Commissioned Research
Top 9 Best Hmm Software of 2026

GITNUXSOFTWARE REVIEW

General Knowledge

Top 9 Best Hmm Software of 2026

Compare hmm software options and key tradeoffs for model building, inference, and reporting, with tools like MATLAB included for teams.

18 min readUpdated Editorially scoped research
How we ranked these tools
01Editorial Scope

Vendors set the topic, but our editors set the inclusion criteria and the evaluation framework before research starts.

02Source Research

AI search agents find sources, extract figures with provenance, and aggregate written reviews. They also transcribe video reviews and podcasts.

Read our full methodology →

Framework: Features 40% · Ease 30% · Value 30%

Gitnux may earn a commission through links on this page. This does not influence our reviews. Editorial policy

HMM software matters because hidden Markov models only deliver when the data model, estimation API, and inference pipeline match the use case. This comparison helps technical teams weigh tradeoffs between statistical modeling libraries and full research platforms, focusing on configuration depth, integration paths, and operational fit for production workflows.

MATLAB is the best pick for research teams building repeatable HMM sequence workflows because its single scripting workspace pairs estimation with custom feature engineering and visualization, whereas TensorFlow Probability is the better fit when you need differentiable HMM distributions that run inside TensorFlow graphs for custom latent-variable inference.

Editor’s top 3 picks

Three quick recommendations before you dive into the full comparison below — each one leads on a different dimension.

Editor pick
1

MATLAB

MATLAB’s end-to-end scripting loop combines HMM estimation with custom feature engineering and visualization in one workspace.

Best for: Fits when research teams need sequence modeling with custom feature engineering and repeatable batch evaluation scripts.

2

GE HealthCare Command Center

Editor pick

Predictive hospital command-center views combine capacity, patient movement, and bottleneck signals for coordinated operational decisions.

Best for: Fits when hospital systems need centralized capacity management and predictive coordination across inpatient operations.

3

TensorFlow Probability

Editor pick

State-space and distribution composition lets HMM-like models reuse TensorFlow execution, gradients, and inference APIs together.

Best for: Fits when research teams need HMM inference inside TensorFlow for custom latent-variable models.

Comparison Table

1
MATLABBest overall
enterprise
9.1/10
Overall
2
8.9/10
Overall
3
8.6/10
Overall
4
enterprise
8.4/10
Overall
5
API-first
8.0/10
Overall
6
API-first
7.8/10
Overall
7
API-first
7.5/10
Overall
8
7.3/10
Overall
9
6.9/10
Overall
#1

MATLAB

enterprise

MATLAB provides hidden Markov model functions through its Statistics and Machine Learning Toolbox.

9.1/10
Overall
Features9.1/10
Ease of Use9.0/10
Value9.2/10
Standout feature

MATLAB’s end-to-end scripting loop combines HMM estimation with custom feature engineering and visualization in one workspace.

MATLAB from MathWorks is a research-focused environment for modeling, estimation, and signal processing workflows that often include HMMs. It provides a scripting and visualization loop around probabilistic modeling so teams can simulate sequences, validate assumptions, and iterate on feature pipelines.

HMM usage is typically implemented through MATLAB code and toolboxes for statistical modeling, with integration into broader data cleaning, feature engineering, and evaluation scripts. Deployment is supported by MATLAB’s build and production workflows for moving analysis code toward repeatable batch runs.

Pros
  • +Mature numerical routines for inference and sequence modeling
  • +Tight scripting workflow for simulation, debugging, and visualization
  • +Strong interoperability with external files, C/C++, and data tooling
  • +Production-oriented batch processing for repeated experiments
Cons
  • –HMM workflows require code assembly rather than a dedicated GUI path
  • –Large training runs can hit memory limits without careful batching
  • –Ecosystem integrations depend on toolbox choices and custom glue code
  • –Governance and RBAC are not designed as HMM-specific admin controls
Use scenarios
  • Instrumentation engineers

    Model instrument signals with HMM

    Repeatable signal modeling

  • Quantitative researchers

    Simulate stochastic system sequences

    Scenario results

Show 2 more scenarios
  • Engineering research teams

    Analyze scientific data for HMM

    Interpretable outputs

    Clean and transform technical datasets in MATLAB, then interpret numerical results alongside HMM evaluation.

  • Software-oriented model developers

    Generate deployable HMM artifacts

    Executable implementations

    Convert validated HMM algorithms into packaging workflows that produce executable artifacts for downstream use.

Best for: Fits when research teams need sequence modeling with custom feature engineering and repeatable batch evaluation scripts.

#2

GE HealthCare Command Center

enterprise

AI-enabled hospital command center software for real-time capacity management, patient flow optimization, and care coordination across nearly 500 hospitals globally.

8.9/10
Overall
Features8.9/10
Ease of Use8.9/10
Value8.9/10
Standout feature

Predictive hospital command-center views combine capacity, patient movement, and bottleneck signals for coordinated operational decisions.

GE HealthCare Command Center gives health systems a shared operational view across beds, transfers, discharges, and departmental demand. Electronic health record integration can supply the operational data needed for real-time dashboards and predictive models. Separate views help executives, command-center staff, and department leaders act on the same facility conditions.

The product requires accurate source-system feeds and extensive local configuration for roles, escalation rules, and operational definitions. During bed shortages, transfer surges, or emergency department congestion, teams can use predictive signals to prioritize actions before capacity constraints spread. Smaller organizations may receive less value from its enterprise-scale operating model.

Pros
  • +Real-time capacity views connect bed status, patient movement, and escalation priorities.
  • +Predictive analytics identify likely bottlenecks before visible capacity failure.
  • +Command-center workflows support hospital-wide transfer and discharge coordination.
  • +Role-specific dashboards align executives, flow teams, and frontline departments.
Cons
  • –Public materials provide limited detail about external developer APIs and custom data schemas.
  • –Accurate source-system feeds remain essential for trustworthy operational metrics.
  • –Large networks may need extensive role and escalation configuration.
  • –Primary focus is inpatient operations rather than outpatient practice workflows.
Use scenarios
  • Hospital operations leaders

    Hospital capacity coordination

    Faster capacity decisions

  • Patient flow teams

    Emergency department congestion management

    Earlier bottleneck intervention

Show 1 more scenario
  • Health system executives

    Network performance oversight

    Comparable facility performance

    Role-specific dashboards show operational variation across facilities and support targeted improvement work.

Best for: Fits when hospital systems need centralized capacity management and predictive coordination across inpatient operations.

#3

TensorFlow Probability

API-first

TensorFlow Probability provides differentiable hidden Markov model distributions for Python.

8.6/10
Overall
Features8.6/10
Ease of Use8.6/10
Value8.6/10
Standout feature

State-space and distribution composition lets HMM-like models reuse TensorFlow execution, gradients, and inference APIs together.

TensorFlow Probability provides probabilistic modeling primitives in the TensorFlow ecosystem for tasks like Bayesian inference, uncertainty quantification, and probabilistic forecasting. It includes HMM-ready building blocks such as state-space models and distribution APIs that support custom likelihoods and latent-variable inference.

The library also exposes MCMC and variational inference workflows through TensorFlow execution, which helps productionize inference graphs with the same tooling used for other TensorFlow deployments. HMM implementations typically combine transition and emission distributions with an inference engine written in TensorFlow.

Pros
  • +Tight integration with TensorFlow graphs and accelerators for inference workloads
  • +Distribution and bijector APIs support custom emission models without leaving TensorFlow
  • +Multiple inference engines including variational inference and MCMC workflows
  • +State-space and sequence modeling components map directly onto HMM structure
Cons
  • –HMMs require assembling primitives rather than using a dedicated HMM high-level module
  • –Modeling and inference configuration demands TensorFlow and probabilistic programming fluency
  • –End-to-end training and deployment automation for clinical pipelines is not included
  • –Performance tuning often depends on graph shape, masking, and sampler configuration
Use scenarios
  • Probabilistic modeling teams

    Build HMM state-space inference

    Structured inference graphs

  • Bayesian inference practitioners

    Run MCMC on HMMs

    Posterior state estimates

Show 1 more scenario
  • Uncertainty quantification engineers

    Train HMM via variational inference

    Uncertainty-aware forecasts

    Set up variational inference workflows that optimize HMM latent models with uncertainty-aware outputs.

Best for: Fits when research teams need HMM inference inside TensorFlow for custom latent-variable models.

#4

Huma

enterprise

EU MDR Class IIb and FDA-cleared remote patient monitoring platform with low-code clinical application configuration and automated triage capabilities.

8.4/10
Overall
Features8.4/10
Ease of Use8.4/10
Value8.4/10
Standout feature

Huma’s study intake to workflow task generation model maps requirement inputs into routed operational actions with traceable outcomes.

Huma focuses on healthcare research workflows that connect study requirements to operational execution through structured intake and configurable orchestration. Core capabilities include configurable forms, task and workflow routing, and integrations that move study data between systems used for patient recruitment and operations.

Pros
  • +Configurable workflow routing turns study intake into assigned operational tasks
  • +Integration hooks support moving study data between external systems used in research
  • +Activity history provides traceability for requests and downstream actions
  • +RBAC supports separating administrative, operational, and data-handling roles
Cons
  • –Automation setup needs careful configuration to avoid inconsistent routing outcomes
  • –Clinical workflow depth for day-to-day care delivery is limited versus EHR-centric tools
  • –Advanced governance reporting can require extra process alignment across teams
  • –Template coverage for research-specific edge cases may require configuration work
Use scenarios
  • Clinical operations leaders

    Route new study intake tasks

    Traceable task execution

  • Study coordinators

    Standardize configurable intake forms

    Consistent study capture

Show 1 more scenario
  • Clinical governance admins

    Control study workstream access

    Controlled governance

    Apply role-based access controls and track actions across study workflows for oversight.

Best for: Fits when research operations teams need configurable intake and workflow automation with audit traceability.

#5

Pyro

API-first

Pyro supports hidden Markov modeling through probabilistic programming with Python and PyTorch.

8.0/10
Overall
Features8.0/10
Ease of Use8.0/10
Value8.0/10
Standout feature

Run configuration and orchestration via API calls that standardize experiment lifecycles across datasets.

Pyro.ai focuses on operationalizing HMM workflows for research teams, with an API-first approach to model orchestration and experiment runs.

Pros
  • +API-first automation for repeatable HMM experiment execution
  • +Configurable pipeline steps for consistent training and inference runs
  • +Batch-friendly orchestration for higher throughput
  • +Clear separation between run configuration and execution logic
Cons
  • –Limited native healthcare integration surfaces like HL7 or FHIR
  • –Requires engineering effort to build custom evaluation loops
  • –Governance controls like audit logs and RBAC are not its strongest area
  • –Workflow UI is thin compared with toolkit-based competitors
Use scenarios
  • Research engineering teams

    Orchestrate HMM training runs

    Managed experiment lifecycles

  • Data science batch processors

    Process HMM inference at scale

    Consistent batch throughput

Show 1 more scenario
  • Programmatic model builders

    Configure inference and likelihood logic

    Deterministic inference pipelines

    Program HMM transition and emission modeling with pipeline configuration controls for training and inference stages.

Best for: Fits when research teams need programmatic HMM pipelines with repeatable runs.

#6

pomegranate

API-first

pomegranate is a Python probabilistic modeling library that includes hidden Markov models.

7.8/10
Overall
Features7.8/10
Ease of Use7.8/10
Value7.8/10
Standout feature

PyTorch-based DenseHMM and SparseHMM classes combine custom emissions, GPU execution, minibatch training, and sparse transitions.

Teams can define transition matrices, start probabilities, end probabilities, and emissions with distributions such as Normal, Categorical, Poisson, and Bernoulli. DenseHMM supports fully connected transitions, while SparseHMM reduces computation for models with limited state connectivity. PyTorch tensors provide automatic differentiation, device selection, and integration with existing neural-network workflows.

Pros
  • +DenseHMM and SparseHMM cover fully connected and constrained transition structures
  • +PyTorch tensors support GPU execution and automatic differentiation
  • +Custom emission distributions extend models beyond built-in probability families
  • +Fit, score, predict, sample, and serialize operations share one Python API
Cons
  • –No graphical interface exists for designing or inspecting HMM structures
  • –PyTorch concepts add dependency and tensor-management overhead
  • –Production monitoring and experiment tracking require external tooling
  • –Sparse transition design requires explicit state-connectivity configuration
Use scenarios
  • Sequence modeling researchers

    Train latent-state models on sensor sequences

    Reproducible state inference

  • Fraud analytics teams

    Detect unusual transaction behavior

    Ranked anomaly candidates

Show 2 more scenarios
  • Bioinformatics developers

    Classify genomic sequence states

    Automated state annotation

    Developers assign nucleotide emissions to hidden states and train models across batched biological sequences.

  • Python ML engineers

    Embed HMMs in neural pipelines

    Integrated sequence models

    Engineers combine differentiable HMM components with PyTorch modules and shared device management.

Best for: Fits when Python research teams need customizable HMMs with GPU-backed training and code-first automation.

#7

hmmlearn

API-first

hmmlearn supplies Python implementations of hidden Markov models for statistical modeling.

7.5/10
Overall
Features7.5/10
Ease of Use7.6/10
Value7.4/10
Standout feature

scikit-learn estimator API for HMM fit, predict, and score workflows using the same pipeline patterns as other models.

hmmlearn is a Python-focused toolkit for training Hidden Markov Models and related Bayesian variants via code-first configuration rather than a GUI workflow. The core capability is estimation for discrete and continuous emissions using the EM family of algorithms, plus decoding tools like Viterbi-style most-likely state sequences.

Pros
  • +scikit-learn compatible estimators for training and inference workflows
  • +Built-in EM training for Gaussian and discrete emission models
  • +Viterbi-style decoding for most-likely hidden state sequences
  • +Extensible model options via subclassing and custom initialization
Cons
  • –No native healthcare interoperability support like FHIR APIs
  • –Limited governance features such as RBAC and audit logs for teams
  • –Requires careful initialization to avoid poor local optima
  • –Not designed for high-throughput production services without custom wrapping
Use scenarios
  • Python ML practitioners

    Train discrete emission HMMs

    Estimated HMM parameters

  • Time series analysts

    Decode most likely hidden states

    State sequence outputs

Show 1 more scenario
  • Data pipeline developers

    Integrate HMMs into scikit workflows

    Pipeline-ready models

    Use scikit-learn compatible estimators to embed HMM training and inference inside existing Python pipelines.

Best for: Fits when researchers need a Python HMM training and decoding engine embedded in ML pipelines, not a clinical system integration layer.

#8

Shivam Medisoft

SMB

Hospital management software with centralized control tower for real-time performance monitoring, revenue tracking, and operational analytics.

7.3/10
Overall
Features7.3/10
Ease of Use7.3/10
Value7.3/10
Standout feature

Configurable task-driven clinical workflow states that map intake steps to documentation and follow-up without custom code.

Shivam Medisoft is a healthcare management and monitoring application built for clinical and administrative workflows. Core modules commonly cover patient registration, appointment and intake management, and internal care coordination tasks.

Pros
  • +Workflow configuration supports multi-step clinical documentation flows
  • +Patient registration and appointment handling cover common front-desk tasks
  • +Interoperability focus uses HL7 messaging patterns for external system exchange
  • +Role-based access support supports separate operational responsibilities
Cons
  • –FHIR API coverage and granularity are limited compared with API-first tools
  • –Automation depends on manual configuration of workflows and forms
  • –Audit log depth for fine-grained clinical actions can be hard to validate
  • –Setup requires careful governance of roles, locations, and workflow states
Use scenarios
  • Clinic coordinators

    Route intake tasks to follow-up

    Consistent follow-up

  • Care coordination teams

    Manage appointment intake flows

    Structured intake handling

Show 1 more scenario
  • Healthcare IT staff

    Integrate with external systems

    Operational data exchange

    Connect to external health systems through messaging interfaces to support operational interoperability.

Best for: Fits when clinics need configurable intake and documentation workflows with HL7 exchange to outside systems.

#9

HealthCare Logic SystemView

enterprise

AI-enabled hospital intelligence platform providing real-time monitoring of ED, theatres, beds, and outpatient departments with predictive demand modeling at 96% accuracy.

6.9/10
Overall
Features6.9/10
Ease of Use6.9/10
Value6.9/10
Standout feature

Real-time command-center workspace linking bed capacity, patient movement, and discharge barriers.

SystemView gives bed managers and nursing leaders a shared view of available beds, blocked beds, patient locations, and pending discharges. Operational dashboards expose bottlenecks by unit, allowing teams to prioritize placement and escalation work. Its value increases when hospitals maintain consistent status definitions across departments and facilities.

Pros
  • +Real-time census and bed-state views support hospital command-center decisions.
  • +Discharge-barrier tracking creates a shared queue for unresolved placement issues.
  • +Configurable alerts route capacity exceptions to assigned operational teams.
  • +HL7 messaging supports feeds from existing hospital information systems.
Cons
  • –Stale source feeds can produce misleading bed-availability information.
  • –Primary workflows target inpatient operations, limiting value for outpatient practices.
  • –The interface prioritizes operational dashboards over detailed clinical documentation.
  • –Cross-facility comparisons depend on consistent definitions across participating sites.
Use scenarios
  • hospital operations leaders

    Manage daily bed capacity

    Quicker bed assignments

  • patient flow coordinators

    Resolve discharge barriers

    Fewer discharge delays

Show 1 more scenario
  • nursing supervisors

    Monitor unit bottlenecks

    Faster escalation

    Unit-level views reveal blocked beds and stalled movement requiring escalation.

Best for: Fits when hospital command centers need live capacity visibility and coordinated discharge follow-up.

Conclusion

After evaluating 9 hmm software, MATLAB stands out as our overall top pick — it scored highest across our combined criteria of features, ease of use, and value, which is why it sits at #1 in the rankings above.

Our Top Pick
MATLAB

Use the comparison table and detailed reviews above to validate the fit against your own requirements before committing to a tool.

Frequently Asked Questions About Hmm Software

Which tools in the list support HMM workflow orchestration via an API rather than notebooks or GUI screens?
Pyro provides API-driven run configuration and experiment orchestration for repeatable HMM pipelines. pomegranate exposes code-first HMM fitting and scoring methods in a Python API, while hmmlearn packages estimators for pipeline-style fit and decode steps.
How do GE HealthCare Command Center and HealthCare Logic SystemView differ in the data they prioritize for operational decision-making?
GE HealthCare Command Center emphasizes hospital capacity signals and predictive bottleneck detection in a centralized command-center view. HealthCare Logic SystemView focuses on real-time census, bed status, patient movement, and discharge barriers tied to configurable alerts and workflow tasks.
How should a research team choose between hmmlearn and pomegranate for training Hidden Markov Models with different emission types?
pomegranate supports custom emission distributions with DenseHMM and SparseHMM, which helps teams define emissions beyond fixed discrete assumptions. hmmlearn targets HMM estimation for discrete and continuous emissions using EM-style algorithms and then provides decoding like most-likely state sequences.
What breaks if an HMM implementation needs GPU execution and minibatch training inside the training loop?
hmmlearn is not designed around DenseHMM and SparseHMM GPU execution patterns with minibatches. pomegranate includes GPU execution plus minibatch training and model fitting calls, so missing GPU minibatch support blocks those throughput goals.
When is HH-suite a better choice than TensorFlow Probability for sequence modeling tasks that focus on protein homology search?
HH-suite builds and uses profile HMMs for high-throughput protein sequence database searches with tunable sensitivity. TensorFlow Probability provides HMM-ready probabilistic building blocks for custom latent-variable inference in TensorFlow, which fits inference modeling rather than homology mapping as a primary output.
Which option is most suitable for HMMs implemented as composable probabilistic distributions within TensorFlow graphs?
TensorFlow Probability supports composing state-space and distribution objects with an inference engine written around TensorFlow execution. Pyro can orchestrate training and inference runs through an API, but TensorFlow Probability is the native inference graph option inside the TensorFlow stack.
How do integration approaches differ across Shivam Medisoft and Huma for connecting operational workflows to external systems?
Shivam Medisoft centers interoperability on messaging interfaces designed to exchange clinical and administrative data with outside health systems. Huma connects study intake and workflow automation to operational execution with integrations that move study data between recruitment and operations systems, with audit-ready traceability of requests and actions.
When does an HMM toolkit need a structured state model for intake-to-documentation workflow configuration rather than analytics-only outputs?
Shivam Medisoft and Huma both map workflow states to operational task routing, where Shivam Medisoft focuses on intake to documentation and follow-up states and Huma maps requirement inputs into routed operational actions. GE HealthCare Command Center and HealthCare Logic SystemView concentrate on operational monitoring and discharge follow-up decisions rather than clinical documentation state machines.
What are the main setup tradeoffs between using MATLAB and TensorFlow Probability for HMM estimation and productionization?
MATLAB provides an end-to-end scripting and visualization loop that supports repeatable batch runs through MATLAB build and production workflows. TensorFlow Probability pushes HMM-like implementations toward TensorFlow graph execution, which changes the productionization shape to inference graphs and execution environments rather than MATLAB-first batch pipelines.

Tools reviewed

Primary sources checked during evaluation.

Referenced in the comparison table and product reviews above.

How to Choose the Right Hmm Software

We expect hmm software to differ most in how they model and tune HMM behavior for your data, how much automation and configuration they expose, and how reliably they report audit-like run details. The gap shows up in model spec control, training diagnostics, and howToCheckable outputs you can reproduce across datasets.

HMM topology and constraint controls

Why it matters

HMM topology and constraint controls determine whether the software can match your expected state count, start/end behavior, and parameter tying rules. Without that control, the model may fit poorly or require awkward workarounds.

How to check

In a test workflow, set the number of hidden states, choose initialization options, and specify any allowed/disallowed transitions or tied emissions if available. Then run again with a different state count and verify the output reflects the change rather than silently overriding it.

Training diagnostics and convergence reporting

Why it matters

Training diagnostics and convergence reporting lets us judge whether learning actually stabilizes or keeps oscillating. That matters for selecting parameters and avoiding “looks good once” runs that fail on new sequences.

How to check

Start a training job and inspect the reported iterations, stopping reason, and log-likelihood (or equivalent) trend. Ask for an option to export the training log and confirm the tool records the seed/config used for that run.

Sequence import and batch run automation

Why it matters

Sequence import and batch run automation affects throughput and consistency when we train across many datasets or segments. If automation is thin, we spend time manually reshaping inputs and risk mismatched settings.

How to check

Provide a small folder of sequence files and check whether the import maps symbols and handles varying lengths without custom scripts. Then run a batch of multiple datasets using one saved configuration and confirm outputs land in predictable locations with consistent naming.

Decoding outputs and probability traces

Why it matters

Decoding outputs and probability traces tell us what states the model chooses at each time step and how confident it is. For hmm software work, those artifacts are often the difference between a usable model and a black box.

How to check

After training, run decoding on a held-out sequence and export per-time-step state assignments plus per-step probabilities (or equivalent). Confirm you can select output granularity (per-step vs summary) and that probability values are interpretable and reproducible between runs.

Automation API and configuration export

Why it matters

Automation API and configuration export determine whether we can integrate model runs into our own pipelines and rerun experiments consistently. We avoid hidden defaults by treating configs as artifacts.

How to check

Look for an API or scripting interface that triggers training and decoding with a saved config file. Export a config from one run, import it into a second run, and verify the outputs (or at least the training trajectory) match when only the input sequences change.

Red flags

  • Topology controls don’t visibly change outputs
  • Training logs omit stopping reason or progress metrics
  • Decoding exports only labels without probabilities

Fit by situation

  • If we need a specific number of hidden states and transition rules, prioritize HMM topology and constraint controls.
  • If we compare multiple datasets and watch for overfitting, prioritize Training diagnostics and convergence reporting.
  • If we plan to run many experiments and rerun them later, prioritize Automation API and configuration export.
Commissioned Research

Funding and competing interests

This study was co-funded by several of the vendors assessed in it. No single vendor funded it. Funders could propose the market studied.

They had no role in selecting vendors, setting or weighting criteria, analysing data or drawing conclusions, and they received the same factual-accuracy check as every other vendor assessed.

Market definition, inclusion rules and criteria weights were fixed by an editor before research began and did not change afterwards. Editorial process →

Keep exploring

FOR SOFTWARE VENDORS

Not on this list? Let’s fix that.

Our best-of pages are how many teams discover and compare tools in this space. If you think your product belongs in this lineup, we’d like to hear from you—we’ll walk you through fit and what an editorial entry looks like.

Apply for a Listing

WHAT THIS INCLUDES

  • Where buyers compare

    Readers come to these pages to shortlist software—your product shows up in that moment, not in a random sidebar.

  • Editorial write-up

    We describe your product in our own words and check the facts before anything goes live.

  • On-page brand presence

    You appear in the roundup the same way as other tools we cover: name, positioning, and a clear next step for readers who want to learn more.

  • Kept up to date

    We refresh lists on a regular rhythm so the category page stays useful as products and pricing change.