The Distribution Channel Every AI Sales Strategy Ignores

Ask most B2B revenue leaders where their AI budget went this year and you will hear about SDR tools, CRM copilots, call intelligence, and lead scoring. Ask the same leaders how much of their actual revenue moves through a distributor or a reseller rather than their own reps, and a surprising number cannot answer quickly. That gap is not a coincidence. It is the shape of a blind spot that sits underneath almost every AI-in-sales conversation happening right now.

The AI-in-sales conversation has a blind spot

Most AI and digital sales investment assumes a direct motion where your own reps own the customer relationship end to end, and that assumption quietly excludes the layer where a lot of B2B revenue actually moves. Companies spend more than four times their AI and SaaS platform budget on channel partners, according to Forrester research. The layer getting the least AI attention is the layer where the most money already flows.

I understand why this happens, because I have done it myself. When you invest in tooling for a direct sales team, you are optimizing a system you can see completely. You own the CRM. You set the comp plan. You can watch a rep’s pipeline move in real time. A channel partner is a different kind of system. It runs its own books, keeps its own customer relationships, and shares only the slice of data it chooses to share. Faced with a system you can fully instrument and a system you can only partially see, almost every operator points the budget at the one they can see.

The problem is that the system you can only partially see is often the bigger one. In categories built around physical distribution, consumer packaged goods, industrial supply, even a meaningful share of enterprise software sold through resellers, the channel is not a side door. It is the main road. Treating it as an afterthought because it resists easy instrumentation is exactly backward from where the revenue actually sits.

This is not an argument that direct-motion AI investment is wasted. It is an argument that the conversation about AI in B2B sales has been narrower than the revenue it is trying to serve. A predictive model that scores your own reps’ pipeline cannot tell you whether a distributor is quietly sitting on six months of your product because a promotion made it cheap to buy, not because anyone downstream wants it yet.

What selling in and selling out actually mean

Selling in is the transaction where you move product to a distributor, who pays you whether or not the product ever reaches an end customer. Selling out is the separate transaction where that distributor actually moves the product off a shelf and into a buyer’s hands. The two numbers can move in opposite directions for months before anyone with a dashboard notices the gap between them.

The HBR case study that crystallized this for me involves a packaged goods company operating in a South American market through more than 100,000 retail outlets, almost entirely through distributors rather than direct sales. To grow revenue, the company ran repeated trade promotions, offering distributors a lower price for buying in bulk. The distributors did exactly what the incentive told them to do. They bought at the discount and stockpiled.

What followed was not growth. It was inventory. Stock piled up in distributor warehouses, carrying costs climbed, and the company’s own production schedule got harder to plan because nobody could tell when the stockpiled product would actually move. The lower price never produced a market-share gain, because the lower price was never the thing standing between the product and the end customer. The product was stuck in the middle of the chain, and the chain had no reason to unstick it as long as the distributor had already been paid.

This is the trap built into selling-in metrics. A rising shipment number looks exactly like a rising sales number on a quarterly report, right up until the quarter the distributor stops ordering because their warehouse is still full of what you sold them last time. By then the slowdown looks sudden. It was not sudden. It had been building since the first promotion that rewarded the wrong behavior.

Why no dashboard catches this until it is too late

A standard sales dashboard tracks orders and shipments, which look like revenue and feel like progress, but they measure what left your warehouse, not what reached a customer. The real signal, whether the end customer is actually buying the product off a shelf, stays invisible until the channel is already full.

This is not a reporting oversight. It is a structural feature of how shipment data gets generated. Sell-in data is produced automatically at the point of transaction between you and the distributor. Sell-out data requires visibility into what happens after that transaction, inside a system you do not own, and most companies never build the integration or the relationship required to see it. So they build their forecasts and their AI models on the data they have, which is exactly the data that lags the real signal by months.

I think about this the same way I think about the no-list discipline I have written about for AI agents: the thing that gets measured easily is rarely the thing that matters most, and the gap between the two compounds quietly until it is expensive. A rep’s pipeline dashboard at least sits inside a system you control end to end. A channel dashboard built only on shipment data is confidently reporting the wrong layer of reality, and it will keep doing so no matter how much AI gets pointed at it.

The fix is not a smarter model trained on the same shipment data. A model trained on sell-in numbers will get very good at predicting sell-in numbers, which is the metric that was misleading everyone in the first place. You need a different input entirely, not a better algorithm applied to the wrong one.

The fix was not more AI. It was realigned incentives plus real-time data

The company in the HBR case fixed its stockpiling problem by connecting distributors to its ERP system for real-time data by SKU, by store, and by distributor order, and by changing how account managers were paid. The second part did the heavier lifting. Account managers stopped being measured on how much they shipped into the channel and started being measured on how much moved out of it.

That single change in compensation is the whole story, more than the technology that came with it. Once account managers were paid on sell-out, flooding a distributor with discounted inventory stopped being a win. Helping a distributor actually move product off shelves became the only way to hit a number. The real-time data made that possible to measure and reward. The data alone, without the compensation change, would have just given everyone better visibility into a problem nobody was incentivized to fix.

This is the load-bearing point, and it is the same argument I keep coming back to about AI as a management problem rather than a capability problem, applied one layer removed from where I usually write about it. Anthropic’s research on alignment makes a related point about current AI systems more broadly: the technology reflects and amplifies observed behavior, it does not independently correct for it. The same dynamic shows up here with a compensation plan standing in for a model. Point sophisticated forecasting at a channel that is paid to stockpile, and you get a more efficient stockpiling operation. Point the same tooling at a channel that is paid on sell-out, and the forecasting starts doing what everyone assumed AI was supposed to do in the first place. The technology did not create the result in the HBR case. The incentive redesign did. The technology just made the redesigned incentive measurable and fast.

Sales grew by double digits in the regions where this ran, which is a real number worth sitting with. It came from a spreadsheet change to a comp plan and an ERP integration, not from a new model or a new tool category. I would not have guessed that six months ago, and I think most revenue leaders reading this would not guess it either, because the instinct is always to reach for new technology before auditing what the existing technology is being asked to optimize.

A second case, and why the same fix does not always apply

Not every channel problem is an incentive problem. A separate case in the same HBR research involved a beverage company operating through distributors across more than 200,000 outlets in Western Europe, and the failure there had nothing to do with stockpiling. The company simply had no structured information about its own customers. It could not see which outlets were ordering, what they were ordering, or how often.

The fix for this company was different from the fix above, and the difference matters. There was no compensation plan to redesign, because the account managers were not being rewarded for the wrong behavior. They were working blind. The company built a sales information system that pulled analytics from fragmented sources into one place, added omnichannel contact options including email, messaging, loyalty programs, and a distributor-facing ordering platform, and used the resulting visibility to route account attention more efficiently. Roughly half of all orders are now placed digitally, which gave the company, for the first time, a real-time view of what was actually selling.

Put the two cases side by side and the lesson sharpens considerably.

Dimension	Stockpiling case	Invisibility case
Core problem	Incentives reward shipment volume over sell-through	No structured data exists on end-customer demand
Root cause	Trade promotions pay distributors to buy in bulk	No system connects distributor activity back to the company
Primary fix	Compensation tied to sell-out, not sell-in	A sales information system plus omnichannel ordering
Data role	Makes the new incentive measurable	Replaces guesswork with real demand signal
Result	Double-digit sales growth in the regions where it ran	Roughly half of orders now placed digitally

A company that diagnosed the stockpiling case correctly but applied the invisibility company’s fix would have shipped a beautiful ordering platform into a channel still paid to hoard product, and nothing would have improved. A company that diagnosed the invisibility case as an incentive problem would have redesigned a compensation plan for account managers who were already doing the right thing, and the real gap, the missing data, would have stayed exactly as wide as before. Diagnosis has to come before the fix, and the two diagnoses look similar from a distance. Both show up as a sluggish channel. Both get blamed on the distributor. Neither gets fixed by assuming the other company’s playbook applies.

Why most B2B software companies never build this muscle

Most B2B software companies run a direct-only go-to-market motion with no distributor layer at all, which is efficient and simple, and it is also why channel management rarely gets discussed with the same rigor as direct sales inside the software industry. The operators writing about AI in B2B sales mostly run companies that never have to answer the sell-in versus sell-out question, so the muscle never gets built, and the handful of companies that do have a channel layer often treat it as a technology problem rather than the management problem it usually is.

I will say the honest version of this about myself. Across every CRM company I have run, my first instinct when a growth number stalled was always to point tooling and analytics at the direct sales team, because that team’s data lived inside a system I already owned and could query in an afternoon. A channel relationship, where one exists, requires a harder conversation: picking up the phone, asking a partner to share data they have no obligation to share, proposing a compensation change that touches an agreement already in place. The direct team is simply easier to instrument, so the direct team gets the attention, even when the harder, slower conversation is where the bigger number was sitting the whole time.

That four-times figure from Forrester should function as a standing correction to that instinct. If a company is spending four dollars through partners for every dollar spent on its own AI and SaaS tooling, and all of its analytical attention is going to the one-dollar side of that ratio, the attention is pointed at the wrong share of the business by a wide margin. McKinsey’s research on B2B distribution economics describes a version of the same pattern at a broader scale: companies consistently underinvest in the partner and distributor layer relative to its share of revenue, because the direct motion is the one with a dashboard already attached to it. This is not a call to abandon direct-motion investment. It is a call to stop treating the channel as the slow, secondary problem that gets addressed once the direct motion is fully optimized, because for a lot of companies the channel never gets its turn.

A short diagnostic for anyone with a channel layer

Anyone running a business with a distributor or reseller layer can run a short diagnostic this quarter instead of assuming the answer. Three questions do most of the work, and the order matters because the first two determine which fix you actually need before you touch the third.

First, do you have sell-out data at the SKU and store level, or only sell-in and shipment data? If the honest answer is only shipment data, you are in the same position the beverage company was in before it built its sales information system, and no compensation change will help you until the data exists to see what you are rewarding.

Second, is any part of your partner compensation tied to inventory moving into the channel rather than inventory moving out of it? If yes, you likely have the stockpiling company’s problem regardless of what your sell-out data shows, because the incentive is actively working against the outcome you want every single day it stays in place.

Third, if you pointed the same AI and analytics investment you have made in your direct sales team at your channel partners this year, what would you actually find? For most companies with a channel layer, the honest answer to that third question is some version of “we do not know,” and that answer is itself the finding. The absence of a confident answer is the blind spot made visible.

The closing thought

The AI conversation keeps pointing inward at the sales team you can see, because that is the system you built and the system you can fully instrument. The channel is harder to see by design, which is exactly why it accumulates the kind of quiet misalignment that a stockpiling problem or an invisibility problem can hide inside for years without tripping a single alert on a direct-sales dashboard. Neither of the two companies in this essay fixed their channel with a new AI product. Both fixed it by first figuring out whether they had an incentive problem or a visibility problem, and only then applying the technology that made the correct fix measurable.

I keep returning to a version of this same argument across founder lessons worth learning the hard way: the instinct to reach for new technology is almost always faster and more comfortable than the instinct to audit what you are already rewarding, and the second instinct is usually the one that actually moves the number. More essays on where the management problem hides inside an apparent technology problem live at dearmer.com.au, and the thinking behind why I keep returning to this thread is on the about page.

If you pointed the same AI and analytics investment you have made in your direct sales team at your channel partners this year, what would you actually find?