Industrial OEM accounts combine several relationships at once. The customer buys equipment, spare parts, projects and service. Different plants can run different generations of the product. Engineering may trust the OEM while procurement pressures price. A distributor may own the commercial contact while the OEM carries the warranty risk. Revenue alone cannot describe that account.
Build the score from events the team can verify: delivery performance, product problems, service response, agreement fulfillment, payment, forecast behavior, stakeholder access, open commitments and installed-base activity. Preserve the underlying facts and show the few signals that changed the result.
This playbook separates present health, near-term risk, growth potential and data confidence. It gives account teams a score they can challenge, trace and act on.
Decide what the score should predict
Name the decision before choosing the inputs. A renewal-risk score predicts whether an agreement will renew. A run-rate risk score predicts whether actual orders will fall below the customer commitment. A relationship score describes access and trust. An opportunity score estimates addressable demand. Those outputs can inform one account review, but they answer different questions.
Choose a time horizon that matches the commercial cycle. A 30-day horizon may suit a parts reorder. A capital-equipment program may need twelve or twenty-four months. Then define the observable outcome: renewal, material order decline, lost platform, executive escalation, credit hold or another event the business can label consistently.
Research on customer evaluations in key-account relationships found that positive relationship strength and relationship vulnerability had different drivers. That finding supports a practical rule: do not assume that the absence of complaints proves loyalty, or that a complaint proves the customer plans to leave.
Do not combine health, risk and potential
Use separate outputs. Health describes current delivery, product, commercial and relationship performance. Risk identifies a credible negative event within a stated horizon. Potential estimates additional value the OEM and customer could create. Confidence shows how much current evidence supports the other three.
A high-potential account can have poor health. A healthy account can have little remaining growth. Missing data should lower confidence, not automatically mark the customer unhealthy. When a dashboard blends all four, the team cannot tell whether to repair service, protect a renewal, pursue an upgrade or verify the data.
| Output | Question | Example result |
|---|---|---|
| Health | How well does the current relationship work? | 62/100; delivery and open quality actions weakened |
| Risk | What negative event may occur, and by when? | High renewal risk within 180 days |
| Potential | What credible demand remains available? | High retrofit and service-agreement potential |
| Confidence | How complete and current is the evidence? | 74/100; two sites lack current contacts |
Score the right level of the account
Define the account hierarchy before calculating anything. A corporate parent can negotiate the master agreement while individual plants place orders, operate equipment and experience service. A manufacturing program can span several customer legal entities. A dealer can mediate the relationship without owning the equipment.
Store health at the lowest level where the evidence and action make sense: site, program, fleet, contract or buying group. Roll up to the parent with the contributing records visible. Do not let strong parts revenue at one plant hide a major failure at another.
For each level, name the commercial owner, service owner, decision-makers, active contracts, installed assets and channels. A score with no accountable owner becomes a report, not a management tool.
| Level | Typical evidence | Typical action |
|---|---|---|
| Corporate parent | Master terms, executive relationship, credit and total exposure | Executive review or contract negotiation |
| Business unit or program | Forecast, platform position, engineering roadmap and sourcing decision | Protect nomination or develop next program |
| Site | Deliveries, service events, contacts, assets and complaints | Recover execution or propose site work |
| Fleet or asset family | Configuration, use, failures, agreements and parts demand | Service, parts or retrofit plan |
| Contract | Entitlement, obligations, SLA performance, margin and renewal date | Correct performance or prepare renewal |
Use five health dimensions
Start with five dimensions that cover the current relationship: delivered value, execution, product and service experience, commercial behavior and relationship access. Add a sixth dimension only when the business model needs it, such as program alignment for a tier supplier or agreement performance for a service-led OEM.
Limit the model to inputs an account owner can explain. Ten reliable measures beat fifty correlated measures that move together after the same late order. Each input needs an owner, source, refresh rule, expected range and action when it fails.
| Health dimension | Useful evidence |
|---|---|
| Delivered value | Accepted output, achieved performance, adoption, utilization or customer result |
| Execution | On-time delivery, promise changes, backlog, response, first-time fix and open actions |
| Product and service experience | Failures, warranty, complaints, repeat visits, returns and severity |
| Commercial behavior | Order performance, payment, margin, disputes, forecast accuracy and agreement compliance |
| Relationship access | Decision-maker coverage, sponsor strength, meeting outcomes and unanswered commitments |
Measure delivered value in the customer’s terms
Use the result the customer bought. For production equipment, that may include throughput, availability, yield, energy use or safety. For a component supplier, it may include launch timing, defect rate, line continuity or engineering response. For a service contract, it may include response, restoration, planned-maintenance completion or cost predictability.
Record the baseline, target, period, data source and factors outside the OEM’s control. A machine cannot meet an output target when the customer runs an unapproved material or withholds required maintenance. The account review still needs that fact, because a disappointed customer may blame the supplier even when the contract excludes the condition.
Use customer-confirmed measures where possible. Internal completion does not prove customer acceptance. Record acceptance, qualification, production release or another event that closes the promised outcome.
Track execution before revenue reacts
Revenue often falls after months of missed promises. Use operational signals that appear earlier: late milestones, repeated delivery-date changes, incomplete shipments, aged technical questions, overdue corrective actions, service response misses and open commissioning items.
Measure severity and customer impact, not ticket count alone. One unresolved safety problem outweighs ten routine requests. A small late shipment can stop a line; a large order can arrive early without helping if installation waits on one missing component.
Show the oldest open commitment and its named owner. Averages hide the issue that the customer remembers.
| Execution signal | Better definition |
|---|---|
| On-time delivery | Complete delivery against the current customer-confirmed promise |
| Promise stability | Number and magnitude of supplier-driven date changes |
| Technical response | Time to a usable answer, by severity |
| Corrective action age | Days since the OEM committed to the action |
| First-time fix | Service visits resolved without repeat labor or missing parts |
| Commissioning closure | Accepted items versus open issues after planned startup |
Separate product pain from relationship response
A failure creates product risk. The OEM’s response can repair or deepen the relationship damage. Track defect severity, affected units, recurrence, containment, root cause, corrective action and customer acceptance. Then track communication speed, ownership and whether the OEM met each commitment.
A candid response to a serious failure can strengthen trust. Silence around a modest repeat issue can destroy it. Keep both facts in the account record so the score does not punish a customer for reporting problems or reward a supplier that closes tickets without solving them.
Research using longitudinal B2B service-quality data found that service pain helped predict churn behavior. The practical point is simple: service history belongs in an industrial account model even when equipment sales produce most of the current revenue.
Read commercial behavior in context
Use planned and actual orders, backlog, cancellations, payment, disputes, margin and forecast changes. Compare each measure with the customer’s contract, seasonality, program phase and market. A 20% order decline can indicate lost share, a scheduled shutdown, inventory correction or the end of a platform.
Salesforce Manufacturing Cloud account forecasts can combine opportunities, orders, sales agreements, historical orders and custom measures. The useful pattern is planned versus actual at the right product, location and period. A health model should use that variance with a reason, not treat every shortfall as relationship decay.
Flag sudden discount demands, invoice disputes and late payment, but preserve the commercial cause. The customer may contest a legitimate defect, suffer a temporary credit issue or use procurement pressure before a renewal. Finance and the account owner should see the same evidence.
Score relationship access with named people
Count roles, not contacts. Identify the economic buyer, operational owner, engineering authority, procurement lead, service user, executive sponsor and detractor. Record who can approve the next decision and the date of the last meaningful interaction.
Email volume is a poor proxy for access. A weekly thread with a buyer can coexist with no relationship to the plant manager who decides whether the OEM remains on the platform. Capture meeting outcome, stated priority, commitment, concern and next step. Let the account owner distinguish confirmed customer statements from internal interpretation.
Watch for changes: sponsor departure, reorganization, acquisition, new procurement leadership, plant closure or a competitor entering an engineering trial. These events can change the relationship before orders move.
Keep installed-base opportunity outside health
The fleet helps explain the account and its potential, but a large aging fleet does not make the relationship healthy. Calculate active units, configuration coverage, agreement coverage, service due, parts demand, obsolescence and retrofit eligibility as opportunity measures.
Konecranes describes installed-base access as a key element of its service business and states that the data helps target both service and equipment sales. Its 2025 report also explains how condition and utilization data can lead sales teams to contact customers about recommended action. That is an opportunity engine. The customer’s response and the OEM’s execution then affect health.
Link each opportunity to the site, asset, technical reason and customer contact. A high opportunity score should produce a short list of credible actions, not a generic instruction to cross-sell.
Let critical conditions cap the score
A weighted average can hide a severe problem. Strong revenue, payment and access can outweigh an unresolved safety incident in the math even though no account leader would call the account healthy. Define hard caps for critical events.
Examples include an active safety or regulatory escalation, production stop caused by the OEM, credit hold, failed acceptance, executive complaint without an owner, material breach, expired mandatory certification or a renewal inside the notice window with no customer contact. The cap should remain until the team meets a clear exit condition.
Show the cap beside the score. Do not silently force a red result that nobody can explain.
| Critical condition | Example cap | Exit condition |
|---|---|---|
| Open safety escalation | Health cannot exceed 30 | Customer accepts containment and corrective-action plan |
| Customer production stopped by OEM issue | Health cannot exceed 40 | Production restored and recovery plan accepted |
| Credit hold | Health cannot exceed 45 | Finance clears payment or approves terms |
| Renewal notice window open with no contact | Renewal risk remains high | Decision-maker confirms review plan and date |
| Failed customer acceptance | Health cannot exceed 50 | Named acceptance criteria pass |
Treat missing data as uncertainty
Do not award full health because nobody recorded a complaint. Do not assign zero health because a site has no service feed. Calculate confidence from coverage, freshness, source reliability and agreement between systems.
Show which missing fact could change the decision. If the team cannot confirm whether a plant still runs twelve machines, that gap weakens opportunity confidence. If no one knows the renewal decision-maker ninety days before notice, it raises relationship risk as well as reducing confidence.
Assign recovery work: confirm active units, map contacts, reconcile open cases or obtain the customer forecast. The score should make poor data visible without turning data cleanup into an abstract project.
| Confidence factor | Question |
|---|---|
| Coverage | Do we have records for the relevant sites, assets, contracts and contacts? |
| Freshness | Did a recent event confirm the record? |
| Authority | Did the owning system or responsible person provide it? |
| Consistency | Do orders, service, CRM and finance agree? |
| Specificity | Does the evidence name the site, product, period and issue? |
Normalize for the customer’s normal rhythm
Compare a customer with its own expected pattern and a relevant peer group. Project OEMs, run-rate component buyers, distributors and service-contract customers behave differently. A quarterly buyer should not turn red after thirty days without an order.
Use product lifecycle and program timing. Orders fall after a launch build, rise before a shutdown and pause during customer inventory correction. Compare the same season or program phase when cycles matter. Record planned events so the model does not rediscover them as risk every month.
Refresh fast-moving signals daily or weekly and slow signals monthly or quarterly. Recomputing executive relationship health every hour adds noise.
Use a score that an account owner can reproduce
A practical first model can score each health dimension from 0 to 100, apply documented weights by account type and enforce critical caps. Publish the formula. Show the raw value, target, direction, age and contribution of each input.
For a service-led account, delivered value and execution may carry most of the weight. For a run-rate component account, order performance and program alignment may matter more. Hold the formula stable long enough to test it. Frequent weight changes make history impossible to interpret.
Do not ask managers to override the score until it “looks right.” Let them record a reasoned assessment beside the model result. Review consistent disagreement as evidence that the formula or data misses something.
| Example service-led account dimension | Weight |
|---|---|
| Delivered value | 25% |
| Execution | 25% |
| Product and service experience | 20% |
| Commercial behavior | 15% |
| Relationship access | 15% |
Show the reason, change and next action
Every account view should answer four questions: what is the current state, what changed, which evidence caused it and what should happen next. Show the top positive and negative drivers, critical caps, missing evidence, trend and owner.
Write the action as a decision or commitment. “Improve relationship” cannot close. “Service director will review the two repeat drive failures with the plant engineering manager by 12 October” can. Link the action to the records that created it and capture the result.
Avoid decorative alerts. If a signal cannot change priority, owner or action, remove it from the health view.
Worked example: stable revenue hides a weakening account
Northfield Components buys drives and control cabinets for three plants. Twelve-month revenue remains flat and payment stays current. A revenue-led dashboard marks the account green. The operating evidence tells a different story.
Plant A received only 63% of order lines complete by the customer-confirmed date during the last quarter. Two repeat drive failures remain under corrective action after 47 days. The engineering sponsor who defended the incumbent platform left the company. Procurement has invited a competitor into a qualification trial. Plant B remains stable. Plant C has twelve aging cabinets that qualify for a controls retrofit, but the OEM has not confirmed the current configuration on five.
The model returns health 54, high platform risk within twelve months, retrofit potential 82 and confidence 76. Delivery and repeat failures drive health down. Sponsor loss and the competitor trial drive risk. The aging cabinets drive potential. Missing configuration data lowers confidence.
The account plan does not begin with a broad sales campaign. Operations owns a delivery recovery plan. Engineering closes the repeat-failure action with Plant A. The account leader maps the new decision group and asks the customer to define qualification criteria. Service validates the five uncertain cabinets before preparing the retrofit case.
| Output | Result | Required action |
|---|---|---|
| Health | 54/100 | Recover delivery and close repeat-failure actions |
| Platform risk | High within 12 months | Map new decision group and competitor qualification |
| Potential | 82/100 | Validate and price the controls retrofit by site |
| Confidence | 76/100 | Confirm five cabinet configurations and Plant C contacts |
Validate the score against real outcomes
Backtest the model on historical accounts. Choose an outcome and prediction date, then use only information available at that date. Measure how often high-risk accounts experienced the event, how many events the model missed and how much warning time it provided.
Review false positives with the account team. Some will expose seasonality, planned program changes or missing context. Review false negatives for signals the model omitted or recorded too late. Test by customer segment because one threshold rarely works across equipment projects, component programs and service agreements.
Track whether the intervention changed the result. A model that predicts risk accurately but sends teams toward ineffective actions has limited value.
Run the score inside the account review
Review material changes weekly and the full account monthly or quarterly according to the business rhythm. Bring sales, service, operations, quality and finance when their evidence or action matters. The account owner should leave with a small set of commitments, not a debate about which dashboard is correct.
Lock the score snapshot used in the review. Preserve later data corrections separately. That record shows what the team knew, what it decided and whether the action worked.
Use the systems that own each fact
ERP should own orders, shipments, invoices, payment and cost. Quality should own defects, containment and corrective action. Field service should own service events and response. The installed-base record should own active assets and configuration. CRM should own contacts, opportunities and account actions. Contract systems should own obligations, entitlement and renewal dates.
Salesforce’s Manufacturing Cloud data model separates fleets, assets, sales agreements, orders, contracts, warranty claims and forecast facts. Regardless of software, that separation matters: the health workflow should reference each source record and avoid inventing a second version of the order, claim or asset.
The calculated account view can live in CRM or a dedicated application. It needs stable IDs, effective dates, source links, refresh status and permission rules across the underlying systems.
How Bourne builds an industrial account view
Bourne reads account evidence from CRM, ERP, quality, service, installed-base, contract and communication systems. It maps each event to the correct parent, site, program, contract and asset, while preserving the source and date.
The application calculates health, risk, potential and confidence separately. It shows the drivers and critical conditions, then prepares the action for the responsible person: resolve a late corrective action, confirm a renewal plan, recover a forecast variance, validate installed equipment or prepare an asset-specific offer. A person approves account judgments and customer-facing actions.
When the team completes the work, Bourne writes the outcome to the owning system and tests whether the underlying condition changed. The account review starts from current operating evidence instead of a manually assembled slide.
Pilot the model on twenty accounts
Select twenty accounts that include healthy renewals, current escalations, run-rate declines, project customers, service agreements, distributors and known lost business. Define one risk outcome and horizon. Build the account hierarchy and limit the first model to a small set of trusted inputs.
Have account owners review every input, score, cap and action. Record disagreement without forcing consensus. Backtest the model, then run it prospectively for one full commercial cycle. Measure prediction quality, warning time, action completion and outcome.
The pilot passes when teams can explain the result, discover material issues earlier, direct work to the right owner and distinguish a troubled relationship from an attractive opportunity with incomplete data.
Arda Bulut
Arda Bulut is the co-founder and CTO of Bourne and HockeyStack. He leads engineering at Bourne, building the platform people use to create AI products, agents and automations.