-town companies with a cut that is a ceiling.
Actual generation in kWh does not appear. The database carries installed capacity, not how much each system generates.
The last year is partial. 2026 runs through August/September.
I write these limits down because I have seen plenty of market presentations sell certainty where there is only approximation. A number with a caveat is worth more than a round number that falls apart at the first question.
How do you use the Mercado Solar dashboard in your city?
Open the Mercado Solar dashboard, pick your city and see the registered systems, the capacity and how many companies from the cut are based there, by period. It is free and asks for no sign-up to browse.
What you can do today:
Filter by state and by period and watch the indicators change: companies opened, registered systems, capacity in MW and systems per company.
See the timeline of companies opened per year side by side with distributed generation registered per year.
Load the map by municipality, with three layers: companies, capacity and systems per company.
Check the ranking of the 12 municipalities with the most systems in the period, plus the table by state.
Type in your company’s CNPJ and see its public profile (size and years of activity), if it is in our cut.
If you want to go further, there is the diagnostic. You enter the range of systems you installed over the last 12 months and your biggest problem today (too few quotes, visits without a sale, customer service, price, after-sales or credit). The dashboard compares your range with the city’s reference, suggests a first step for the problem you flagged, shows up to 3 companies of similar size and activity in the same city, and hands you a kit of social posts with the municipality’s numbers, ready to copy.
The diagnostic is only released with your explicit consent for the use of the data, which serves only to generate the requested diagnostic. The similar companies come from public Receita Federal data and are not “direct competitors”. It is a profile comparison, not a ranking.
One use that tends to go unnoticed: after-sales. More than 1.6 million systems in the country were registered in 2022 or earlier, and the dashboard shows that number by city. Greener recorded that 85% of installers have already been approached by customers of other installers, and 53% already offer operation and maintenance (O&M). An old installed base is a customer for cleaning and inspection.
How can Evolutiva help people working in solar energy?
With data, technology and consulting, without trying to sell one more solar energy CRM. This market already has quoting and customer service tools, and good ones. What most companies lack is knowing where the demand is and having a digital operation that turns that demand into a conversation.
If you run an installer or integrator business
Start with the free diagnostic. In Mercado Solar you open your city, see the installed systems and who operates there, and get a diagnostic of your company.
If it makes sense, we build it together: a website and customer acquisition in the right cities, paid traffic measured from the click to the quote, and custom tools built on the public data for your region.
Before Evolutiva, I spent five years at Oak Energia, my renewable energy startup. That background now turns into two kinds of help:
Renewable energy consulting: working out which source and which model make sense for your consumption, comparing proposals and choosing with confidence who will install it.
Custom technology: simulators, dashboards and systems to track generation and return on investment, connected to the data your company already has.
How many solar energy companies are there in Brazil?
There is no official number, because there is no CNAE code specific to solar energy. Public estimates range from 12,000 to 30,000. In our cut of the Receita Federal data (solar name and installation activity), there are 6,307 active headquarters, a number that should be treated as a ceiling.
Does ANEEL’s data show which company installed each system?
No. ANEEL publishes the system’s holder, not the installer. That is why you cannot build an installer ranking from open data. To compare your company with the city, you need the information only you have, such as how many systems you installed.
How do I see how many solar energy systems there are in my city?
Open the Mercado Solar dashboard and pick the city under “My city”. It shows the solar distributed generation systems registered with ANEEL, the capacity in MW and the companies from the cut headquartered there, with a period filter.
Is the solar energy market still growing in 2026?
It is growing in installed base, but the fight for customers got harder. Brazil passed 4.5 million systems, but the average number of quotes per installer fell from 42 a month (first half of 2022) to 25 (first half of 2026), according to Greener, and company openings in the sector have fallen since 2021.
Is the Mercado Solar dashboard free? What happens to my data?
It is free, and browsing the dashboard requires no sign-up. The diagnostic asks for email, municipality and explicit consent. The data serves only to generate the requested diagnostic, and the public dashboard shows only aggregate numbers by municipality and state, never personal data.
Why do some cities with many systems have no company in the dashboard?
Because the cut only finds headquarters with a solar energy name and an installation activity based in the city. Whoever serves that market may be a company from a neighboring municipality, a branch office or a company with a neutral name. For a regional installer, those cities are usually an expansion opportunity.
If you run an installer or integrator business and want to know where your next customer is, open your city in Mercado Solar. And if your city’s number surprises you, send me a message. I want to understand what the data does not show.
Median first response time is the sales metric almost nobody tracks — and the one that degrades fastest without warning. When Harvard researchers audited 2,241 US companies by sending test inquiries, the average first response took 42 hours and 23% never responded at all. Not by decision: by drift. Fixing it costs zero in media spend, because these are inquiries the company already paid to receive.
The wrong diagnosis almost every company makes
The conversation repeats with almost comic regularity. The company invests in ads, the ads generate inquiries, the inquiries land on WhatsApp — and three months later someone says, in a meeting, that “the leads are bad.” From there the discussion turns to targeting, to creative, to keywords, to switching agencies. Everyone looks at the top of the funnel.
Almost nobody looks at the gap between the customer’s message and the company’s first reply.
This isn’t negligence. It happens because that gap doesn’t show up anywhere. It isn’t in the Google Ads dashboard, it isn’t in Meta Ads Manager, it isn’t in GA4, and it isn’t in the monthly report the agency sends. The media dashboard ends at the click; the CRM, when there is one, starts at the qualified lead. Response time lives exactly in the space between the two — and no in-between space has an owner.
The result is that the company ends up with a comfortable, wrong explanation. “The leads are bad” is comfortable because it outsources the blame to the media vendor and suggests a solution you can buy: more budget, another campaign, another vendor. “We take half an hour to respond” is uncomfortable because the solution can’t be bought — it has to be organized. And organizing is harder than spending.
This article is about that gap: what the available research has already shown about it, why almost every number circulating on the subject is bad, where exactly the time is lost in an operation, and how to measure your own — with no new tool and no IT project.
What is median first response time, and why not the average?
It’s the interval between the customer’s first message and the first human reply, for the case in the middle of the line. Half the inquiries waited less than that; half waited more. The average, by contrast, is wrecked by a handful of conversations answered days later — and ends up describing an operation that doesn’t exist.
The difference is practical, not statistical. Picture ten inquiries: nine answered in 5 minutes and one answered three days later, because someone found the lost message on Monday. The median of that set is 5 minutes — which is, in fact, the typical customer’s experience. The average comes out above seven hours. If you use the average to manage the team, you’ll conclude the operation is slow when it isn’t; if you use the average to defend yourself, you’ll hide one serious case inside a pile of good ones.
In an operation with intermittent coverage — the norm at any company without a dedicated contact center — average and median frequently tell opposite stories. Hence the rule: the median describes the operation, and the tail describes the risk. Both matter, and neither replaces the other.
There’s also a third measure that almost never shows up and is the most revealing of all: the share of inquiries that simply never get a reply. That isn’t slowness. It’s total loss, and it doesn’t enter any average, because there’s no interval to measure. It vanishes from the report precisely because it’s the worst possible case — and, as you’ll see below, it’s also the number that shocks people most the first time they measure.
How long do you have before you lose the inquiry?
Less than almost anyone assumes. The most solid reference on the subject comes from an MIT Sloan study led by James Oldroyd: waiting 30 minutes instead of 5 minutes to call a web-generated lead cuts the odds of qualifying that lead by a factor of 21.
One precision point is worth making, because most marketing articles get it wrong, and it’s useful for telling apart who read the source from who copied it. The “5 minutes, 21 times” number comes from the MIT Sloan Lead Response Management study, from 2007 — not from the Harvard Business Review article usually cited alongside it. The HBR article is a different piece of work, from March 2011, by James Oldroyd, Kristina McElheran and David Elkington, titled The Short Life of Online Sales Leads. It did something else: it audited 2,241 US companies by sending test inquiries and measuring how long each one took to respond.
What the HBR audit found is the data point that matters most to anyone reading this article:
HBR audit (2011) — 2,241 companies
Result
Average first response time
42 hours
Responded within 1 hour
37%
Responded between 1 and 24 hours
16%
Took more than 24 hours
24%
Never responded
23%
Read that last line again. Nearly a quarter of 2,241 companies — real companies, with sales teams, spending money to generate those inquiries — left a sales inquiry with no reply at all. They didn’t reply slowly. They didn’t reply.
Two honest caveats, before anyone uses them against the argument. First: these studies measured form inquiries at US companies, mostly B2B, and the response channel was the phone. WhatsApp is a different channel, a different country and a different time expectation — the direction of the finding transfers, the exact magnitude does not. Second: these are studies more than a decade old. Nothing suggests customers have grown more patient since then; the reasonable assumption is the opposite, and it works in the argument’s favor, not against it. Use them as an order of magnitude, never as a contractual target.
Why is almost every number circulating on this bad?
Because most of them have no traceable primary source — they’re statistics recycled between articles that cite each other until the origin disappears. Search the subject and you’ll find dozens of round, attractive numbers. It’s worth knowing how to tell which ones are worth anything.
The pattern is always similar. A number appears in a software vendor’s sales material. A blog cites the material. Another blog cites the blog. By the third hop, the attribution has become “according to studies” or the name of a well-known company that may never have published it. The number circulates for years, gets repeated in presentations, makes it into a sales proposal — and nobody can reach the original document, because it doesn’t exist or was never public.
Four questions separate usable data from a decorative number:
Who measured it, by name? Identifiable author, identifiable institution. “According to research” is not a source.
Measured what, with what sample? “2,241 US companies, test inquiries submitted through forms” is verifiable. “Leading companies” is not.
Who paid for it? A study published by whoever sells the solution to the problem being measured isn’t useless — but it calls for a discount, and the discount has to be declared.
How old is it, and does the context still hold? A 2011 data point about the phone applied to WhatsApp in 2026 needs an explicit caveat. Without it, it’s contraband.
That’s why this article leans on two studies only, and returns to them insistently: they’re identifiable, they have authors, they have a described sample and any reader can check them. Two numbers you can verify are worth more than twenty you have to accept.
And there’s a practical consequence, which is the point of everything that follows: no external benchmark, however good, tells you what your number is. It tells you the problem exists and is large. Only your own measurement tells you whether you have the problem, and how big it is. That’s why the most useful part of this text isn’t the part citing studies — it’s the part that teaches you how to measure.
Why does slowness set in without anyone deciding?
Because nobody decides to get worse. Operations rarely get slow through an identifiable decision — a headcount cut, a policy change, a tool swap. They get slow by accumulation: dozens of small, reasonable choices, none of which had “respond more slowly” as a stated consequence.
That’s the signature of an entire class of operational problems, and it’s worth recognizing. When a number gets worse by decision, someone can be questioned and the number comes back. When it gets worse by drift, there’s no moment where someone made a mistake — and so there’s no moment where someone fixes it. Drift is only interrupted by continuous measurement, never by attention.
Add to that the effect of growth. When inquiry volume rises, the people answering feel busier, which is true, and feel they’re delivering more, which is also true in absolute terms. What the feeling doesn’t capture is that the queue grew faster than the capacity to answer it. Higher volume with the same effort shows up as waiting — and waiting makes no noise: the customer who gives up doesn’t complain, they just don’t come back.
There’s also a trap specific to answering by message. The conversation doesn’t disappear. It sits there, marked unread, available to be answered at any time. That creates the illusion that nothing was lost — the message is still there, after all. But the commercial value of a conversation is not conserved over time. Replying three hours later is, for most intents, the same as not replying — with the added cost of occupying someone for ten minutes to find that out.
The four points where time is lost
“We’re slow to respond” is a symptom, not a diagnosis. In practice, the delay accumulates at four distinct points, with different causes and different fixes. Finding out which one is yours avoids spending money in the wrong place — which is usually hiring people.
1. The message arrives and nobody is notified
The inquiry lands on a shared device, in an open tab nobody is watching, or in an inbox someone checks “every now and then.” The time here isn’t service time: it’s discovery time. This is the most common point and the cheapest to fix, because the solution is configuration, not hiring. It’s also the most deceptive: the team is available, the customer is waiting, and the two never meet.
2. Everyone sees it and nobody owns it
The notification reaches five people and none of them is responsible for that specific inquiry. Each one reasonably assumes someone else will take it. It’s the bystander effect applied to customer service — and it gets worse as the team grows, which makes the instinct to “hire more people” not just expensive but sometimes counterproductive.
3. Everything goes into the same queue
The person who needs the service now waits behind the person researching for six months from now, because nothing separates them. Without triage, the queue is served first come, first served — the ordering least correlated with value there is. Here the median may even look fine: the problem is that it hides the fact that the most valuable inquiries are in the tail.
4. Demand happens outside the schedule
Some share of inquiries arrives when nobody is scheduled — and “nobody is scheduled” usually means “nobody is scheduled because we set the hours by convention, not by the curve.” That’s the subject of the next section.
All four produce the same symptom in the report and require completely different fixes. That’s why measurement has to come before the solution: without knowing which of the four the time accumulates in, any intervention is a bet.
How your own data usually points to which one is yours:
What the data shows
Likely point of loss
Fix
Similar delay at any hour of the day, including with the team present
1 — nobody is notified
Active notification
Delay got worse as the team grew
2 — nobody owns it
Routing with a named owner
Good median, but valuable inquiries in the tail
3 — single queue
Triage by urgency
Delay concentrated in specific time windows
4 — outside the schedule
Schedule built on the real curve
Inquiries with no reply at all, scattered
1 and 2 combined
Notification + owner, in that order
What hours should your team be available?
The hours when demand exists — which rarely match standard business hours. Staffing 8am to 6pm Monday through Friday is a convention inherited from on-site work, not a reading of how people looking for your company behave.
The real curve usually has two features the convention ignores. It has a concentrated core — a few hours of the day account for a large share of volume — and it has a tail in the early evening, made up of people who spent the day working and can only handle personal matters after hours. Staffing by convention misses at both ends at once: too many people in the early morning and too few exactly when the inquiry arrives from the person with the least free time.
But there’s a trap before you change any schedule, and it deserves a rule of its own: one month of data describes that month, not the operation.
A short window often shows a convincing pattern — a spike on an unexpected day, an hour that looks new. Much of that, when checked against a larger window, turns out to be the effect of a specific campaign, a seasonal swing, a one-off event. It isn’t demand: it’s the trace of something you yourself did that month. Reallocating people based on it means moving the team against the true curve and then having to move them again.
Scheduling decision based on
Risk
Convention (business hours)
Surplus in the early morning, shortage in the evening tail
A one-month window
Mistakes campaign effects for demand patterns
A window of 12 months or more
Low — requires having the data exported and read
Schedules are expensive to change once and more expensive still to change twice. Before touching staffing, confirm the pattern survives a large window.
Why doesn’t this metric exist in your stack?
It’s worth understanding why such a consequential number manages to stay invisible at a company with GA4 configured, active campaigns and a monthly report. It isn’t carelessness: it’s a predictable consequence of how the tools divide the world among themselves.
The media platform knows everything up to the click and nothing after it. It can record that the click went to a message link, but the conversation happens in an app it doesn’t own and that sends nothing back. For the ads dashboard, the click is the end of the story.
GA4 knows everything that happens inside the site and nothing outside it. It records the click on the chat button, if someone instrumented that event — and plenty of people haven’t. From the moment the browser opens the app, GA4 is blind by design.
The CRM starts too late. It begins at the qualified lead, meaning after someone has talked, assessed and decided to create a record. And here’s the most perverse distortion of all: the inquiries lost to slowness disappear from the CRM precisely because they were lost. The company ends up with no record of its own losses, and any funnel analysis built only on the CRM is conditioned on survival — it measures who made it through, never who was left behind.
That leaves the customer service layer, which is where the data actually is — and which almost never has an owner. The service tool usually reports to operations or to sales, not to marketing; the report it produces is about agent productivity, not about acquisition performance. Nobody is in the wrong. The metric lives exactly on the border between two departments and three systems, and a border with no owner is not measured.
The fix is architectural, not a matter of tooling. Three instrumentations, in order of effort:
A click event on the chat channel, with the source. Every message link on the site fires an event identifying the page it came from. It connects media to the start of the conversation, and it’s the bare minimum.
A source marker in the text of the first message. A pre-filled snippet, different by page or campaign, that the customer sends without editing. It’s what crosses the app boundary — and it doesn’t depend on cookies or on tracking consent, because it’s content the customer sends themselves.
Response time reported alongside the media metrics, in the same document and at the same cadence.
The third item looks bureaucratic and is the one that changes behavior most. A metric that shows up in the report leadership reads is a metric someone defends. A metric that lives in a dashboard nobody opens is a metric that drifts for years without anyone seeing it.
What does this cost in money?
You can estimate it with four numbers your company already has, and the math tends to be uncomfortable. What you can’t do is use another company’s numbers — including the studies cited here. They prove the problem exists; only your own measurement says what it costs you.
The four numbers are: inquiries per month, percentage with no reply, close rate on the inquiries you do answer and average ticket. The estimate of lost revenue is the product of the four — inquiries × percentage lost × close rate × ticket.
Three caveats keep this math from turning into fiction. First: applying the close rate of answered inquiries to the unanswered ones is an optimistic approximation, because some share of the unanswered probably had weaker intent. Treat the result as a ceiling, not as confirmed loss. Second: the urgent inquiry has a different ticket and a different close rate from the rest — if you can separate them, do, because that’s where the loss concentrates. Third: the math ignores the reputational effect of not being answered, which is real and can’t be estimated from those four numbers.
Once the math is done, the number that matters isn’t the absolute figure. It’s the comparison: what it costs to recover an inquiry you already have versus what it costs to buy a new one. The inquiry lost to slowness has already been paid for — the media that brought it already left the bank account. Recovering it costs notification setup and automated-reply design, which are one-time costs. Buying a new inquiry costs your campaign’s cost per lead, every month, forever.
That’s why this is almost always the highest-return intervention available in an acquisition operation — and almost always the last one done. It doesn’t look like a marketing initiative: no asset, no campaign, no good-looking slide. It produces more revenue on the same budget, which is what the paid media was trying to buy all along.
How to fix it, in the right order
The order matters more than the measures, and it follows the four points of loss described above — from the cheapest fix to the most expensive.
1. Notification and ownership before hiring
Make sure an incoming inquiry produces a signal, and that the signal has a named recipient. Routing with an owner solves the first two points of loss at once, usually cuts the median without adding anyone to payroll, and is configuration — not a project.
2. An automated reply that actually replies
“Hi! We got your message and will get back to you shortly” isn’t an automated reply — it’s a wait notice dressed up as service. A useful automated reply delivers information: price range, real operating hours, lead time, next step. It fully resolves a slice of the inquiries and shortens the queue left for a human.
On price specifically, the classic fear is worth facing. Publishing a range drives away people who weren’t going to buy at that range — and those people would have consumed team time to reach the same conclusion. Whoever stays arrives with calibrated expectations, which is the best possible condition for closing. If your operation has a queue, filtering before the human isn’t a loss: it’s giving capacity back to the people with real intent.
3. Routing by urgency
Simple triage on the first message — one question, two or three options — separates the queue that tolerates half an hour from the queue that doesn’t tolerate three minutes. Keeping everything in one queue is a business decision, not a technical limitation.
One note on automation, because this is where many projects derail: every automation that talks to customers needs to be born with its scope declared in writing. What it answers on its own, what it must hand off to a human, and what it never does under any circumstances. Without that, the bot meant to save response time becomes the cause of the next problem — and then the company shuts it all down and goes back to square one.
The trap of optimizing the metric instead of the outcome
As soon as first response time becomes a tracked number, a predictable temptation shows up — and it deserves a warning before someone falls for it in good faith.
It’s trivial to improve this metric without improving anything. Just reply quickly with anything: “we got your message, a consultant will assist you.” The number plummets, the chart looks great in the meeting, and the customer still waits exactly as long for what they came for. The company has started measuring the speed of the wave, not the speed of the service.
It’s the classic effect of any indicator that becomes a target: when the measure becomes the goal, it stops being a good measure. And here the distortion is especially easy because the degraded version is indistinguishable from the good one in the report — both show up as “we responded in 2 minutes.”
Two safeguards fix it, and both are cheap:
Measure time to the useful reply, not to the first reply. A useful reply is one that delivers information the customer can act on — a price, a lead time, an availability, a question that moves the conversation forward. Acknowledge receipt all you want; just don’t count that as service.
Always track it alongside an outcome number. Inquiries that turned into real conversations, or into sales. Response time falling with outcomes flat is a sign that the metric is being managed, not the operation.
This doesn’t contradict the recommendation to use an automated reply — the distinction is the content. An automation that delivers a price range and real hours is a useful reply, and it counts. One that only announces someone is coming isn’t, and shouldn’t count. The question that separates the two: after reading this, does the customer know something they didn’t know before?
How to measure this in your operation this week
You don’t need a new tool to get the first number. You need a conversation export and an afternoon. If your operation uses a customer service platform, the export exists; if it uses the regular app, you can do it by hand on a sample — and that’s already enough to decide.
Export at least 12 months. Not three, not one — for the reason given in the section on scheduling. If the export comes in several files, merge them all before counting: “full year” exports often come truncated without saying so, and a truncated slice looks a lot like a complete one.
Group by conversation, not by message, and identify in each one the customer’s first message and the first reply from a human agent — not the automated one.
Calculate three numbers, not one: the median interval, the percentage of conversations with a recorded human reply, and the percentage with no reply at all. The third is usually the most shocking, and it’s the one nobody had.
Break it down by hour and by day of week to find the real demand curve, and compare it with the current schedule.
Read 50 full conversations. This is the part nobody does and the one that pays off most.
On the fifth step, it’s worth insisting. Counting conversations answers how many; reading conversations answers what. Reading is how you discover the customer’s real vocabulary — which is almost never your site’s vocabulary — which friction shows up before all the others, and what kinds of things people ask for that you don’t sell. Demand you never mapped doesn’t appear in a media report by design: the report can only tell you about the campaigns that exist and the keywords you bought. What the customer writes unprompted is the only source that has that information.
Two cautions. Customer service conversations contain personal data — work in aggregate and without identification, and treat the export with the same care you’d give a customer database, because that’s what it is. And record the date of the measurement: the value of this number is in the series, not in the point. Measured once, it’s a curiosity; measured every month, it’s control.
Is a bot replying bad for the customer experience?
It’s worse than a fast human and much better than a human who never shows up. The comparison that matters isn’t between a bot and an ideal agent — it’s between a bot and the real alternative, which for nearly a quarter of the companies HBR audited was no reply at all.
What makes automation bad is almost never the fact that it’s automation. It’s automation that resolves nothing: long menus, generic answers, no exit to a human. A first automated reply that delivers real information and offers a clear route to a person is perceived as efficiency, not as coldness.
The practical rule: the bot is good at what’s predictable and terrible at what’s sensitive. Price, hours, address and lead time are predictable. Negotiation, exceptions and any conversation with emotional weight are sensitive, and the design has to guarantee they reach a human fast — not that they get handled by a script faking empathy.
Frequently asked questions
What’s a good WhatsApp response time?
There’s no universal number, and be suspicious of anyone offering one — for the reasons in the section on source quality. What does exist is a direction: the closer to minutes, the better, and the difference between 5 and 30 minutes was large enough to show up in academic research. Measure your current median and treat the next target as a percentage reduction against it, not as a number copied from another company.
Do I need a 24-hour team?
Almost certainly not. You need coverage on your real demand curve. Before discussing an overnight shift, look at the early-evening tail — that’s usually where the unanswered volume sits, and it’s far cheaper to cover.
How do I measure this if my WhatsApp is the regular app?
By hand, on a sample. Take 50 conversations from the last few months, note the time of the customer’s first message and of your first reply, calculate the median. It’s less precise than a full export and it’s enough to reveal an order-of-magnitude problem — which is the kind of problem most operations have.
Doesn’t publishing prices on the site drive customers away?
It drives away people who weren’t going to buy at that range — who would have consumed team time to reach the same conclusion. And it qualifies whoever stays. If your operation has a waiting queue in customer service, filtering before the human isn’t a loss: it’s what gives capacity back to the people with real intent.
Does the customer who disappears come back later?
Some do, and it’s precisely the least urgent group. The urgent inquiry doesn’t come back — they resolve it with whoever answers first. Since that’s usually the inquiry with the highest unit value, the loss from slowness concentrates disproportionately in the segment the company can least afford to lose.
Does this apply to B2B or only to local services?
The studies cited here measured exactly US B2B companies responding to forms. The mechanics are the same in both cases; what changes between contexts is the tolerable window, not the principle that it exists and that almost nobody measures it.
My company gets few inquiries per month. Does this still apply?
It applies more, not less. At low volume, each lost inquiry is a bigger slice of the month’s revenue. The upside is that the diagnosis gets simpler: you can read every conversation instead of sampling, and the fix is almost always notification and named responsibility, not automation. Low volume makes the problem cheaper to solve and more expensive to ignore.
What does fixing it cost?
The first front — notification and routing with a named owner — is usually configuration, not hiring. The second and third involve automation and carry project costs. The relevant point is that none of the three involves increasing media spend: you’re recovering inquiries you’ve already paid for.
Why has my agency never shown me this?
Because the data isn’t in the media dashboard. It’s in the customer service tool, which usually belongs to another department, and it requires exporting and reading content instead of pulling a ready-made report. In most cases it isn’t bad faith — it’s a border between systems that nobody was assigned to cross.
What to do with this
If you take only one thing from this article, let it be this: before adding one more dollar of media spend, measure the median of your first response time and the percentage of inquiries that never got a reply. Both numbers take an afternoon to obtain and they change the entire conversation about campaign performance.
Investing in media with a customer service operation in drift is buying more inquiries for a queue that already can’t handle the current ones. The extra return doesn’t show up, the conclusion becomes “the leads are bad,” and the cycle starts over — with more budget.
The good news is that problems like this are among the cheapest there are to solve. The customer already wants to talk to you. They already reached out. They’re already waiting. All that’s missing is someone getting there first.
Sources
Oldroyd, J. B.; McElheran, K.; Elkington, D. The Short Life of Online Sales Leads. Harvard Business Review, March 2011 — an audit of 2,241 US companies by sending test inquiries. · Lead Response Management Study, MIT Sloan School of Management (Oldroyd, 2007) — the origin of the 5-minute finding and the factor of 21. Both studies are identifiable, have named authors and a described sample, and can be checked by the reader — the criterion Evolutiva applies to any number it publishes.
If you’d rather not measure it alone. This article ends by asking for two numbers: the median of your first response time and the percentage of inquiries that never got a reply. Producing both is exactly what Evolutiva does in the free diagnostic — where your inquiries come from, how long they wait and where they’re lost. You walk away with the numbers in hand, whether you work with us afterwards or not.
We use cookies to improve your experience and analyze traffic, in accordance with our Privacy Policy (LGPD, the Brazilian data protection law). Learn more
