From Whack-a-Mole to Program: Ongoing Shadow IT Visibility
The first shadow IT hunt is exciting. You pull the DNS census and find a file-storage service with two dozen hosts on it. You have a good conversation with the team, migrate them, and feel like you fixed something. The third hunt is less exciting, because the same category of finding is back under a different name. By the tenth, you are the person who keeps turning up at desks with awkward questions. The findings arrive faster than you can talk to anyone. Nothing about the reasons people go around IT has changed. I got to about the eighth before I noticed. That is whack-a-mole, and it is what happens when discovery is treated as a series of incidents rather than a program.
A program is not a bigger tool or a stricter rule. It is a small set of habits that run on a schedule, and a request path that is genuinely easier than going rogue. It includes a handful of honest measurements and an agreement with management about what is allowed. It also has a written boundary about what the whole thing will never do. This guide describes each piece and ends with a checklist you can use to see which ones you already have.
It assumes you have read the earlier guides in this topic, or at least know how a finding is produced and what the conversation afterwards looks like. Those are the moves. This guide is about the game.
In this guide
- Why whack-a-mole fails
- The recurring reviews
- A request path easier than going rogue
- Metrics that matter
- Policy alignment: allowed, blocked, tolerated
- Privacy boundaries, written down
- Block, monitor, or adopt?
- The program checklist
- Where to go next
Why whack-a-mole fails
Handling each finding on its own feels responsive, but it has four built-in failures. First, nothing is learned: each conversation ends, its outcome lives in someone's memory, and the next similar finding starts from scratch. Second, the underlying need is never addressed, so the same category of tool returns under a new name as soon as the old one is removed. Third, the reviewer burns out, because the queue only ever grows and every item is an interruption. Fourth, and most damaging, the organization learns that IT's role in this is to say no after the fact. That is precisely the belief that creates shadow IT in the first place.
The program fixes each failure with a specific habit. Findings are reviewed on a schedule and recorded in one place. The record feeds a catalog and a request path that remove the reasons to go around IT. A few metrics show whether the situation is improving. A written policy makes IT's role predictable. None of it is complicated. All of it has to be written down and owned, or it decays back into whack-a-mole within a quarter.
The recurring reviews
Two meetings, or one meeting with two agendas, carry the whole program. Neither should take more than an hour. Meetings grow if you let them, so put the hour in the invitation.
The monthly unknown-service review
This is the review queue from finding unsanctioned SaaS, walked by a small group instead of one admin alone. Invite whoever owns the sanctioned-service list and someone from the business side who knows what teams are actually working on. If you have one, invite whoever handles security or compliance. A workable agenda:
- New unknowns since last month, top of the risk ranking first. Each gets a disposition: sanction, ignore, evaluate, converse, or block. Five minutes per item, no more; anything needing longer becomes an action item.
- Open items from last month. Did the migration happen? Did the host count fall? Did the evaluation finish? Items that are still open after two reviews get escalated or explicitly tolerated with a reason.
- Repeat findings. Anything that came back after a disposition. These are the most informative rows on the whole report, because they say a need was not met.
- The metrics, described below, on one slide or one page.
- List maintenance. Aliases added, ignore entries added, catalog entries changed. Boring and essential.
The quarterly inventory diff
Hardware moves slowly, so the diff from finding unofficial hardware can run weekly for alerting purposes but only needs a human review once a quarter. The agenda is shorter. Review devices seen but not on the books, and devices on the books not seen in months. Check anything previously marked "tolerated, convenience" to see whether it has quietly become a data path. Once a year, walk the floor anyway. Some things never talk on the network and you only find them with your eyes.
A note on the reviewer. Both reviews need a named owner who prepares the report, runs the meeting, and records the dispositions. Programs die when this role is "whoever has time," because nobody ever does. It should be a specific person with the task in their job description, even if the task is an hour a month.
A request path easier than going rogue
Here is the uncomfortable arithmetic. A free trial and a company card take four minutes. If your process for getting a tool approved takes four weeks of silence, people will do the four-minute thing every time. No amount of discovery will change that. The most effective shadow IT control is not detection at all. It is a request path that competes on speed and convenience with going rogue. Two pieces make it work.
The fast evaluation lane
A time-boxed review of a proposed tool that answers a fixed set of questions and returns a decision within a promised number of working days. The questions are the same every time, which is what makes it fast:
- What data will it hold? Public, internal, confidential, or regulated. This single answer decides most cases.
- How do people sign in, and how do they get removed? Can it use your central login? Can you list and disable accounts when someone leaves?
- Can you get the data out? An export function, in a usable format, without a support ticket to the vendor.
- Does it overlap with something you already have? If so, why is the existing tool not meeting the need? Sometimes the answer improves the existing tool rather than adding a new one.
- What does it cost, and who pays? Including the cost of the three other teams who will want it once it is approved.
Publish the promised turnaround and keep it. A lane that promises five days and delivers five days is trusted. One that promises five and delivers thirty is worse than none, because it teaches people that asking is pointless.
Ashgrove Housing had a software request form. It asked for a business case, a data classification, a budget code, and a sponsor's signature. The signature needed a committee that met monthly, so an answer took somewhere past forty working days. A free trial took four minutes and asked for an email address. The first DNS census found nineteen unknown services; the form had received three requests all year. The fast lane replaced it with five questions and a promised five days. Requests rose to a dozen a month, and the unknown count fell for the first time. The form is still available on request.
The published catalog
A page, wherever your staff already look for answers, that lists the approved tools by need, not by product name. "I need to send a large file to a client" leads to the answer; nobody should have to know the product's name to find it. Alongside the approved list, keep a short "asked and declined" list with the reason for each decline. Keep a "being evaluated" list so people can see that their request is moving. The catalog is what makes "sanctioned" a real word. Without it, the sanctioned list lives only in IT's head. Staff cannot be expected to follow a list they cannot see.
Tip: the first time a finding turns into an approved catalog entry, tell the team that found it, and thank them. Word travels. Nothing improves the honesty of your census faster than the news that IT sometimes says yes.
Metrics that matter
A program needs a few numbers to know whether it is working. The right ones are the ones that change when the program does its job. Report them monthly, as trends rather than single values. Resist the urge to set targets before you have a few months of baseline to know what normal looks like.
| Metric | How to compute it | What a good trend looks like |
|---|---|---|
| Unknown destinations | Count of services in the unknown bucket at month end | Falls steeply at first, then settles into a small, stable number |
| Time to sanction | Working days from first-seen (or request) to a decision | Short and predictable; the promised turnaround is met |
| Repeat findings | Services or categories that reappear after a disposition | Near zero; each repeat is investigated as a failed fix |
| Request-lane volume | Requests submitted per month | Rises as people learn the lane exists, then stabilizes |
| Catalog coverage | Share of the most-used external services that appear in the catalog | Rises toward "nearly all of the top of the list" |
| Tolerated items aging | Count of tolerated items past their review date | Zero, or a short list with reasons |
Read the metrics together. A falling unknown count with a rising request-lane volume means the desire paths are moving onto the pavement, which is exactly the goal. A falling unknown count with no requests means people have moved their workarounds somewhere you cannot see, which is a warning, not a win. Repeat findings that climb while time-to-sanction grows mean the evaluation lane has stalled and people have stopped waiting for it. (Nobody will tell you this. The graph will.)
Just as important is what not to measure. Do not track findings per person or per department as a league table. Suppose a department head sees their team's name at the top of a "worst offenders" chart. At that moment, the program has become a surveillance program in everyone's eyes, and cooperation ends. Measure the system, not the people.
Policy alignment: allowed, blocked, tolerated
Every disposition in the monthly review is really a policy decision, and policy decisions need a policy behind them or they become one admin's opinion. Sit down with management and, if you have them, legal or HR, and agree on three lists and the rules for moving between them.
Allowed is the catalog: tools with an owner, a support path, a known data-handling answer, and a way to offboard accounts. Blocked is the short list of things that cannot be used regardless of need, with a stated reason for each. Usually, the reason is the data they would hold or the way they would open the network. Technical blocking at the firewall or resolver belongs only to this list. Every block should point at a catalog entry that meets the same need, or at a written explanation of why no tool can. Tolerated is everything in between: known, recorded, not yet decided, with a review date. Tolerated is not a dumping ground; every item on it has a reason and an expiry.
The policy should also say who decides. A good default: the reviewer can sanction low-risk tools and tolerate anything pending evaluation. Blocking, and any decision about regulated data, goes to management. Write down how a tool moves from tolerated to allowed or blocked. Also write down what happens when an agreement from a conversation is not kept. Escalation then follows a known path rather than a personal one. Keep the whole policy short. A policy nobody reads is a policy nobody follows, and nobody reads page three. The guide on the diplomatic conversation describes the individual escalation cases; the policy is where those cases get their authority.
Privacy boundaries, written down
Everything in this topic depends on one sentence being true: the program looks at services and devices, not at people. That sentence has to be written down, agreed with management and whoever handles legal questions, and shown to anyone who asks. One page is enough, and it should answer these questions:
- What data sources are used? DNS query logs, flow records, TLS server names, switch and DHCP tables. Nothing else without a change to this document.
- At what level is it aggregated? By service and host count, with department grouping where it helps route a conversation. No per-person reports, no browsing histories, no content.
- What categories are excluded entirely? Personal categories — news, shopping, health, personal email — are dropped before any report is produced.
- How long is data kept? Raw logs follow the existing security retention; program summaries are kept as long as the program runs.
- Who sees the reports? The review group, by name or role, and nobody else without a stated reason.
- What is never done? No decrypting traffic for this purpose. No identifying an individual except to find the team for a conversation. No using program data for performance management.
- How are findings handled? Conversation first; escalation by the policy's rules; findings recorded as services and outcomes, not as people.
This document does two jobs. It protects staff from the program drifting into something it was never meant to be. It also protects you. If anyone ever asks what IT was doing with all that network data, you hand them the page. In some places the law requires something like it. Everywhere, it is what allows you to say "we look at services, not people" and be believed.
Block, monitor, or adopt?
The three possible fates of a finding, and the conditions that point to each:
| Action | Choose it when | Do not choose it when | What must accompany it |
|---|---|---|---|
| Block | Regulated or contractually protected data is involved; the finding opens the network (a bridged router); there is no legitimate need and a real hazard | It is the opening move; a team's live work depends on it and no migration exists; the block is really just impatience | A catalog entry meeting the same need, or a written reason none exists; notice to affected teams before, not after |
| Monitor (tolerate) | Risk is low or medium; an evaluation is under way; a migration is agreed and dated; a project would be harmed by an abrupt change | There is no review date; the item has been "temporary" for a year; the data is regulated | A recorded owner, reason, and review date; a check of host count at the next review |
| Adopt | The need is real and unmet; the tool passes the evaluation lane; it is better than or equal to the incumbent; the cost is understood | It duplicates an existing tool that would work if fixed; the data-handling answer is unknown; it cannot offboard accounts | A catalog entry, an owner, central login where possible, and a thank-you to the team that found it |
Two patterns are worth naming because they are so common. The first is the block that should have been an adoption. A genuinely good tool is removed because it was unapproved. Three worse ones appear within a quarter. The second is the tolerate that should have been a block. A consumer router is left in place for months because the conversation was awkward. Then, one day, something arrives through it. Blocking is rare and adoption is frequent in a healthy program, and "tolerated" always has a date. Not "soon." A date.
The program checklist
Use this to see what you have and what is missing. A program with every row filled in is unusual; a program with an owner for every row is what matters.
| Component | What it is | Cadence | Owner |
|---|---|---|---|
| Sanctioned-service list with aliases | The definition of "known," in machine-matchable form | Updated at every review | Reviewer |
| Ignore list | Infrastructure noise excluded from the diff | Updated at every review | Reviewer |
| Nightly collect-group-diff job | DNS census, flow rollup, and diff producing the review queue | Nightly, automated | Admin |
| Monthly unknown-service review | Dispositions for new, open, and repeat findings | Monthly | Reviewer plus group |
| Weekly hardware diff, quarterly review | Network versus register, with a human look each quarter | Weekly / quarterly | Admin / reviewer |
| Fast evaluation lane | Fixed questions, promised turnaround | On request | Reviewer plus decision owner |
| Published catalog | Approved tools listed by need; declined and in-evaluation lists | Updated at every review | Reviewer |
| Metrics page | The six trends above | Monthly | Reviewer |
| Allowed / blocked / tolerated policy | Who decides what, and how items move | Reviewed yearly | Management |
| Privacy boundary document | The one page that says "services, not people" | Reviewed yearly | Management plus legal or HR |
| Conversation records | Finding, need, agreement, follow-up, for every conversation | Per conversation | Whoever had it |
If you are starting from nothing, begin with the privacy document (because everything else needs its permission). Next come the sanctioned list and the nightly job, then the monthly review. Follow with the catalog and evaluation lane, then the metrics. The policy can be a few paragraphs at first and grow as real cases test it. What you should not do is start with blocking, because a program that begins by saying no has spent its goodwill before it has earned any.
Where to go next
You now have the shape of a program: two reviews, a request path that beats going rogue, six honest trends, and a three-list policy. You also have a written boundary that keeps the whole thing about services rather than people. The technical pieces it depends on are in finding unsanctioned SaaS and finding unofficial hardware. The human piece is in the diplomatic conversation. If you need to explain to someone why any of this matters, what shadow IT looks like on the wire is the page to send. All of it lives under Discovering Shadow IT Through Network Data. The first hunt was the exciting part. The program is the part that works.
Put this into practice with SysPeep Monitor
A program like this needs its device-side data collected on a schedule without anyone remembering to do it. SysPeep Monitor auto-discovers the devices and interfaces on your network and polls them over SNMP and WMI. It keeps reports and dashboards in one web console, and lets you set alert rules for the changes you care about. The quarterly inventory diff starts from data that is already there. The free trial installs on any Windows machine in minutes.
Download Free TrialLearn About SysPeep