Microsoft Ads can report 14 conversions while your sales inbox holds two real enquiries. If those numbers describe different outcomes, more optimization can make the account better at buying the wrong one.
Say you imported your best Google campaign three months ago and now spend around $3,500 a month on Microsoft. Clicks cost less, the conversion column has numbers in it, and sales still wants to know where the business is. The tempting question is which agency can take over. I would ask what the account has been told to count.
Paying for a click does not tell Microsoft what happened next. It knows the search, the auction and the click. It only knows about the enquiry, the qualification call and the sale if a feedback path tells it. Reporting, conversion tracking and sales qualification describe different stages of the same funnel. Changing managers helps only if the replacement connects them.
Follow one cleaning enquiry from search to sale
I’ll use a Chicago cleaning company throughout. Someone searches “office cleaning services quote,” sees its Microsoft ad and clicks. From there, the chain looks like this:
- Search: someone expresses an interest.
- Click: the ad brings them to the site.
- Landing page: they reach the quote page.
- Enquiry: they submit the form or call.
- Qualification: the office establishes whether the job fits.
- Sale: the prospect becomes a customer.
The qualification step matters. An 800-square-foot site wanting once-a-month service is not the same opportunity as 22,000 square feet needing cleaning five nights a week. Both people can fill in the same form. Only one may deserve a walkthrough.
Microsoft records the search and click automatically. The business outcomes need a path back into the account. Without it, the conversion column cannot tell the difference between those two jobs.
Importing a working Google campaign does not solve this. It brings campaign structure across, not a complete, working feedback loop. The structure can look familiar while the measurement underneath behaves differently. Budgets, bid strategies, partner networks, location settings and tracking all need post-import checks. An ongoing Google auto-import can also overwrite changes made in Microsoft, depending on what is being synced.
Before anyone touches bids, I want to know which part of that chain broke.
Three failures that look like one bad campaign
A spend graph cannot distinguish these failures:
- No enquiry occurred. The traffic or page failed to produce one.
- An enquiry occurred but was not recorded. The business received something the account cannot see.
- A conversion was recorded but was not valuable. The account counted an outcome sales would reject.
An account can contain more than one. That is why “just optimize it” is not a diagnosis.
1. No enquiry occurred: inspect the traffic and page
For the Chicago cleaning company, imagine 400 Microsoft clicks and one phone call. The failure happens before a useful enquiry reaches the business. No amount of tidying the conversion report creates the missing calls.
Imported keywords carry assumptions from Google into a smaller auction with a different audience and traffic mix. Partner traffic may put the ad on sites the Google campaign never touched. A campaign can retain its familiar name and keywords without retaining the conditions that made it work.
The landing page can finish the job. I used to tell clients a copied campaign deserved a copied landing page. I was wrong. A 12-field quote form that works with brand traffic behind it may struggle with cold, partner-mixed traffic in front of it. The form has not changed. The visitor has.
My first checks are concrete:
- Read the search-term report, not just the keyword list.
- Segment performance by network.
- Check location settings and budgets against the traffic available.
- Check what the import continues to sync.
- Compare clicks, sessions and form starts.
If sessions match clicks but form starts do not, stop talking about bidding. Read 100 actual queries before changing a bid. Establish whether the right people reached a page they could reasonably use.
2. An enquiry occurred but was not recorded: inspect the wiring
Now consider a different version of the same account: the inbox holds 11 quote requests from Microsoft traffic this month, but Microsoft reports two conversions. That gap points toward measurement, not an absence of demand.
A UET base tag can show Tag Active simply because the tag is present. That does not establish that the intended conversion event fires or that its details match the goal. For a custom-event goal, I check the event code and goal definition separately. A green tag check is not a completed enquiry.
I work through the chain in order:
- Confirm the base tag loads.
- Complete the enquiry action and check that the intended event fires.
- Compare that event with the conversion goal’s conditions.
- Check whether the completed enquiry appears in reporting.
If those checks do not explain the gap, investigate whether Microsoft’s industry-related UET filtering applies. Rebuilding the same tag repeatedly will not resolve filtering on Microsoft’s side.
Revenue needs its own check. A counted conversion and a recorded conversion value are separate things. If values are missing, inspect the revenue event and goal configuration, rather than assuming the destination URL supplies everything.
If enquiries exist in the CRM but not in Microsoft, fix measurement before adjusting bids. Otherwise, the next manager starts with the same blind spot.

3. A conversion was recorded but was not valuable: inspect the instruction
Return to the original problem: 14 reported conversions, two real enquiries. Suppose the other 12 are one newsletter signup, three PDF downloads, two phone-button taps and six duplicate counts from thank-you-page reloads.
The reporting looks healthy because the account counted what it was configured to count. That is precisely the problem.
If informational actions are included in the goals used for bidding, automated bidding can pursue those easier outcomes instead of the ones that generate business. A download is not a booked walkthrough. A tap on a phone button is not necessarily a conversation.
The fix starts with the goal list:
- Keep purchases, qualified leads and booked calls as the outcomes the account is meant to pursue.
- Keep informational actions available for reporting without treating them as equivalent successes.
- Remove duplicate counting so one enquiry does not become several conversions.
The goal list is not just a report. It is an order slip. Before asking the account to deliver more, check what you ordered.
Choose the deepest outcome you can measure reliably
Think of the outcomes used for bidding as training data. If every form fill counts equally, the account cannot learn from a distinction you never send back: 10 of 100 leads became sales-qualified leads, while 90 went nowhere.
The useful principle is to define the deepest reliably measurable outcome before changing bids. That might be a marketing-qualified lead, sales-qualified lead, opportunity or revenue. The word reliably does real work here.
For the cleaning company, a raw quote request says someone showed interest. A qualified walkthrough says the enquiry meets the business’s requirements. Those are different instructions. The mechanism described in this SaaS pipeline case study follows the same logic: send back outcomes closer to revenue instead of asking the account to infer them from lead volume.

But a better-defined outcome does not manufacture more examples of it. Our cleaning company has 14 reported conversions across three months, before removing the weak ones. That is a thin signal for automated bidding.
The practitioner benchmark cited for automated bidding is at least 30 conversions in 30 days. I treat that as a warning about sparse data, not a magic switch that turns learning on at conversion 30. A low-volume account still needs decisions. It does not need false confidence in a target CPA built on a handful of mixed outcomes.
Timing matters too. Add a 60-day sales cycle, and yesterday’s reported CPA cannot tell the whole story. Judging that business on seven days of results risks cutting keywords before their enquiries have had time to close.
Choose an outcome you can measure consistently, then judge it on a timeline that fits the sale. Deeper is useful only when the signal arrives intact.
Return the CRM’s evidence to the account
The CRM may know which cleaning enquiries became walkthroughs and contracts. Microsoft will not know unless that information travels back.
With auto-tagging enabled, the landing URL carries a unique msclkid. That identifier needs to survive the journey into the CRM so the outcome can be uploaded against an offline conversion goal. This is how the click meets the sale.
For the Chicago company, I want the chain to work like this:
- Capture the
msclkidfrom the visit and preserve it in a first-party cookie. - Pass it into the CRM with the enquiry.
- Record qualification and closed-won as separate stages.
- Upload the relevant outcomes against their corresponding goals.
- Keep that feed running daily rather than treating it as a one-off setup.
The details still matter. Allow the required delay after creating an offline goal before the first upload. Check that conversion times follow the click and fall within the applicable upload and conversion windows. Microsoft’s 90-day upload limit is not a reason to ignore the actual sales cycle.
I also reconcile the records before trusting the result. Email, phone and timestamp provide more useful matching evidence than a name alone. Blank click identifiers need investigation. Consent belongs in the same review, not in a separate drawer nobody opens.
The SaaS attribution rebuild described here uses the same underlying mechanism: replace raw form-fill feedback with sales-qualified events whose identifiers and timing remain intact.
If sales cannot match a Microsoft enquiry to a quote or a lost reason, the feedback loop still has a CRM gap. A new bid strategy cannot fill it.
Separate bidding feedback from proof of growth
Once a sale reaches the account, there is one more distinction to keep straight: attribution is not proof of incremental revenue.
A Microsoft click can receive credit inside the chosen attribution window even when the buyer also clicked Google, opened three emails and drove past the cleaning company’s vans for a year. That attribution can help train bidding. It does not establish that Microsoft created a sale that otherwise would not have happened.
I keep two questions separate:
- What should the account bid for? Put qualified opportunities and revenue beside the campaigns that receive credit.
- Did the spend grow the business? Examine qualified pipeline with tracking held constant, using a holdout or a before-and-after rather than relying on last-click screenshots.
Clean feedback makes the first question answerable. It does not automatically settle the second. Keep both questions visible, or a better-looking report can pass for a better business.
Make a takeover manager own the diagnosis
A manager who opens with a new bid strategy has skipped the work I want done first. My takeover checklist follows the funnel:
- Reconcile measurement. Check click-ID capture, UET coverage, event matching and duplicate counts.
- Reconcile outcomes. Establish which reported conversions became genuine enquiries, qualified opportunities and sales.
- Name the failure in writing. Put the evidence beside the diagnosis.
- Fix the broken stage. Traffic and page problems need different work from missing events or rejected leads.
- Keep checking the feed. A working setup needs ongoing verification, not a screenshot from launch day.
This is the operating model I want from groas: continuous execution watches the feed, a named strategist owns the diagnosis, and changes come with reasoning in plain language. A percentage-of-spend freelancer checking in monthly is a weaker fit for that cadence, particularly on $3,500 in Microsoft spend.
The handover should transfer accountability for the feedback loop, not just access to the dashboard.
Test the model on a garage-door company
Now take a new case. A Dallas garage-door company spends $2,000 a month on Microsoft, reports 22 conversions at roughly $90 each, and books three jobs. The owner wants cheaper clicks.
I would start by matching those 22 conversions to call recordings and form entries. Say that produces nine genuine enquiries, six duplicate counts from page reloads and seven brochure downloads included among the bidding goals.
That identifies failure three: the account counted outcomes the business did not mean to buy. The reported $90 CPA is not a cost per genuine enquiry. That figure is about $222.
Next comes volume and timing. Nine enquiries in 30 days sits well below the practitioner benchmark discussed earlier. I would not treat a target CPA based on the original 22 conversions as a reliable instruction. It bundles enquiries, duplicates and downloads into one number.
The fix follows the cause: take downloads out of the bidding goals, eliminate duplicate counts, create separate offline goals for qualified leads and booked jobs, preserve msclkid in the CRM, and upload outcomes daily. Hold the budget flat while repairing the signal, then assess what the qualified enquiries actually produce.
Cheaper clicks were the owner’s proposed solution. They were not the diagnosed problem. Changing managers only pays when the new manager repairs the feedback loop instead of optimizing the broken one.

